feat(tui): live Markdown rendering during text streaming (v0.9.0)

Replaces v0.8.2's drop-Markdown patch with proper in-place Markdown
rendering. The transcript becomes a VerticalScroll container; each
turn's response body lives as a single Static widget whose content
is updated as Text deltas arrive — Markdown is re-rendered in place
rather than re-printed on Done. Eliminates the v0.8.x double-print
without sacrificing rich formatting.

- transcript: RichLog → VerticalScroll (#transcript-scroll)
- Text deltas: mount Static(Markdown(buffer)) on first delta;
  Static.update(Markdown(buffer)) on subsequent deltas
- --raw mode: bypass Markdown, mount Static(plain_str) for the same
  in-place update semantics
- Terminal events (Done/Error/Cancelled) mount styled label Statics
- _cancel_via_sse: write → mount Static on the new container
- _write_turn_headers: transcript gets a styled RichText Static
  ("── turn N ──"); other panes still receive Rule renderables
- Test suite reshape: bulk rename `log` → `transcript` for the
  presenter contract, `_mounted_renderables` helper extracts
  Static.content for assertion, `_spy_writes` captures both
  RichLog.write and VerticalScroll.mount

docs/contracts/issues/13.contract.md

@@ -161,7 +161,7 @@ New `Static(id="pane-name")` widget alongside the existing `identity` + `hint` w
 - **INV-019** *(amended v0.6.0)*: Three TabPanes in the right column: `Tools` (id `tools-tab`, contains `#tools-log`) + `Debug` (id `debug-tab`, contains `#debug-log`) + `Thinking` (id `thinking-tab`, contains `#thinking-log`). Ctrl+1/Ctrl+2/Ctrl+3 activate respective tabs. `pane-name` Static reflects active tab name dynamically.
 - **INV-020** *(amended v0.6.0)*: Render-exception fallback (INV-009) preserves routing per event class: `ToolStart` / `ToolResult` → `tools_log`; `Thinking` → `thinking_log`; `WorkerPhase` / `TextBoundary` → `debug_log`; everything else → `log`.
 - **INV-021** *(new v0.6.0)*: `Text` events do NOT route to `log` per-delta. They accumulate into `TuiPresenterState.text_buffer` and update a single `current_text` Static (docked above the prompt). On terminal event (`Done`/`Error`/`Cancelled`), `current_text` is cleared and (raw mode) accumulated text or (non-raw) post-Done `Markdown(response)` is written to `log`. The pre-v0.6.0 per-token RichLog spam is retired.
-- **INV-022** *(amended v0.6.5)*: Thinking deltas stream DIRECTLY into `thinking_log` (one delta = one RichLog line). The first delta of a run writes `Rule(title=f"turn N · thinking #K start")`; subsequent deltas write their raw content as lines; the run closes on the next non-thinking event with `Rule(title=f"turn N · thinking #K end")`. Pre-v0.6.5 markdown re-render dropped — the streamed deltas ARE the content; the whole pane scrolls naturally as content arrives.
+- **INV-022** *(amended v0.7.1)*: Thinking deltas COALESCE on `\n` boundaries before writing to `thinking_log`. The first delta of a run writes `Rule(title=f"turn N · thinking #K start")`; subsequent deltas accumulate in `TuiPresenterState.thinking_chunk_buffer`; whenever the buffer contains `\n`, the leading line(s) flush as RichLog entries (one entry per natural paragraph). The run closes on the next non-thinking event: any tail in the buffer flushes as a final line, then `Rule(title=f"turn N · thinking #K end")`. Pre-v0.7.1 per-delta-per-line caused token-spam (Worldtree emits thinking at token granularity); coalescing produces one log line per natural paragraph, not per token.
 - **INV-023** *(new v0.6.0)*: Turn-ID header `Rule(title=f"turn N")` is written to all four log panes (`log`, `tools_log`, `debug_log`, `thinking_log`) by `_stream_turn_worker` on the first event of each turn — enables cross-pane visual correlation during multi-turn debugging.
 - **INV-024** *(amended v0.6.5)*: `thinking-current` Static REMOVED. v0.6.1 placed it inside the Thinking pane (docked bottom); operators reported the bottom-docked Static "scrolling a little section at the bottom" (its 200-char tail acting as a scroll-window) instead of letting the whole pane scroll. v0.6.5 deletes the Static entirely and streams Thinking deltas directly into `thinking_log` (the scrollable RichLog) — the whole pane scrolls naturally as content arrives. The Rule(start) at the first delta of a run is now the live "thinking is happening" indicator.
 

docs/contracts/issues/15.contract.md

@@ -0,0 +1,290 @@
+---
+contract_version: "2.1"
+target_module: "ratatoskr.tier3"
+scope: "New module `ratatoskr.tier3` exposing Worldtree's Tier 3 (consumer-defined) agent lifecycle: `define_agent` (POST /agents/define), `patch_agent` (PATCH /agents/<id>), `delete_agent` (DELETE /agents/<id>), plus `Tier3AgentInfo` frozen dataclass. Plus a thin CLI entry point (`python -m ratatoskr.tier3 <define|patch|delete>`) that mirrors `ratatoskr.cli`'s env-var posture (`WORLDTREE_API_URL`, `WORLDTREE_API_KEY`). Convention-aligned with `ratatoskr.sessions` (issue #2): caller-owned httpx.AsyncClient, no Worldtree imports, response parsing into frozen dataclass, exception `.body` truncated to `[:1024]`. Picker stays generic — agents with `:` in agent_id show in the list like any other per issue #8's out-of-scope clause. Goal: ratatoskr operators can define, mutate, and delete Tier 3 agents from the command line, then exercise the full session flow against them to observe how Tier 3 agent_ids (colon-containing) flow through the picker / session-create / SSE stream."
+depends_on:
+  - "httpx"
+used_by: []
+language: "python"
+complexity: "low"
+estimated_loc: 250
+confidence: 0.9
+assumptions:
+  - "Tier 3 endpoints land at the same `WORLDTREE_API_URL` as the rest of the Conversation API — no separate hostname / port. Auth via the same bearer key. The caller's user_id is derived server-side from the API key's owner; the agent's `agent_id` is constructed as `<auth_user_id>:<agent_name>`. Live probe against personal Worldtree (2026-05-25) confirmed: POST with `{agent_name: 'smoke-test', ...}` and `Authorization: Bearer <key>` returned `agent_id=ratatoskr:smoke-test`, `user_id=ratatoskr`."
+  - "Per Worldtree spec §2576-2750: `agent_name` is a strict slug `[a-z][a-z0-9-]{2,63}` and immutable after definition. `user_id` is derived from the auth, must be slug-safe (`[a-z][a-z0-9-]{2,63}` per Phase 2.0 gate). PATCH accepts ONLY `system_prompt` and/or `model`; any other key (including the immutable `agent_name`, `user_id`, or layer fields `persona`/`motivational`/`valence`/`memory` — even with `null` value) returns 422 `field_not_mutable` BEFORE the DB lookup."
+  - "**Layer fields are explicitly null** on define. Phase 2.0 ships baseline addressing + ownership + lifecycle only; `persona` / `motivational` / `valence` / `memory` are schema-reserved. Non-null on these → 422 `layer_deferred`. The module's `define_agent` does NOT expose these as parameters at all — sending them would require an amendment when a future Phase enables them."
+  - "**`model` field is a provider model ID, not a profile alias.** Live probe found: `model='default'` (an llm_profiles profile name) returns 422 `model_not_available`; `model='qwen3.6-35-a3b'` (an actual provider model ID) returns 201. The CLI / module take the string verbatim and pass through — validation is server-side. Operators discover valid IDs via the model `metadata` on existing sessions or out-of-band."
+  - "**Quota: 50 Tier 3 agents per Heimdall key.** 51st define → 429 `agent_quota_exceeded` with `Retry-After: 0`. The module raises `Tier3QuotaExceeded(retry_after=0)` — the retry_after field captures the header value verbatim for forward-compat if Worldtree later returns a non-zero throttle."
+  - "**Key-revocation cascade is server-side.** When an API key is revoked (`DELETE /admin/keys/{key_id}`), every Tier 3 agent with `owner_key_hash` equal to the revoked key's hash is soft-deleted in the same SQL transaction. Active sessions on those agents return 401 `auth_revoked` on next message. The ratatoskr module doesn't track or simulate this — operators discover it via runtime 401s and the admin-side audit log."
+  - "**Picker integration is implicit** — no changes to `ratatoskr.tui.AgentPickerApp` for this issue. Tier 3 agents appear in `GET /agents` if defined and the picker's existing format `{agent_id} · {name} — {description}` renders the colon-containing agent_id without special-casing. Per issue #8 out-of-scope clause, ratatoskr does not visually distinguish Tier 1 vs Tier 3 in the picker — same UX surface."
+  - "**Session-create with colon-containing agent_id works unchanged.** Issue #5 already routes `end_user_id` into the POST /sessions body, which Tier 3 session-create requires from Phase 2.0 (per spec §2649-2664). No `ratatoskr.sessions` change needed."
+  - "**CLI uses argparse with subparsers** (define / patch / delete). The subparsers entry point lives at `python -m ratatoskr.tier3` via `__main__.py`. Output on success: prints a one-line summary (`defined ratatoskr:wizard (qwen3.6-35-a3b)` / `patched ratatoskr:wizard` / `deleted ratatoskr:wizard`). Output on error: `[<error_code>] <message>` to stderr + non-zero exit. Exit codes mirror `ratatoskr.cli`: 0 happy / 10 usage / 11 auth / 20 api-failure / 21 network."
+  - "**No `list` subcommand in v1.** A `tier3 list` operation would have to filter `GET /agents` by prefix-matching the caller's user_id, but that prefix isn't exposed in the response — only the agent_id is, and you'd have to introspect the auth's user_id. Operators discover their own Tier 3 agents by reading the `GET /agents` list (which the picker already surfaces) and looking for `<their-user-id>:*` entries. Add `list` in a follow-up if operators report friction."
+  - "**Module is standalone**: does NOT import or interact with `ratatoskr.sessions` / `ratatoskr.sse_client` / `ratatoskr.tui` / `ratatoskr.cli` beyond reusing the `USER_AGENT` constant from `ratatoskr.cli`. Cross-module use is one-way (cli supplies the user-agent string; tier3 does not import sessions). This keeps the module surface minimal and testable in isolation."
+  - "**The CLI's `python -m ratatoskr.tier3` entry point uses sys.argv handling that mirrors `ratatoskr.cli`** — a top-level `main(argv: list[str] | None = None) -> int` function that argparse-dispatches to subcommand handlers. Each subcommand handler is an async coroutine wrapped by `asyncio.run(...)`. Auth resolution: `--api-key` flag > `$WORLDTREE_API_KEY` env > `_AuthError` (exit 11). Server URL: `--server` > `$WORLDTREE_API_URL` > default `http://localhost:8000` (same default as `ratatoskr.cli`)."
+  - "**Tests use `respx` for HTTP mocking** (same pattern as `tests/test_sessions.py`). New test file: `tests/test_tier3.py`. Cover all success + error response codes per the ERROR_ROUTING matrix below. No live network in unit tests — the live smoke is in the acceptance criteria, not the unit tests."
+open_questions:
+  - "Should `define_agent` accept an optional `bifrost` parameter for Bifrost-bound Tier 3 sessions? The spec §2658 shows `bifrost` as a session-create field (not define-time). Draft: no — Bifrost binding is per-session; if a Tier 3 agent needs Bifrost on every session, that's an orthogonal feature on POST /sessions, not POST /agents/define. Issue #5's `--end-user-id` already covers the session-create-side parameters."
+  - "Should the CLI also offer `--end-user-id` for sessions created via tier3 + ratatoskr-cli composition? Draft: no — once an agent is defined, operators use the main `ratatoskr --new --agent <id> --end-user-id <eid>` flow; tier3 CLI is define/patch/delete only."
+  - "Should `delete_agent` support a `--force` flag for 'really delete even if active sessions exist'? Per spec §2634-2639, `DELETE` already cancels active sessions and revokes the per-resource scope grant on the owner — there's no soft fail. Draft: no — the spec's hard-delete-with-cascade behavior is the right shape; ratatoskr doesn't need to wrap it."
+prd:
+  issue: 15
+  issue_url: "https://gitea.phasefinal.com/vh/ratatoskr/issues/15"
+  body_sha256_16: "03367d7b451ab17f"
+  lock_in_comment_id: null
+  lock_in_sha256_16: null
+  lock_in_at: null
+  pinned_at: "2026-05-25T03:21:38+00:00"
+dependencies:
+  - issue: 2
+    path: "src/ratatoskr/sessions.py"
+    reason: "Convention dependency, not a code dependency. Issue #2 (`ratatoskr.sessions`) is the posture template: caller-owned httpx client, async-native, no Worldtree imports, response-parsing into frozen dataclasses, exception body truncation to [:1024]. `ratatoskr.tier3` follows the same shape verbatim."
+  - issue: 3
+    path: "src/ratatoskr/cli.py"
+    reason: "Convention dependency only. `ratatoskr.tier3.__main__` mirrors `ratatoskr.cli`'s argparse + env-fallback + exit-code shape. Imports `USER_AGENT` from `ratatoskr.cli` so outbound HTTP carries the same identity string."
+---
+
+# Tier 3 — Consumer-defined agent lifecycle module
+
+## Context
+
+Worldtree's Tier 3 (Phase 2.0, spec §2576-2750) lets the consumer define their own agents at `<user_id>:<agent_name>`. The agent's `user_id` is the auth's user identity (derived from the API key's owner); the `agent_name` is supplied at define-time. The lifecycle is owner-only — only the key that defined an agent can patch / delete it (modulo the key-revocation cascade).
+
+`ratatoskr.tier3` exposes this lifecycle as a Python module + small CLI tool. Picker integration is implicit (Tier 3 agents already appear in `GET /agents` per issue #8). Session-create works unchanged through `ratatoskr.sessions.create_session` since the colon-containing agent_id is opaque to that layer.
+
+## Public surface
+
+```python
+@dataclass(frozen=True)
+class Tier3AgentInfo:
+    """Worldtree Tier 3 agent envelope returned by define / patch."""
+
+    agent_id: str       # f"{user_id}:{agent_name}"
+    user_id: str
+    agent_name: str
+    system_prompt: str
+    model: str
+    created_at: str     # ISO 8601 with offset
+    updated_at: str     # ISO 8601 with offset
+
+
+async def define_agent(
+    client: httpx.AsyncClient,
+    *,
+    agent_name: str,
+    system_prompt: str,
+    model: str,
+) -> Tier3AgentInfo:
+    """POST /agents/define → 201 with Tier3AgentInfo. See FN define_agent."""
+
+
+async def patch_agent(
+    client: httpx.AsyncClient,
+    agent_id: str,
+    *,
+    system_prompt: str | None = None,
+    model: str | None = None,
+) -> Tier3AgentInfo:
+    """PATCH /agents/<id> → 200 with updated Tier3AgentInfo. See FN patch_agent."""
+
+
+async def delete_agent(client: httpx.AsyncClient, agent_id: str) -> None:
+    """DELETE /agents/<id> → 204. See FN delete_agent."""
+```
+
+## Exception classes
+
+```python
+class Tier3QuotaExceeded(Exception):
+    """429 agent_quota_exceeded — 50-agent cap reached on the Heimdall key."""
+    def __init__(self, *, retry_after: int) -> None: ...
+    retry_after: int
+
+class Tier3UserIdUnsupported(Exception):
+    """403 tier3_user_id_unsupported — auth's user_id not slug-safe."""
+
+class Tier3FieldNotMutable(Exception):
+    """422 field_not_mutable — PATCH carrying an immutable key."""
+    def __init__(self, *, field: str | None) -> None: ...
+    field: str | None
+
+class Tier3LayerDeferred(Exception):
+    """422 layer_deferred — define carrying non-null layer field."""
+    def __init__(self, *, field: str | None) -> None: ...
+    field: str | None
+
+class Tier3AgentNotFound(Exception):
+    """404 — patch/delete on non-existent agent."""
+    def __init__(self, *, agent_id: str) -> None: ...
+    agent_id: str
+
+# Reused from ratatoskr.sessions (one-way import — sessions doesn't depend on tier3):
+# SessionApiFailed(status, body) for all other non-2xx responses.
+```
+
+## Functions
+
+### FN define_agent
+
+```
+FN define_agent(
+    client: httpx.AsyncClient,
+    *, agent_name: str, system_prompt: str, model: str,
+) -> Tier3AgentInfo
+BRIEF: POST /agents/define → 201 with Tier3AgentInfo.
+
+PRE-001: agent_name matches `[a-z][a-z0-9-]{2,63}` (slug guard — client-side
+         assert; the server enforces too, but this prevents wire round-trip
+         for trivially-bad input).
+PRE-002: system_prompt is non-empty.
+PRE-003: model is non-empty.
+
+STEPS:
+  1. assert PRE-001/002/003.
+  2. body = {
+       "agent_name": agent_name,
+       "system_prompt": system_prompt,
+       "model": model,
+     }
+  3. resp = await client.post("/agents/define", json=body)
+  4. ROUTE response status:
+     201 → parse body into Tier3AgentInfo, return.
+     422 → inspect error_code:
+       layer_deferred → raise Tier3LayerDeferred(field=err.get("field"))
+       (others) → raise SessionApiFailed(status=422, body=resp.content)
+     403 + tier3_user_id_unsupported → raise Tier3UserIdUnsupported
+     429 → raise Tier3QuotaExceeded(retry_after=int(resp.headers.get("Retry-After", 0)))
+     other → raise SessionApiFailed(status, body)
+
+POST-001: returned Tier3AgentInfo has agent_id of shape "<user_id>:<agent_name>".
+```
+
+### FN patch_agent
+
+```
+FN patch_agent(
+    client: httpx.AsyncClient, agent_id: str,
+    *, system_prompt: str | None = None, model: str | None = None,
+) -> Tier3AgentInfo
+BRIEF: PATCH /agents/<id> → 200 with updated Tier3AgentInfo.
+
+PRE-001: agent_id contains `:` (Tier 3 shape).
+PRE-002: at least one of system_prompt or model is non-None (no-op patches
+         are still server-accepted but client-side assert avoids the round-trip).
+
+STEPS:
+  1. assert PRE-001/002.
+  2. body = {}; if system_prompt is not None: body["system_prompt"] = system_prompt;
+     if model is not None: body["model"] = model.
+  3. resp = await client.patch(f"/agents/{agent_id}", json=body)
+  4. ROUTE response status:
+     200 → parse, return.
+     404 → raise Tier3AgentNotFound(agent_id=agent_id)
+     422 + field_not_mutable → raise Tier3FieldNotMutable(field=err.get("field"))
+     other → raise SessionApiFailed(status, body)
+```
+
+### FN delete_agent
+
+```
+FN delete_agent(client: httpx.AsyncClient, agent_id: str) -> None
+BRIEF: DELETE /agents/<id> → 204.
+
+PRE-001: agent_id contains `:` (Tier 3 shape).
+
+STEPS:
+  1. assert PRE-001.
+  2. resp = await client.delete(f"/agents/{agent_id}")
+  3. ROUTE response status:
+     204 → return None.
+     404 → raise Tier3AgentNotFound(agent_id=agent_id)
+     other → raise SessionApiFailed(status, body)
+```
+
+## CLI surface (`python -m ratatoskr.tier3`)
+
+```
+$ python -m ratatoskr.tier3 define --name wizard \
+    --system-prompt "You are a guided-elicitation wizard..." \
+    --model qwen3.6-35-a3b
+defined ratatoskr:wizard (qwen3.6-35-a3b)
+
+$ python -m ratatoskr.tier3 patch ratatoskr:wizard --system-prompt "New prompt"
+patched ratatoskr:wizard
+
+$ python -m ratatoskr.tier3 delete ratatoskr:wizard
+deleted ratatoskr:wizard
+```
+
+Auth + server URL: same env-var fallback as `ratatoskr.cli`. Exit codes: 0 / 10 (usage) / 11 (auth) / 20 (api-failure) / 21 (network).
+
+## Invariants
+
+- **INV-001**: `define_agent` request body carries exactly `{agent_name, system_prompt, model}` — no layer fields, no `bifrost`, no `metadata`. Phase 2.0 baseline shape only.
+- **INV-002**: `patch_agent` request body carries ONLY `system_prompt` and/or `model` — every other key is omitted. Server-side 422 `field_not_mutable` is the safety net; client-side body-construction is the first line.
+- **INV-003**: `delete_agent` is fire-and-confirm — no body, no retry, no soft-delete. Cascade handling is server-side; ratatoskr doesn't track it.
+- **INV-004**: All exceptions carry a `[:1024]` body cap (when applicable) per the issue #2 convention.
+- **INV-005**: CLI auth resolution mirrors `ratatoskr.cli`: `--api-key` flag > `$WORLDTREE_API_KEY` > exit 11.
+- **INV-006**: CLI server URL resolution mirrors `ratatoskr.cli`: `--server` > `$WORLDTREE_API_URL` > `http://localhost:8000`.
+- **INV-007**: Module never imports `ratatoskr.sessions` / `ratatoskr.sse_client` / `ratatoskr.tui` (one-way: only `cli.USER_AGENT` is imported, and only by `__main__.py` for the outbound User-Agent header).
+- **INV-008**: All HTTP through caller-owned `httpx.AsyncClient` — module never constructs its own client. (`__main__` constructs one for the CLI entry point per ratatoskr.cli's pattern.)
+
+## TESTS (tests/test_tier3.py — new file)
+
+```
+- test_define_happy: 201 + full response shape → Tier3AgentInfo populated.
+- test_define_quota_exceeded: 429 + Retry-After header → Tier3QuotaExceeded(retry_after=N).
+- test_define_user_id_unsupported: 403 tier3_user_id_unsupported → Tier3UserIdUnsupported.
+- test_define_layer_deferred_persona: 422 layer_deferred → Tier3LayerDeferred (would only fire if the body sent a layer field; the module never sends one, so this asserts server-side defense but reflecting a 422 we don't actually generate. Test exercises the response path, not the request).
+- test_define_bad_slug: PRE-001 assertion fires before HTTP for agent_name="X" (uppercase) or "ab" (too short).
+- test_define_empty_prompt: PRE-002 assertion fires for empty system_prompt.
+- test_define_other_5xx: 503 → SessionApiFailed(status=503).
+- test_patch_happy_both_fields: 200 + updated body → Tier3AgentInfo.
+- test_patch_happy_single_field: 200 with only system_prompt set; body omits model.
+- test_patch_field_not_mutable: 422 field_not_mutable → Tier3FieldNotMutable.
+- test_patch_404: 404 → Tier3AgentNotFound(agent_id=...).
+- test_patch_no_args: PRE-002 assertion fires (both None).
+- test_patch_non_tier3_id: PRE-001 assertion fires for agent_id without `:`.
+- test_delete_happy: 204 → returns None.
+- test_delete_404: 404 → Tier3AgentNotFound.
+- test_delete_non_tier3_id: PRE-001 assertion fires.
+- test_delete_other_5xx: 500 → SessionApiFailed.
+- test_cli_define_happy: argv → 201 mock → stdout="defined ratatoskr:wizard (qwen3.6-35-a3b)" + exit 0.
+- test_cli_patch_happy: argv → 200 mock → stdout="patched ratatoskr:wizard" + exit 0.
+- test_cli_delete_happy: argv → 204 mock → stdout="deleted ratatoskr:wizard" + exit 0.
+- test_cli_missing_auth: no API key → stderr "[auth_error]" + exit 11.
+- test_cli_api_failed: 500 mock → stderr "[api_failed]" + exit 20.
+```
+
+## ERROR_ROUTING (module + CLI)
+
+| HTTP shape | error_code | Exception (module) | CLI label | Exit |
+|---|---|---|---|---|
+| 201 / 200 / 204 | — | (none — happy) | one-line confirmation on stdout | 0 |
+| 429 | agent_quota_exceeded | `Tier3QuotaExceeded(retry_after=N)` | `[quota_exceeded] retry_after=N` | 20 |
+| 403 | tier3_user_id_unsupported | `Tier3UserIdUnsupported` | `[user_id_unsupported]` | 20 |
+| 404 | — | `Tier3AgentNotFound(agent_id=...)` | `[agent_not_found] <id>` | 20 |
+| 422 | field_not_mutable | `Tier3FieldNotMutable(field=...)` | `[field_not_mutable] field=...` | 20 |
+| 422 | layer_deferred | `Tier3LayerDeferred(field=...)` | `[layer_deferred] field=...` | 20 |
+| any other non-2xx | — | `SessionApiFailed(status, body)` | `[api_failed] status=N body=...` | 20 |
+| httpx.ConnectError / ReadTimeout / TransportError | — | propagates | `[network_error] T: M` | 21 |
+| PRE-001/002/003 assertion violation | — | `AssertionError` | `[usage_error] <msg>` | 10 |
+| no auth | — | `_AuthError` (reused from cli) | `[auth_error] no API key` | 11 |
+
+## Layout after this module lands
+
+```
+src/ratatoskr/
+  __init__.py
+  cli.py            (existing, unchanged)
+  sessions.py       (existing, unchanged)
+  sse_client.py     (existing, unchanged)
+  tui.py            (existing, unchanged)
+  tier3.py          NEW
+  __main__/         (no change — main cli still entry-point)
+
+# CLI invocation:
+$ python -m ratatoskr.tier3 define --name wizard ...
+$ python -m ratatoskr.tier3 patch ratatoskr:wizard ...
+$ python -m ratatoskr.tier3 delete ratatoskr:wizard
+```

persistent-memory.md

@@ -32,9 +32,9 @@ separate dev team rather than an in-tree Worldtree tool.
 
 ## Current state / in-flight
 
-_As of 2026-05-24 (post-v0.6.5 thinking streams into whole pane):_
+_As of 2026-05-25 (post-v0.8.2 drop double-print; v0.9.0 live-md next):_
 
-**Status: v0.6.5 shipped.** Nine core issues complete (`sse_client`
+**Status: v0.8.2 shipped.** Eleven core features complete (`sse_client`
 #1, `sessions` #2, `cli` #3, `tui` #4, `--end-user-id` #5, TUI
 startup error visibility #6, presenter contract semantics amendment
 #12, startup agent picker #8, §5 layout reshape + Tools pane #13)
@@ -51,7 +51,12 @@ Static in the footer (static "Tools" v1; dynamic when more tabs
 land). CLI mode (--send) unaffected by design — INV-018.
 
 Last commits on `main`:
-- v0.6.5 refactor(tui): thinking streams into thinking-log (no Static)
+- v0.8.2 fix(tui): drop post-Done Markdown body re-render (no double-print)
+- `11ef683` fix(tui,sse): inline Text streaming + empty-id keepalive skip (v0.8.1)
+- `9fade55` feat(local_agents): JSON-backed local tier-3 index + picker merge (v0.8.0)
+- `9918c10` fix(tui): coalesce thinking deltas on `\n` (v0.7.1)
+- `c086ae2` feat(tier3): ratatoskr.tier3 module + CLI (v0.7.0)
+- `d356990` refactor(tui): thinking streams into thinking-log (v0.6.5)
 - `82437bd` style(tui): picker highlighted item → Aurora blue (v0.6.4)
 - `ac690c1` style(tui): restore Australis palette, only $background → pure black (v0.6.3)
 - `d845b20` style(tui): neutralize Australis dark palette (v0.6.2, reverted)

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "ratatoskr"
-version = "0.6.5"
+version = "0.9.0"
 description = "Worldtree Conversation API debug TUI — multi-pane observability dashboard"
 readme = "README.md"
 requires-python = ">=3.12"

src/ratatoskr/local_agents.py

@@ -0,0 +1,134 @@
+"""Local index of tier-3 agents defined via `python -m ratatoskr.tier3`.
+
+Workaround for Worldtree's ``GET /agents`` not returning consumer-defined
+agents (the public list excludes tier-3 per-spec; see issue #15 smoke
+findings). Local file maintains a list of agent_ids + display metadata so
+the picker can show them alongside foundational agents.
+
+Storage shape: JSON at ``$XDG_CONFIG_HOME/ratatoskr/local_agents.json``
+(default ``~/.config/ratatoskr/local_agents.json``). Override via
+``$RATATOSKR_LOCAL_AGENTS`` env var for tests / per-machine isolation.
+
+If Worldtree later starts returning tier-3 agents in ``GET /agents``, this
+module's role narrows to redundant local cache; can be removed cleanly
+since the picker's dedup-by-agent-id keeps remote-wins behavior.
+
+Failure modes are lenient: missing file → empty index; corrupt JSON or
+schema mismatch → empty index (no crash). The picker continues to show
+foundational agents either way; the local-tier-3 surface degrades to
+"operator passes --agent ratatoskr:<name> explicitly" — the
+pre-v0.8.0 workflow.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from dataclasses import asdict, dataclass
+from pathlib import Path
+
+_SCHEMA_VERSION = 1
+
+
+@dataclass(frozen=True)
+class LocalAgentEntry:
+    """One row in the local tier-3 agent index.
+
+    Schema:
+      - ``agent_id``: full "user_id:agent_name" string (Worldtree-owned).
+      - ``agent_name``: slug from define (display name).
+      - ``model``: provider model ID at last define/patch.
+      - ``description``: synthetic display string (typically derived from
+        the system_prompt's first line + a "(tier 3)" prefix; the picker
+        uses this in its ``{id} · {name} — {description}`` rendering).
+      - ``defined_at``: ISO-8601 timestamp from the Tier3AgentInfo response.
+    """
+
+    agent_id: str
+    agent_name: str
+    model: str
+    description: str
+    defined_at: str
+
+
+def _local_agents_path() -> Path:
+    """Resolve the local index file path with XDG + env-var override."""
+    override = os.environ.get("RATATOSKR_LOCAL_AGENTS")
+    if override:
+        return Path(override)
+    xdg = os.environ.get("XDG_CONFIG_HOME")
+    base = Path(xdg) if xdg else (Path.home() / ".config")
+    return base / "ratatoskr" / "local_agents.json"
+
+
+def load_local_agents() -> list[LocalAgentEntry]:
+    """Read the local index. Returns ``[]`` on missing file, corrupt JSON,
+    schema mismatch, or any read error — never raises.
+    """
+    path = _local_agents_path()
+    if not path.exists():
+        return []
+    try:
+        raw = json.loads(path.read_text())
+    except (json.JSONDecodeError, OSError):
+        return []
+    if not isinstance(raw, dict) or raw.get("version") != _SCHEMA_VERSION:
+        return []
+    agents = raw.get("agents", [])
+    if not isinstance(agents, list):
+        return []
+    out: list[LocalAgentEntry] = []
+    for item in agents:
+        if not isinstance(item, dict):
+            continue
+        try:
+            out.append(LocalAgentEntry(**item))
+        except TypeError:
+            # Malformed row (missing/extra fields) — skip silently.
+            continue
+    return out
+
+
+def _save_local_agents(agents: list[LocalAgentEntry]) -> None:
+    """Persist the index. Creates parent dir as needed."""
+    path = _local_agents_path()
+    path.parent.mkdir(parents=True, exist_ok=True)
+    payload = {"version": _SCHEMA_VERSION, "agents": [asdict(a) for a in agents]}
+    path.write_text(json.dumps(payload, indent=2))
+
+
+def add_local_agent(entry: LocalAgentEntry) -> None:
+    """Add (or replace) an agent in the local index. agent_id is the key."""
+    agents = [a for a in load_local_agents() if a.agent_id != entry.agent_id]
+    agents.append(entry)
+    _save_local_agents(agents)
+
+
+def update_local_agent(entry: LocalAgentEntry) -> None:
+    """Update an existing entry. Identical semantics to ``add_local_agent``
+    (agent_id is the dedup key), exposed separately so callers can
+    self-document intent.
+    """
+    add_local_agent(entry)
+
+
+def remove_local_agent(agent_id: str) -> None:
+    """Remove an entry by agent_id. No-op if absent (idempotent)."""
+    agents = [a for a in load_local_agents() if a.agent_id != agent_id]
+    _save_local_agents(agents)
+
+
+def make_description(system_prompt: str) -> str:
+    """Synthesize a one-line description for the picker from a system prompt.
+
+    Strategy: first non-empty line, stripped of leading markdown heading
+    markers and whitespace, prefixed with "(tier 3) ", truncated to 80
+    chars. Falls back to "(tier 3) custom system prompt" if the prompt is
+    empty (defensive — define rejects empty prompts at PRE-002).
+    """
+    for line in system_prompt.splitlines():
+        stripped = line.lstrip("# ").strip()
+        if stripped:
+            label = f"(tier 3) {stripped}"
+            return label[:80] + ("…" if len(label) > 80 else "")
+    return "(tier 3) custom system prompt"

src/ratatoskr/sse_client.py

@@ -309,6 +309,14 @@ async def _iter_events(
             # with a bad id is still a keepalive). Don't reorder.
             if sse.data == "":
                 continue
+            # v0.8.1: empty-id frames are also treated as keepalives. Worldtree
+            # SOMETIMES emits events without an `id:` line (observed mid-stream
+            # on the qwen3.6-35-a3b-heretic provider, 2026-05-25). Per the SSE
+            # RFC, events without ids are legitimate (they just don't update
+            # Last-Event-ID); the previous strict behavior crashed every turn
+            # on the offending agent. Treat same as empty-data: skip silently.
+            if sse.id == "":
+                continue
             try:
                 sse_id = _parse_sse_id(sse.id)
             except ValueError as exc:

src/ratatoskr/tier3.py

@@ -0,0 +1,471 @@
+"""Worldtree Tier 3 (consumer-defined) agent lifecycle client.
+
+Implements docs/contracts/issues/15.contract.md. Caller-owned httpx.AsyncClient
+posture (same as ratatoskr.sessions). Exposes three lifecycle operations:
+
+- ``define_agent`` — POST /agents/define
+- ``patch_agent`` — PATCH /agents/<id>
+- ``delete_agent`` — DELETE /agents/<id>
+
+Plus a frozen ``Tier3AgentInfo`` dataclass for the response shape. The picker
+already handles colon-containing agent_ids generically (issue #8); session
+creation works unchanged via ``ratatoskr.sessions.create_session``.
+
+Spec reference: ``docs/conversation-api-spec.md`` §2576-2750 (Phase 2.0).
+"""
+
+from __future__ import annotations
+
+import argparse
+import re
+from dataclasses import dataclass
+
+import httpx
+
+from ratatoskr.sessions import SessionApiFailed
+
+# Per spec §2627: agent_name + user_id slugs are `[a-z][a-z0-9-]{2,63}`.
+_SLUG_RE = re.compile(r"^[a-z][a-z0-9-]{2,63}$")
+
+
+@dataclass(frozen=True)
+class Tier3AgentInfo:
+    """Worldtree Tier 3 agent envelope returned by define / patch.
+
+    INV-001: ``agent_id`` is always shape ``"<user_id>:<agent_name>"`` —
+    constructed server-side from the auth's user_id + the supplied agent_name.
+    """
+
+    agent_id: str
+    user_id: str
+    agent_name: str
+    system_prompt: str
+    model: str
+    created_at: str
+    updated_at: str
+
+
+class Tier3QuotaExceeded(Exception):
+    """Raised on HTTP 429 ``agent_quota_exceeded`` — 50-agent cap reached
+    on the Heimdall key. ``retry_after`` captures the Retry-After header
+    verbatim (defaults to 0 per spec §2675; forward-compat for non-zero)."""
+
+    def __init__(self, *, retry_after: int) -> None:
+        super().__init__(f"Tier 3 agent quota exceeded (retry_after={retry_after})")
+        self.retry_after = retry_after
+
+
+class Tier3UserIdUnsupported(Exception):
+    """Raised on HTTP 403 ``tier3_user_id_unsupported`` — auth's user_id
+    is not slug-safe per Phase 2.0 gate (spec §2626)."""
+
+    def __init__(self) -> None:
+        super().__init__("tier3 caller user_id is not slug-safe")
+
+
+class Tier3FieldNotMutable(Exception):
+    """Raised on HTTP 422 ``field_not_mutable`` — PATCH request body
+    carried a key that's immutable post-define (``agent_name``, ``user_id``,
+    or any layer field). Server rejects BEFORE the DB lookup (spec §2644)."""
+
+    def __init__(self, *, field: str | None) -> None:
+        super().__init__(f"field not mutable on Tier 3 patch: {field!r}")
+        self.field = field
+
+
+class Tier3LayerDeferred(Exception):
+    """Raised on HTTP 422 ``layer_deferred`` — define request carried a
+    non-null layer field (``persona`` / ``motivational`` / ``valence`` /
+    ``memory``). Phase 2.0 ships baseline only; layers are schema-reserved.
+
+    Note: ``define_agent`` never sends layer fields, so this exception is
+    defense-against-server-side-changes / forward-compat. INV-001 in the
+    request body construction is the first line of defense.
+    """
+
+    def __init__(self, *, field: str | None) -> None:
+        super().__init__(f"tier3 layer field deferred: {field!r}")
+        self.field = field
+
+
+class Tier3AgentNotFound(Exception):
+    """Raised on HTTP 404 — PATCH or DELETE on a non-existent agent_id
+    (spec §2634 + §2641)."""
+
+    def __init__(self, *, agent_id: str) -> None:
+        super().__init__(f"tier3 agent not found: {agent_id!r}")
+        self.agent_id = agent_id
+
+
+def _extract_error_code(resp: httpx.Response) -> str | None:
+    """Pluck the ``detail.error_code`` from a Worldtree error envelope.
+
+    Worldtree wraps API errors in ``{"detail": {"error_code": "...", ...}}``
+    per the spec. Returns None on shape mismatch (so callers fall through
+    to the generic ``SessionApiFailed`` branch).
+    """
+    try:
+        body = resp.json()
+    except ValueError:
+        return None
+    detail = body.get("detail") if isinstance(body, dict) else None
+    if isinstance(detail, dict):
+        code = detail.get("error_code")
+        if isinstance(code, str):
+            return code
+    return None
+
+
+def _extract_error_field(resp: httpx.Response) -> str | None:
+    """Pluck ``detail.field`` from a Worldtree error envelope (used for
+    ``field_not_mutable`` and ``layer_deferred`` to surface which field
+    triggered the rejection). Returns None on shape mismatch.
+    """
+    try:
+        body = resp.json()
+    except ValueError:
+        return None
+    detail = body.get("detail") if isinstance(body, dict) else None
+    if isinstance(detail, dict):
+        field = detail.get("field")
+        if isinstance(field, str):
+            return field
+    return None
+
+
+def _parse_tier3_agent_info(body: dict) -> Tier3AgentInfo:
+    """Parse a Worldtree Tier 3 agent JSON body into the frozen dataclass."""
+    return Tier3AgentInfo(
+        agent_id=body["agent_id"],
+        user_id=body["user_id"],
+        agent_name=body["agent_name"],
+        system_prompt=body["system_prompt"],
+        model=body["model"],
+        created_at=body["created_at"],
+        updated_at=body["updated_at"],
+    )
+
+
+async def define_agent(
+    client: httpx.AsyncClient,
+    *,
+    agent_name: str,
+    system_prompt: str,
+    model: str,
+) -> Tier3AgentInfo:
+    """POST /agents/define — create a Tier 3 agent.
+
+    See contract FN define_agent. Validates the agent_name slug client-side
+    before the network round-trip; server-side validation is the safety net.
+    Returns a fully populated Tier3AgentInfo on 201. Routes documented error
+    codes to typed exceptions; unknown non-2xx → SessionApiFailed.
+    """
+    assert client is not None
+    assert _SLUG_RE.match(agent_name), (
+        f"agent_name must match [a-z][a-z0-9-]{{2,63}}: {agent_name!r}"
+    )
+    assert system_prompt, "system_prompt must be non-empty"
+    assert model, "model must be non-empty"
+
+    body = {
+        "agent_name": agent_name,
+        "system_prompt": system_prompt,
+        "model": model,
+    }
+    resp = await client.post("/agents/define", json=body)
+
+    if resp.status_code == 201:
+        return _parse_tier3_agent_info(resp.json())
+    if resp.status_code == 429:
+        # Spec §2675: 51st define → 429 with Retry-After: 0.
+        try:
+            retry_after = int(resp.headers.get("Retry-After", "0"))
+        except (TypeError, ValueError):
+            retry_after = 0
+        raise Tier3QuotaExceeded(retry_after=retry_after)
+    if resp.status_code == 403:
+        if _extract_error_code(resp) == "tier3_user_id_unsupported":
+            raise Tier3UserIdUnsupported()
+    if resp.status_code == 422:
+        code = _extract_error_code(resp)
+        if code == "layer_deferred":
+            raise Tier3LayerDeferred(field=_extract_error_field(resp))
+    raise SessionApiFailed(status=resp.status_code, body=resp.content)
+
+
+async def patch_agent(
+    client: httpx.AsyncClient,
+    agent_id: str,
+    *,
+    system_prompt: str | None = None,
+    model: str | None = None,
+) -> Tier3AgentInfo:
+    """PATCH /agents/<id> — mutate system_prompt and/or model.
+
+    See contract FN patch_agent. Per spec §2641: only system_prompt + model
+    are mutable in Phase 2.0; any other key returns 422 field_not_mutable.
+    """
+    assert client is not None
+    assert ":" in agent_id, f"tier 3 agent_id must contain ':': {agent_id!r}"
+    assert system_prompt is not None or model is not None, (
+        "patch requires at least one of system_prompt or model"
+    )
+
+    body: dict[str, str] = {}
+    if system_prompt is not None:
+        body["system_prompt"] = system_prompt
+    if model is not None:
+        body["model"] = model
+    resp = await client.patch(f"/agents/{agent_id}", json=body)
+
+    if resp.status_code == 200:
+        return _parse_tier3_agent_info(resp.json())
+    if resp.status_code == 404:
+        raise Tier3AgentNotFound(agent_id=agent_id)
+    if resp.status_code == 422:
+        code = _extract_error_code(resp)
+        if code == "field_not_mutable":
+            raise Tier3FieldNotMutable(field=_extract_error_field(resp))
+    raise SessionApiFailed(status=resp.status_code, body=resp.content)
+
+
+async def delete_agent(client: httpx.AsyncClient, agent_id: str) -> None:
+    """DELETE /agents/<id> — owner hard-delete (cancels active sessions
+    server-side per spec §2636).
+
+    See contract FN delete_agent. 204 on success; 404 if the agent_id
+    doesn't exist; other non-2xx → SessionApiFailed.
+    """
+    assert client is not None
+    assert ":" in agent_id, f"tier 3 agent_id must contain ':': {agent_id!r}"
+
+    resp = await client.delete(f"/agents/{agent_id}")
+    if resp.status_code == 204:
+        return
+    if resp.status_code == 404:
+        raise Tier3AgentNotFound(agent_id=agent_id)
+    raise SessionApiFailed(status=resp.status_code, body=resp.content)
+
+
+# ---- CLI (`python -m ratatoskr.tier3 <subcommand>`) ------------------------
+#
+# Auth + server URL resolution mirrors ratatoskr.cli verbatim. Exit codes
+# mirror ratatoskr.cli: 0 happy / 10 usage / 11 auth / 20 api-failure /
+# 21 network. Outbound requests carry the same User-Agent string.
+
+
+class _Tier3UsageError(Exception):
+    """Argparse usage violation → exit 10."""
+
+
+class _Tier3AuthError(Exception):
+    """No API key resolvable → exit 11."""
+
+
+def _build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        prog="python -m ratatoskr.tier3",
+        description="Worldtree Tier 3 (consumer-defined) agent lifecycle.",
+    )
+    parser.add_argument("--api-key", dest="api_key", default=None)
+    parser.add_argument("--server", dest="server", default=None)
+    sub = parser.add_subparsers(dest="cmd", required=True)
+
+    p_define = sub.add_parser("define", help="Create a Tier 3 agent.")
+    p_define.add_argument("--name", required=True, help="agent_name (slug).")
+    p_define.add_argument(
+        "--system-prompt", dest="system_prompt", required=True,
+        help="System prompt the agent ships with.",
+    )
+    p_define.add_argument(
+        "--model", required=True,
+        help="Provider model ID (NOT a profile alias; e.g., qwen3.6-35-a3b).",
+    )
+
+    p_patch = sub.add_parser("patch", help="Mutate system_prompt and/or model.")
+    p_patch.add_argument("agent_id", help='Full "<user_id>:<agent_name>" form.')
+    p_patch.add_argument("--system-prompt", dest="system_prompt", default=None)
+    p_patch.add_argument("--model", default=None)
+
+    p_delete = sub.add_parser("delete", help="Hard-delete a Tier 3 agent.")
+    p_delete.add_argument("agent_id", help='Full "<user_id>:<agent_name>" form.')
+
+    return parser
+
+
+def _resolve_auth(ns: argparse.Namespace) -> tuple[str, str]:
+    """Resolve API key + server URL with the same env-var fallback as cli.py."""
+    import os
+
+    api_key = ns.api_key or os.environ.get("WORLDTREE_API_KEY") or ""
+    if not api_key:
+        raise _Tier3AuthError("no API key (set --api-key or WORLDTREE_API_KEY)")
+    server_url = (
+        ns.server or os.environ.get("WORLDTREE_API_URL") or "http://localhost:8000"
+    )
+    return api_key, server_url
+
+
+async def _run_define(ns: argparse.Namespace) -> int:
+    api_key, server_url = _resolve_auth(ns)
+    from ratatoskr.cli import USER_AGENT
+    from ratatoskr.local_agents import (
+        LocalAgentEntry,
+        add_local_agent,
+        make_description,
+    )
+
+    async with httpx.AsyncClient(
+        base_url=server_url,
+        headers={
+            "Authorization": f"Bearer {api_key}",
+            "User-Agent": USER_AGENT,
+        },
+        timeout=httpx.Timeout(connect=10.0, read=30.0, write=10.0, pool=10.0),
+    ) as client:
+        info = await define_agent(
+            client,
+            agent_name=ns.name,
+            system_prompt=ns.system_prompt,
+            model=ns.model,
+        )
+        # v0.8.0: persist to local index so the picker can show it.
+        add_local_agent(
+            LocalAgentEntry(
+                agent_id=info.agent_id,
+                agent_name=info.agent_name,
+                model=info.model,
+                description=make_description(info.system_prompt),
+                defined_at=info.created_at,
+            )
+        )
+        print(f"defined {info.agent_id} ({info.model})")
+        return 0
+
+
+async def _run_patch(ns: argparse.Namespace) -> int:
+    api_key, server_url = _resolve_auth(ns)
+    from ratatoskr.cli import USER_AGENT
+    from ratatoskr.local_agents import (
+        LocalAgentEntry,
+        make_description,
+        update_local_agent,
+    )
+
+    if ns.system_prompt is None and ns.model is None:
+        raise _Tier3UsageError(
+            "patch requires at least one of --system-prompt or --model"
+        )
+    async with httpx.AsyncClient(
+        base_url=server_url,
+        headers={
+            "Authorization": f"Bearer {api_key}",
+            "User-Agent": USER_AGENT,
+        },
+        timeout=httpx.Timeout(connect=10.0, read=30.0, write=10.0, pool=10.0),
+    ) as client:
+        info = await patch_agent(
+            client,
+            ns.agent_id,
+            system_prompt=ns.system_prompt,
+            model=ns.model,
+        )
+        # v0.8.0: refresh local index with the post-patch state.
+        update_local_agent(
+            LocalAgentEntry(
+                agent_id=info.agent_id,
+                agent_name=info.agent_name,
+                model=info.model,
+                description=make_description(info.system_prompt),
+                defined_at=info.updated_at,
+            )
+        )
+        print(f"patched {info.agent_id}")
+        return 0
+
+
+async def _run_delete(ns: argparse.Namespace) -> int:
+    api_key, server_url = _resolve_auth(ns)
+    from ratatoskr.cli import USER_AGENT
+    from ratatoskr.local_agents import remove_local_agent
+
+    async with httpx.AsyncClient(
+        base_url=server_url,
+        headers={
+            "Authorization": f"Bearer {api_key}",
+            "User-Agent": USER_AGENT,
+        },
+        timeout=httpx.Timeout(connect=10.0, read=30.0, write=10.0, pool=10.0),
+    ) as client:
+        await delete_agent(client, ns.agent_id)
+        # v0.8.0: drop from local index so the picker stops listing it.
+        remove_local_agent(ns.agent_id)
+        print(f"deleted {ns.agent_id}")
+        return 0
+
+
+def main(argv: list[str] | None = None) -> int:
+    """Sync entry point — argparse + dispatch + error → exit-code mapping.
+
+    Mirrors ratatoskr.cli.main()'s error-routing matrix:
+      0  happy
+      10 usage error
+      11 auth error
+      20 api-failure (typed exception or generic SessionApiFailed)
+      21 network error
+    """
+    import asyncio
+    import sys
+
+    parser = _build_parser()
+    try:
+        ns = parser.parse_args(argv)
+    except SystemExit as exc:
+        return int(exc.code) if exc.code is not None else 0
+
+    handler = {
+        "define": _run_define,
+        "patch": _run_patch,
+        "delete": _run_delete,
+    }[ns.cmd]
+
+    try:
+        return asyncio.run(handler(ns))
+    except _Tier3UsageError as exc:
+        sys.stderr.write(f"[usage_error] {exc}\n")
+        return 10
+    except _Tier3AuthError as exc:
+        sys.stderr.write(f"[auth_error] {exc}\n")
+        return 11
+    except AssertionError as exc:
+        sys.stderr.write(f"[usage_error] {exc}\n")
+        return 10
+    except Tier3QuotaExceeded as exc:
+        sys.stderr.write(f"[quota_exceeded] retry_after={exc.retry_after}\n")
+        return 20
+    except Tier3UserIdUnsupported:
+        sys.stderr.write("[user_id_unsupported]\n")
+        return 20
+    except Tier3AgentNotFound as exc:
+        sys.stderr.write(f"[agent_not_found] {exc.agent_id}\n")
+        return 20
+    except Tier3FieldNotMutable as exc:
+        sys.stderr.write(f"[field_not_mutable] field={exc.field}\n")
+        return 20
+    except Tier3LayerDeferred as exc:
+        sys.stderr.write(f"[layer_deferred] field={exc.field}\n")
+        return 20
+    except SessionApiFailed as exc:
+        sys.stderr.write(
+            f"[api_failed] status={exc.status} body={exc.body!r}\n"
+        )
+        return 20
+    except (httpx.ConnectError, httpx.ReadTimeout, httpx.TransportError) as exc:
+        sys.stderr.write(f"[network_error] {type(exc).__name__}: {exc}\n")
+        return 21
+
+
+if __name__ == "__main__":
+    import sys
+
+    sys.exit(main())

src/ratatoskr/tui.py

@@ -10,13 +10,13 @@ from __future__ import annotations
 
 import asyncio
 import sys
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from typing import ClassVar, Literal
 
 import httpx
 from textual.app import App, ComposeResult
 from textual.binding import Binding
-from textual.containers import Horizontal, Vertical
+from textual.containers import Horizontal, Vertical, VerticalScroll
 from textual.theme import Theme
 from textual.widgets import (
     Footer,
@@ -186,18 +186,28 @@ class TuiPresenterState:
     """
 
     thinking_open: bool = False
-    # v0.6.0: per-turn streaming text buffer. Text deltas accumulate here
-    # and update `current_text` Static in place — no per-token RichLog spam.
-    text_buffer: list[str] = field(default_factory=list)
     # Thinking-run counter for turn-scoped start/end markers.
     thinking_run_index: int = 0
+    # v0.7.1: thinking-content accumulator. Worldtree emits Thinking deltas
+    # at token granularity; flushing each delta as its own RichLog line
+    # produces per-token-per-newline visual spam. Buffer here and flush
+    # only on `\n` boundaries (one written line per natural paragraph) or
+    # when the run closes (any leftover tail).
+    thinking_chunk_buffer: str = ""
+    # v0.9.0: Text accumulator for live Markdown rendering. Worldtree emits
+    # Text deltas at token granularity; each delta appends to this buffer
+    # and the current_response_widget re-renders Markdown(text_chunk_buffer)
+    # in place. On terminal event the widget is finalized + reference clears.
+    text_chunk_buffer: str = ""
+    # v0.9.0: reference to the Static widget holding the current turn's
+    # response Markdown Renderable. None between turns.
+    current_response_widget: object = None
 
     def render(
         self,
         event: Event,
         *,
-        log: RichLog,
-        current_text: Static,
+        transcript: "VerticalScroll",
         tools_log: RichLog,
         debug_log: RichLog,
         thinking_log: RichLog,
@@ -205,18 +215,14 @@ class TuiPresenterState:
     ) -> None:
         """Render one Worldtree SSE event with the TUI hierarchy + coalescing.
 
-        v0.6.5 routing:
-          - `log` (transcript) = content only: user-prompt echo (written
-            outside the presenter), terminal labels, post-Done Markdown body.
-          - `current_text` (Static below transcript) = live-streaming Text
-            deltas accumulated into one growing line; cleared on terminal.
-          - `tools_log` = ToolStart + ToolResult.
-          - `debug_log` = WorkerPhase + TextBoundary.
-          - `thinking_log` = streaming Thinking deltas inline (each chunk =
-            one line in the scrollable log). Rule(start)/Rule(end) markers
-            wrap each run. The whole pane scrolls naturally — no separate
-            tail-scrolling Static at the bottom (v0.6.5 removed
-            `thinking-current`).
+        v0.9.0 routing:
+          - `transcript` (VerticalScroll) = chat content: each turn mounts
+            child widgets (turn-header / prompt-echo / response Markdown /
+            done-label). Live Markdown rendering during Text streaming.
+          - `tools_log` (RichLog) = ToolStart + ToolResult.
+          - `debug_log` (RichLog) = WorkerPhase + TextBoundary.
+          - `thinking_log` (RichLog) = streaming Thinking deltas inline
+            (coalesced on `\n`); Rule(start)/Rule(end) wrap each run.
 
         Exceptions caught at the presenter boundary (INV-009 fallback).
         """
@@ -234,10 +240,11 @@ class TuiPresenterState:
             return RichText(s, style=_AU_DEMOTED)
 
         try:
-            # v0.6.5: Thinking deltas stream directly into thinking_log.
-            # First delta of a run writes the Rule(start) header; each
-            # subsequent delta writes its content as a line; the run closes
-            # on the next non-thinking event with a Rule(end).
+            # v0.7.1: Thinking deltas coalesce by newline before flushing.
+            # Worldtree emits Thinking events at token granularity; per-delta
+            # RichLog writes produce one visual line per token (per-token-per-
+            # newline spam). Buffer the deltas and flush only on `\n` (one
+            # written line per natural paragraph) or run close.
             if isinstance(event, Thinking):
                 from rich.rule import Rule
 
@@ -249,13 +256,24 @@ class TuiPresenterState:
                         style=_AU_DEMOTED,
                     ))
                     self.thinking_open = True
-                # Stream the delta content (chunk-of-tokens) as one line.
-                thinking_log.write(event.content)
+                self.thinking_chunk_buffer += event.content
+                # Flush every complete line in the buffer. Whatever's after
+                # the final `\n` stays buffered for the next delta or close.
+                while "\n" in self.thinking_chunk_buffer:
+                    line, _, rest = self.thinking_chunk_buffer.partition("\n")
+                    if line:  # skip empty lines (blank paragraph separators)
+                        thinking_log.write(line)
+                    self.thinking_chunk_buffer = rest
                 return
             # Non-thinking event: close any open thinking run with Rule(end).
             if self.thinking_open:
                 from rich.rule import Rule
 
+                # Flush the tail (content with no trailing `\n`) before the
+                # end rule so nothing gets lost on close.
+                if self.thinking_chunk_buffer:
+                    thinking_log.write(self.thinking_chunk_buffer)
+                    self.thinking_chunk_buffer = ""
                 turn_id = event.sse_id.turn_id if hasattr(event, "sse_id") else (
                     event.turn_id if hasattr(event, "turn_id") else "?"
                 )
@@ -266,51 +284,77 @@ class TuiPresenterState:
                 self.thinking_open = False
             # Now render the non-thinking event itself.
             if isinstance(event, Text):
-                # v0.6.0: streaming text accumulates into current_text Static
-                # — one growing live line, NOT per-delta RichLog entries.
-                self.text_buffer.append(event.content)
-                current_text.update("".join(self.text_buffer))
+                # v0.8.1: stream Text deltas into transcript directly,
+                # coalesced on `\n`. Same pattern as Thinking (v0.7.1).
+                # The pre-v0.8.1 #current-text Static is gone — its dock-
+                # bottom growth was overlapping the transcript visually.
+                #
+                # v0.9.0: Text deltas accumulate in text_chunk_buffer and
+                # the current_response_widget renders Markdown(buffer) in
+                # place. First Text delta of the turn mounts a fresh Static
+                # holding the Markdown Renderable; subsequent deltas update
+                # the same widget. Live markdown rendering — no post-Done
+                # re-render needed.
+                from rich.markdown import Markdown
+
+                self.text_chunk_buffer += event.content
+                # --raw bypasses Markdown rendering — useful for debugging
+                # the raw text stream surface, and matches the pre-v0.9.0
+                # --raw semantics (which dropped the post-Done Markdown re-
+                # render). In raw mode the response widget holds plain str.
+                rendered = (
+                    self.text_chunk_buffer if raw else Markdown(self.text_chunk_buffer)
+                )
+                if self.current_response_widget is None:
+                    self.current_response_widget = Static(
+                        rendered, classes="response-md"
+                    )
+                    transcript.mount(self.current_response_widget)
+                else:
+                    self.current_response_widget.update(rendered)
+                transcript.scroll_end(animate=False)
                 return
             if isinstance(event, (Done, Error, Cancelled)):
-                # Terminal event: clear the streaming Static first so the
-                # live-preview band collapses. Then write the colored label
-                # + (non-raw) Markdown body / (raw) accumulated plain text
-                # to the transcript.
-                accumulated = "".join(self.text_buffer)
-                self.text_buffer.clear()
-                current_text.update("")
-                # Terminal labels tinted per outcome (Aurora green / Dawn red
-                # / Dawn yellow) for at-a-glance scanning.
+                # Terminal event: finalize the response widget (clear ref so
+                # the next turn mounts a fresh one). The accumulated text is
+                # already rendered as Markdown in the widget — no post-Done
+                # re-render, no double-print.
+                self.text_chunk_buffer = ""
+                self.current_response_widget = None
+                # Terminal labels mount as styled Statics. Tinted per outcome
+                # (Aurora green / Dawn red / Dawn yellow) for at-a-glance
+                # scanning.
                 if isinstance(event, Done):
-                    log.write(RichText(
-                        f"[done] turn_id={event.sse_id.turn_id} model={event.model} "
-                        f"duration={_format_duration_ms(event.duration_ms)} "
-                        f"usage {_format_usage(event.usage, arrow='→')}",
-                        style=_AU_SUCCESS,
+                    transcript.mount(Static(
+                        RichText(
+                            f"[done] turn_id={event.sse_id.turn_id} "
+                            f"model={event.model} "
+                            f"duration={_format_duration_ms(event.duration_ms)} "
+                            f"usage {_format_usage(event.usage, arrow='→')}",
+                            style=_AU_SUCCESS,
+                        ),
+                        classes="done-label",
                     ))
-                    if raw:
-                        # Raw mode: emit the accumulated streamed text verbatim
-                        # so the operator has a record after the Static clears.
-                        if accumulated:
-                            log.write(accumulated)
-                    else:
-                        from rich.markdown import Markdown
-                        from rich.rule import Rule
-
-                        log.write(Rule(style=_AU_DEMOTED))
-                        log.write(Markdown(event.response))
                 elif isinstance(event, Error):
-                    log.write(RichText(
-                        f"[error] turn_id={event.sse_id.turn_id} code={event.error_code} "
-                        f"message={event.message!r}",
-                        style=_AU_ERROR,
+                    transcript.mount(Static(
+                        RichText(
+                            f"[error] turn_id={event.sse_id.turn_id} "
+                            f"code={event.error_code} message={event.message!r}",
+                            style=_AU_ERROR,
+                        ),
+                        classes="error-label",
                     ))
                 else:  # Cancelled
-                    log.write(RichText(
-                        f"[cancelled] turn_id={event.turn_id} reason={event.reason!r} "
-                        f"partial_message_id={event.partial_message_id}",
-                        style=_AU_WARNING,
+                    transcript.mount(Static(
+                        RichText(
+                            f"[cancelled] turn_id={event.turn_id} "
+                            f"reason={event.reason!r} "
+                            f"partial_message_id={event.partial_message_id}",
+                            style=_AU_WARNING,
+                        ),
+                        classes="cancelled-label",
                     ))
+                transcript.scroll_end(animate=False)
                 return
             if isinstance(event, WorkerPhase):
                 # v0.5.0: telemetry → Debug pane, not transcript.
@@ -342,22 +386,24 @@ class TuiPresenterState:
             # the original event AND a render_error line with the class name only
             # (NO exception message — security clause). Volva F1 fix.
             #
-            # v0.6.0 routing-under-failure preservation — fallback writes go
-            # to the same destination the successful render would have used:
-            #   - ToolStart/ToolResult       → tools_log
-            #   - Thinking                   → thinking_log
-            #   - WorkerPhase/TextBoundary   → debug_log
-            #   - everything else            → log
+            # v0.9.0 routing-under-failure: panes (RichLog) still write Strip
+            # lines; transcript (VerticalScroll) mounts a Static instead.
             if isinstance(event, (ToolStart, ToolResult)):
-                target = tools_log
+                tools_log.write(_plain_label(event))
+                tools_log.write(f"[render_error] {type(exc).__name__}")
             elif isinstance(event, Thinking):
-                target = thinking_log
+                thinking_log.write(_plain_label(event))
+                thinking_log.write(f"[render_error] {type(exc).__name__}")
             elif isinstance(event, (WorkerPhase, TextBoundary)):
-                target = debug_log
+                debug_log.write(_plain_label(event))
+                debug_log.write(f"[render_error] {type(exc).__name__}")
             else:
-                target = log
-            target.write(_plain_label(event))
-            target.write(f"[render_error] {type(exc).__name__}")
+                # Transcript-bound event (Text / Done / Error / Cancelled).
+                transcript.mount(Static(_plain_label(event), classes="error-label"))
+                transcript.mount(
+                    Static(f"[render_error] {type(exc).__name__}", classes="error-label")
+                )
+                transcript.scroll_end(animate=False)
 
 
 class AgentPickerApp(App[str | None]):
@@ -546,21 +592,43 @@ class RatatoskrApp(App[int]):
     }
     /* v0.6.5: thinking-current Static removed; thinking now streams
        directly into thinking-log so the whole pane scrolls naturally. */
-    #transcript {
+    /* v0.9.0: transcript is a VerticalScroll container holding dynamically
+       mounted Statics + Markdown widgets per turn. Live Markdown rendering
+       replaces the v0.8.x RichLog approach which couldn't render Markdown
+       in-flight (only on Done as a re-render → double-print bug). */
+    #transcript-scroll {
         height: 1fr;
         background: $background;
         padding: 0 1;
     }
-    /* v0.6.0: streaming-text Static carries in-flight assistant tokens.
-       Replaces per-token RichLog spam — one growing line that updates in
-       place. Cleared on terminal event; final Markdown body lands in the
-       transcript. */
-    #current-text {
-        dock: bottom;
+    /* Per-turn mounted widgets carry id-prefix conventions:
+       - .turn-header   "── turn N ──" (dim)
+       - .prompt-echo   "❯ user input"  (aurora bright cyan)
+       - .response-md   Markdown(accumulated_text) — updated live
+       - .done-label    "[done] turn_id=…" (aurora green)
+       - .error-label   "[error] …"        (dawn red)
+       - .cancelled-label "[cancelled] …"  (dawn yellow)
+    */
+    .turn-header {
+        height: auto;
+        padding: 0 1;
+        color: $au-dark-60;
+    }
+    .prompt-echo {
         height: auto;
-        background: $background;
         padding: 0 1;
     }
+    .response-md {
+        height: auto;
+        padding: 0 1;
+    }
+    .done-label, .error-label, .cancelled-label {
+        height: auto;
+        padding: 0 1;
+    }
+    /* v0.8.1: #current-text Static removed. Streaming text now coalesces
+       on `\n` and writes directly to #transcript (same pattern as v0.7.1
+       thinking fix). Eliminates the dock-bottom-growth-overlap bug. */
     #tools-log, #debug-log, #thinking-log {
         background: $background;
         padding: 0 1;
@@ -659,8 +727,12 @@ class RatatoskrApp(App[int]):
         # work without widget-level markup=True.
         with Horizontal(id="main-row"):
             with Vertical(id="left-column"):
-                yield RichLog(id="transcript", wrap=True, markup=False, highlight=False)
-                yield Static("", id="current-text")
+                # v0.9.0: transcript is a VerticalScroll holding per-turn
+                # mounted widgets (turn header, prompt echo, response Markdown,
+                # done label). Live Markdown rendering happens via Static
+                # widgets holding `Markdown` Renderables, updated as Text
+                # deltas arrive.
+                yield VerticalScroll(id="transcript-scroll")
                 yield Input(id="prompt", placeholder="Type a message and press Enter")
             with Vertical(id="right-column"):
                 with TabbedContent(id="side-panes"):
@@ -723,20 +795,30 @@ class RatatoskrApp(App[int]):
         self._set_hint(self.HINT_IDLE)
 
     def _write_turn_headers(self, turn_id: int) -> None:
-        """v0.6.0: Write `── turn N ──` Rule headers across every pane so
-        operators can visually correlate sections during cross-pane
-        debugging. Called from `_stream_turn_worker` on first event of
-        each new turn (idempotent per turn via active_turn_id guard).
+        """v0.6.0: turn-ID headers across every pane for cross-pane
+        correlation. v0.9.0: transcript is a VerticalScroll; mounts a
+        Static with rule-style text instead of writing a Rule Renderable
+        to RichLog. Other panes still use RichLog.write(Rule).
         """
         from rich.rule import Rule
+        from rich.text import Text as RichText
 
         title = f"turn {turn_id}"
         rule = Rule(title=title, style=_AU_DEMOTED)
         try:
-            self.query_one("#transcript", RichLog).write(rule)
+            # Transcript (VerticalScroll): mount a styled Static.
+            transcript = self.query_one("#transcript-scroll", VerticalScroll)
+            transcript.mount(
+                Static(
+                    RichText(f"── turn {turn_id} ──", style=_AU_DEMOTED),
+                    classes="turn-header",
+                )
+            )
+            # Other panes (RichLog): write the Rule Renderable.
             self.query_one("#tools-log", RichLog).write(rule)
             self.query_one("#debug-log", RichLog).write(rule)
             self.query_one("#thinking-log", RichLog).write(rule)
+            transcript.scroll_end(animate=False)
         except Exception:
             # Defensive: widget tree may be tearing down — never let a
             # turn-header write block the SSE consumer.
@@ -752,12 +834,19 @@ class RatatoskrApp(App[int]):
             pass
 
     async def on_input_submitted(self, event: Input.Submitted) -> None:
-        """Echo user prompt, spawn stream worker; busy notice if not idle."""
+        """Echo user prompt, spawn stream worker; busy notice if not idle.
+
+        v0.9.0: prompt echo mounts as a Static in the transcript VerticalScroll
+        (was log.write to RichLog).
+        """
         if event.input.id != "prompt":
             return
-        log = self.query_one("#transcript", RichLog)
+        transcript = self.query_one("#transcript-scroll", VerticalScroll)
         if self.state != "idle":
-            log.write("[busy] turn in flight; input ignored")
+            transcript.mount(
+                Static("[busy] turn in flight; input ignored", classes="error-label")
+            )
+            transcript.scroll_end(animate=False)
             event.input.value = ""
             return
         content = event.input.value.strip()
@@ -766,7 +855,13 @@ class RatatoskrApp(App[int]):
         # v0.4.1 retheme: operator's voice gets Australis bright cyan so it
         # stands out against the default-foreground assistant text below it.
         from rich.text import Text as RichText
-        log.write(RichText(f"❯ {content}", style=_AU_USER_ECHO))  # noqa: RUF001
+        transcript.mount(
+            Static(
+                RichText(f"❯ {content}", style=_AU_USER_ECHO),  # noqa: RUF001
+                classes="prompt-echo",
+            )
+        )
+        transcript.scroll_end(animate=False)
         event.input.value = ""
         self.state = "streaming"
         self._set_hint(self.HINT_STREAMING)
@@ -775,28 +870,37 @@ class RatatoskrApp(App[int]):
         )
 
     async def _stream_turn_worker(self, content: str) -> None:
-        """Drive stream_turn, render events via TuiPresenterState (issue #12)."""
+        """Drive stream_turn, render events via TuiPresenterState.
+
+        v0.9.0: transcript is a VerticalScroll; the presenter's `transcript`
+        argument is the container, and the presenter mounts Static / Markdown-
+        backed widgets directly. Wire-error labels mount as `error-label`
+        Statics into the transcript-scroll.
+        """
         assert self.state == "streaming"
         assert self.client is not None
         assert content
-        log = self.query_one("#transcript", RichLog)
-        current_text = self.query_one("#current-text", Static)
+        transcript = self.query_one("#transcript-scroll", VerticalScroll)
         tools_log = self.query_one("#tools-log", RichLog)
         debug_log = self.query_one("#debug-log", RichLog)
         thinking_log = self.query_one("#thinking-log", RichLog)
         presenter = TuiPresenterState()
+
+        def _mount_wire_error(label: str) -> None:
+            try:
+                transcript.mount(Static(label, classes="error-label"))
+                transcript.scroll_end(animate=False)
+            except Exception:
+                pass
+
         try:
             async for event in stream_turn(self.client, self.session_id, content):
                 if self.active_turn_id is None:
                     self.active_turn_id = event.sse_id.turn_id
-                    # v0.6.0: turn-ID headers across all panes so the
-                    # operator can visually correlate sections during
-                    # cross-pane debugging.
                     self._write_turn_headers(self.active_turn_id)
                 presenter.render(
                     event,
-                    log=log,
-                    current_text=current_text,
+                    transcript=transcript,
                     tools_log=tools_log,
                     debug_log=debug_log,
                     thinking_log=thinking_log,
@@ -805,15 +909,15 @@ class RatatoskrApp(App[int]):
                 if isinstance(event, (Done, Error, Cancelled)):
                     break
         except SseConnectFailed as exc:
-            log.write(f"[sse_connect_failed] status={exc.status} body={exc.body!r}")
+            _mount_wire_error(f"[sse_connect_failed] status={exc.status} body={exc.body!r}")
         except SseConnectionDropped as exc:
-            log.write(f"[connection_dropped] last_seen={exc.last_seen_sse_id}")
+            _mount_wire_error(f"[connection_dropped] last_seen={exc.last_seen_sse_id}")
         except MalformedSseId as exc:
-            log.write(f"[malformed_sse_id] raw={exc.raw!r}")
+            _mount_wire_error(f"[malformed_sse_id] raw={exc.raw!r}")
         except MalformedSseData as exc:
-            log.write(f"[malformed_sse_data] raw={exc.raw!r}")
+            _mount_wire_error(f"[malformed_sse_data] raw={exc.raw!r}")
         except TurnIdFlip as exc:
-            log.write(f"[turn_id_flip] expected={exc.established} got={exc.got}")
+            _mount_wire_error(f"[turn_id_flip] expected={exc.established} got={exc.got}")
         finally:
             self.state = "idle"
             self.active_turn_id = None
@@ -836,9 +940,12 @@ class RatatoskrApp(App[int]):
                 return
             self.state = "cancelling"
             self._set_hint(self.HINT_CANCELLING)
-            log = self.query_one("#transcript", RichLog)
+            transcript = self.query_one("#transcript-scroll", VerticalScroll)
             self.run_worker(
-                _cancel_via_sse(self.client, self.session_id, self.active_turn_id, log=log)
+                _cancel_via_sse(
+                    self.client, self.session_id, self.active_turn_id,
+                    transcript=transcript,
+                )
             )
         elif self.state == "cancelling":
             if self.stream_worker is not None:
@@ -910,6 +1017,14 @@ async def _resolve_then_run(args: ParsedArgs) -> int:
         # Issue #8: startup agent picker — fetch GET /agents and prompt when
         # --new is passed without --agent. list_agents errors land on real
         # stderr before any alt-screen opens (preserves issue #6 INV-001).
+        #
+        # v0.8.0: merge in local tier-3 agent index. Worldtree's GET /agents
+        # doesn't return consumer-defined agents (issue #15 smoke finding);
+        # ratatoskr keeps its own JSON-backed index of agents the operator
+        # defined via `python -m ratatoskr.tier3 define`. Merged here so the
+        # picker shows foundational + local-tier-3 in one list. Dedup by
+        # agent_id (remote wins on conflict, since a server-listed agent
+        # is the authoritative source).
         chosen_agent_id: str | None = args.agent_id
         if args.new and args.agent_id is None:
             try:
@@ -922,6 +1037,23 @@ async def _resolve_then_run(args: ParsedArgs) -> int:
             except (httpx.ConnectError, httpx.ReadTimeout, httpx.TransportError) as exc:
                 sys.stderr.write(f"[network_error] {type(exc).__name__}: {exc}\n")
                 return 21
+            # v0.8.0: append local tier-3 entries not already in the remote list.
+            from ratatoskr.local_agents import load_local_agents
+
+            remote_ids = {a.agent_id for a in agents}
+            for entry in load_local_agents():
+                if entry.agent_id in remote_ids:
+                    continue
+                agents.append(AgentInfo(
+                    agent_id=entry.agent_id,
+                    name=entry.agent_name,
+                    description=entry.description,
+                    version=None,
+                    capabilities=[],
+                    supported_models=[],
+                    persona_traits={},
+                    ui_hints={},
+                ))
             if not agents:
                 sys.stderr.write("[no_agents] server returned empty agent list\n")
                 return 13
@@ -962,12 +1094,23 @@ async def _cancel_via_sse(
     session_id: str,
     turn_id: int,
     *,
-    log: RichLog,
+    transcript: VerticalScroll,
 ) -> None:
-    """Fire-and-forget cancel; never raises (mirrors cli._cancel_and_log; #3 INV-009)."""
+    """Fire-and-forget cancel; never raises (mirrors cli._cancel_and_log; #3 INV-009).
+
+    v0.9.0: mounts a `[cancel_failed]` Static into the transcript-scroll
+    container on failure (was log.write to RichLog).
+    """
     assert client is not None
     assert isinstance(turn_id, int) and turn_id > 0
     try:
         await cancel_turn(client, session_id, turn_id)
     except (CancelFailed, CancelTurnNotFound, CancelAlreadyCompleted, httpx.RequestError) as exc:
-        log.write(f"[cancel_failed] {type(exc).__name__}: {exc}")
+        try:
+            transcript.mount(Static(
+                f"[cancel_failed] {type(exc).__name__}: {exc}",
+                classes="error-label",
+            ))
+            transcript.scroll_end(animate=False)
+        except Exception:
+            pass

tests/test_local_agents.py

@@ -0,0 +1,187 @@
+"""Tests for ratatoskr.local_agents.
+
+Use ``$RATATOSKR_LOCAL_AGENTS`` env-var override + pytest tmp_path to
+isolate from the operator's real ``~/.config/ratatoskr/local_agents.json``.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+
+from ratatoskr.local_agents import (
+    LocalAgentEntry,
+    _local_agents_path,
+    add_local_agent,
+    load_local_agents,
+    make_description,
+    remove_local_agent,
+    update_local_agent,
+)
+
+
+@pytest.fixture
+def local_path(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Path:
+    """Point $RATATOSKR_LOCAL_AGENTS at a fresh tmp file for the test."""
+    path = tmp_path / "local_agents.json"
+    monkeypatch.setenv("RATATOSKR_LOCAL_AGENTS", str(path))
+    return path
+
+
+def _entry(
+    agent_id: str = "ratatoskr:wizard",
+    agent_name: str = "wizard",
+    model: str = "qwen3.6-35-a3b",
+    description: str = "(tier 3) test agent",
+    defined_at: str = "2026-05-25T00:00:00+00:00",
+) -> LocalAgentEntry:
+    return LocalAgentEntry(
+        agent_id=agent_id,
+        agent_name=agent_name,
+        model=model,
+        description=description,
+        defined_at=defined_at,
+    )
+
+
+class TestPathResolution:
+    def test_env_override(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.setenv("RATATOSKR_LOCAL_AGENTS", "/tmp/custom-agents.json")
+        assert _local_agents_path() == Path("/tmp/custom-agents.json")
+
+    def test_xdg_config_home(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.delenv("RATATOSKR_LOCAL_AGENTS", raising=False)
+        monkeypatch.setenv("XDG_CONFIG_HOME", "/tmp/xdg-config")
+        assert (
+            _local_agents_path()
+            == Path("/tmp/xdg-config/ratatoskr/local_agents.json")
+        )
+
+    def test_default_home(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.delenv("RATATOSKR_LOCAL_AGENTS", raising=False)
+        monkeypatch.delenv("XDG_CONFIG_HOME", raising=False)
+        path = _local_agents_path()
+        assert path == Path.home() / ".config" / "ratatoskr" / "local_agents.json"
+
+
+class TestLoadEmpty:
+    def test_missing_file_returns_empty(self, local_path: Path) -> None:
+        assert not local_path.exists()
+        assert load_local_agents() == []
+
+    def test_corrupt_json_returns_empty(self, local_path: Path) -> None:
+        local_path.parent.mkdir(parents=True, exist_ok=True)
+        local_path.write_text("not json at all")
+        assert load_local_agents() == []
+
+    def test_wrong_schema_version_returns_empty(self, local_path: Path) -> None:
+        local_path.parent.mkdir(parents=True, exist_ok=True)
+        local_path.write_text(json.dumps({"version": 999, "agents": []}))
+        assert load_local_agents() == []
+
+    def test_missing_version_key_returns_empty(self, local_path: Path) -> None:
+        local_path.parent.mkdir(parents=True, exist_ok=True)
+        local_path.write_text(json.dumps({"agents": []}))
+        assert load_local_agents() == []
+
+    def test_malformed_row_skipped(self, local_path: Path) -> None:
+        local_path.parent.mkdir(parents=True, exist_ok=True)
+        local_path.write_text(
+            json.dumps(
+                {
+                    "version": 1,
+                    "agents": [
+                        {"agent_id": "incomplete"},  # missing required fields
+                        {
+                            "agent_id": "ratatoskr:good",
+                            "agent_name": "good",
+                            "model": "m",
+                            "description": "d",
+                            "defined_at": "t",
+                        },
+                    ],
+                }
+            )
+        )
+        entries = load_local_agents()
+        assert len(entries) == 1
+        assert entries[0].agent_id == "ratatoskr:good"
+
+
+class TestAdd:
+    def test_add_one(self, local_path: Path) -> None:
+        add_local_agent(_entry())
+        entries = load_local_agents()
+        assert len(entries) == 1
+        assert entries[0].agent_id == "ratatoskr:wizard"
+
+    def test_add_two_different(self, local_path: Path) -> None:
+        add_local_agent(_entry(agent_id="ratatoskr:a", agent_name="a"))
+        add_local_agent(_entry(agent_id="ratatoskr:b", agent_name="b"))
+        ids = {e.agent_id for e in load_local_agents()}
+        assert ids == {"ratatoskr:a", "ratatoskr:b"}
+
+    def test_add_replaces_same_id(self, local_path: Path) -> None:
+        add_local_agent(_entry(model="old-model"))
+        add_local_agent(_entry(model="new-model"))
+        entries = load_local_agents()
+        assert len(entries) == 1
+        assert entries[0].model == "new-model"
+
+    def test_creates_parent_dirs(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        nested = tmp_path / "deep" / "nested" / "path" / "agents.json"
+        monkeypatch.setenv("RATATOSKR_LOCAL_AGENTS", str(nested))
+        add_local_agent(_entry())
+        assert nested.exists()
+
+
+class TestUpdate:
+    def test_update_changes_existing(self, local_path: Path) -> None:
+        add_local_agent(_entry(model="v1"))
+        update_local_agent(_entry(model="v2"))
+        entries = load_local_agents()
+        assert len(entries) == 1
+        assert entries[0].model == "v2"
+
+
+class TestRemove:
+    def test_remove_existing(self, local_path: Path) -> None:
+        add_local_agent(_entry())
+        remove_local_agent("ratatoskr:wizard")
+        assert load_local_agents() == []
+
+    def test_remove_missing_is_noop(self, local_path: Path) -> None:
+        add_local_agent(_entry())
+        remove_local_agent("ratatoskr:doesnotexist")
+        assert len(load_local_agents()) == 1
+
+
+class TestMakeDescription:
+    def test_first_nonempty_line(self) -> None:
+        prompt = "\n\n# IDENTITY\nYou are a test agent..."
+        desc = make_description(prompt)
+        assert desc.startswith("(tier 3) IDENTITY")
+
+    def test_strips_heading_markers(self) -> None:
+        prompt = "# A nice heading\nMore prompt..."
+        desc = make_description(prompt)
+        assert "(tier 3) A nice heading" == desc
+
+    def test_truncates_long(self) -> None:
+        prompt = "x" * 200
+        desc = make_description(prompt)
+        # 80 char cap including the prefix
+        assert len(desc) == 81  # 80 + ellipsis char
+        assert desc.endswith("…")
+
+    def test_empty_prompt_fallback(self) -> None:
+        desc = make_description("")
+        assert desc == "(tier 3) custom system prompt"
+
+    def test_whitespace_only_fallback(self) -> None:
+        desc = make_description("   \n\n  ")
+        assert desc == "(tier 3) custom system prompt"

tests/test_sse_client.py

@@ -711,6 +711,47 @@ def _sse_raw_chunk(sse_id: str, raw_data: str) -> bytes:
     return f"id: {sse_id}\ndata: {raw_data}\n\n".encode()
 
 
+def _sse_no_id_chunk(data: str) -> bytes:
+    """SSE frame with NO id line + arbitrary data (v0.8.1: keepalive shape)."""
+    return f"data: {data}\n\n".encode()
+
+
+class TestEmptyIdSkipped:
+    @respx.mock
+    async def test_empty_id_on_first_event_skipped(self) -> None:
+        """empty_id_on_first_event_skipped [v0.8.1]: stream starts with an
+        event carrying NO `id:` line → httpx_sse exposes sse.id == ''
+        (no prior id to inherit). Pre-v0.8.1: MalformedSseId raw='' crashed
+        the turn. v0.8.1: treat same as empty-data keepalive — skip silently.
+
+        Observed 2026-05-25 on Worldtree's qwen3.6-35-a3b-heretic provider:
+        the first stream frame had no id line, every turn died with
+        `[malformed_sse_id] raw=''`.
+        """
+        from ratatoskr.sse_client import Done as _Done
+        from ratatoskr.sse_client import Text as _Text
+
+        # First frame: no id line (httpx_sse → sse.id = ""). Skip it.
+        # Subsequent frames have ids; normal processing resumes.
+        stream = (
+            _sse_no_id_chunk('{"type":"keepalive"}')  # ← skipped (sse.id == "")
+            + _sse_chunk("42:1", {"type": "text", "content": "first"})
+            + _sse_chunk("42:2", _DONE_42_6)
+        )
+        respx.post("https://w.example/sessions/s1/messages").mock(
+            return_value=httpx.Response(
+                200, headers={"content-type": "text/event-stream"}, content=stream
+            )
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            events = [e async for e in stream_turn(client, "s1", "hi")]
+        # 2 events — the no-id frame is invisible (no MalformedSseId crash).
+        assert len(events) == 2
+        assert isinstance(events[0], _Text)
+        assert events[0].content == "first"
+        assert isinstance(events[1], _Done)
+
+
 class TestEmptyDataSkipped:
     @respx.mock
     async def test_empty_data_skipped(self) -> None:

tests/test_tier3.py

@@ -0,0 +1,491 @@
+"""Tests for ratatoskr.tier3 per docs/contracts/issues/15.contract.md."""
+
+from pathlib import Path
+
+import httpx
+import pytest
+import respx
+
+from ratatoskr.sessions import SessionApiFailed
+from ratatoskr.tier3 import (
+    Tier3AgentInfo,
+    Tier3AgentNotFound,
+    Tier3FieldNotMutable,
+    Tier3LayerDeferred,
+    Tier3QuotaExceeded,
+    Tier3UserIdUnsupported,
+    define_agent,
+    delete_agent,
+    main,
+    patch_agent,
+)
+
+_FULL_AGENT_RESP = {
+    "agent_id": "ratatoskr:wizard",
+    "user_id": "ratatoskr",
+    "agent_name": "wizard",
+    "system_prompt": "You are a wizard.",
+    "model": "qwen3.6-35-a3b",
+    "created_at": "2026-05-25T03:20:09.703601+00:00",
+    "updated_at": "2026-05-25T03:20:09.703601+00:00",
+}
+
+
+class TestDefineAgent:
+    @respx.mock
+    async def test_happy_define(self) -> None:
+        """happy_define [happy,tracer]: 201 → fully populated Tier3AgentInfo."""
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(201, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            info = await define_agent(
+                client,
+                agent_name="wizard",
+                system_prompt="You are a wizard.",
+                model="qwen3.6-35-a3b",
+            )
+        assert isinstance(info, Tier3AgentInfo)
+        assert info.agent_id == "ratatoskr:wizard"
+        assert info.user_id == "ratatoskr"
+        assert info.agent_name == "wizard"
+        assert info.model == "qwen3.6-35-a3b"
+
+    @respx.mock
+    async def test_request_body_shape(self) -> None:
+        """request_body_shape [trace]: outbound JSON is exactly the three keys."""
+        import json as _json
+
+        route = respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(201, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            await define_agent(
+                client,
+                agent_name="wizard",
+                system_prompt="You are a wizard.",
+                model="qwen3.6-35-a3b",
+            )
+        body = _json.loads(route.calls[0].request.content)
+        # INV-001: exactly these three keys — no layer fields, no metadata.
+        assert body == {
+            "agent_name": "wizard",
+            "system_prompt": "You are a wizard.",
+            "model": "qwen3.6-35-a3b",
+        }
+
+    @respx.mock
+    async def test_quota_exceeded(self) -> None:
+        """quota_exceeded [error]: 429 + Retry-After → Tier3QuotaExceeded."""
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(
+                429,
+                headers={"Retry-After": "0"},
+                json={"detail": {"error_code": "agent_quota_exceeded"}},
+            )
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(Tier3QuotaExceeded) as exc:
+                await define_agent(
+                    client,
+                    agent_name="overflow",
+                    system_prompt="x",
+                    model="m",
+                )
+        assert exc.value.retry_after == 0
+
+    @respx.mock
+    async def test_user_id_unsupported(self) -> None:
+        """user_id_unsupported [error]: 403 + error_code → Tier3UserIdUnsupported."""
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(
+                403, json={"detail": {"error_code": "tier3_user_id_unsupported"}}
+            )
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(Tier3UserIdUnsupported):
+                await define_agent(
+                    client, agent_name="wizard", system_prompt="x", model="m"
+                )
+
+    @respx.mock
+    async def test_layer_deferred(self) -> None:
+        """layer_deferred [error]: 422 + layer_deferred → Tier3LayerDeferred(field)."""
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(
+                422,
+                json={"detail": {"error_code": "layer_deferred", "field": "persona"}},
+            )
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(Tier3LayerDeferred) as exc:
+                await define_agent(
+                    client, agent_name="wizard", system_prompt="x", model="m"
+                )
+        assert exc.value.field == "persona"
+
+    @respx.mock
+    async def test_bad_slug_assert(self) -> None:
+        """bad_slug_assert [adversarial]: agent_name with uppercase → AssertionError, no HTTP."""
+        route = respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(201, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(AssertionError):
+                await define_agent(
+                    client, agent_name="Wizard", system_prompt="x", model="m"
+                )
+        assert route.call_count == 0
+
+    @respx.mock
+    async def test_short_slug_assert(self) -> None:
+        """short_slug_assert [adversarial]: agent_name len < 3 → AssertionError."""
+        route = respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(201, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(AssertionError):
+                await define_agent(
+                    client, agent_name="ab", system_prompt="x", model="m"
+                )
+        assert route.call_count == 0
+
+    @respx.mock
+    async def test_empty_prompt_assert(self) -> None:
+        """empty_prompt_assert [adversarial]: empty system_prompt → AssertionError."""
+        route = respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(201, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(AssertionError):
+                await define_agent(
+                    client, agent_name="wizard", system_prompt="", model="m"
+                )
+        assert route.call_count == 0
+
+    @respx.mock
+    async def test_other_5xx(self) -> None:
+        """other_5xx [error]: 503 → SessionApiFailed(status=503)."""
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(503, content=b"upstream out")
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(SessionApiFailed) as exc:
+                await define_agent(
+                    client, agent_name="wizard", system_prompt="x", model="m"
+                )
+        assert exc.value.status == 503
+
+
+class TestPatchAgent:
+    @respx.mock
+    async def test_happy_patch_both_fields(self) -> None:
+        """happy_patch_both_fields: both fields set → request body has both."""
+        import json as _json
+
+        updated = {
+            **_FULL_AGENT_RESP,
+            "system_prompt": "new prompt",
+            "model": "different-model",
+        }
+        route = respx.patch("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(200, json=updated)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            info = await patch_agent(
+                client,
+                "ratatoskr:wizard",
+                system_prompt="new prompt",
+                model="different-model",
+            )
+        body = _json.loads(route.calls[0].request.content)
+        assert body == {"system_prompt": "new prompt", "model": "different-model"}
+        assert info.system_prompt == "new prompt"
+        assert info.model == "different-model"
+
+    @respx.mock
+    async def test_happy_patch_single_field(self) -> None:
+        """happy_patch_single_field: omit model → body has system_prompt only."""
+        import json as _json
+
+        updated = {**_FULL_AGENT_RESP, "system_prompt": "only this"}
+        route = respx.patch("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(200, json=updated)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            await patch_agent(client, "ratatoskr:wizard", system_prompt="only this")
+        body = _json.loads(route.calls[0].request.content)
+        # INV-002: body omits the None-valued field entirely
+        assert body == {"system_prompt": "only this"}
+
+    @respx.mock
+    async def test_field_not_mutable(self) -> None:
+        """field_not_mutable [error]: 422 + error_code → Tier3FieldNotMutable(field)."""
+        respx.patch("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(
+                422,
+                json={
+                    "detail": {"error_code": "field_not_mutable", "field": "agent_name"}
+                },
+            )
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(Tier3FieldNotMutable) as exc:
+                await patch_agent(
+                    client, "ratatoskr:wizard", system_prompt="x"
+                )
+        assert exc.value.field == "agent_name"
+
+    @respx.mock
+    async def test_404(self) -> None:
+        """404 [error]: PATCH on non-existent agent → Tier3AgentNotFound."""
+        respx.patch("https://w.example/agents/ratatoskr:ghost").mock(
+            return_value=httpx.Response(404, content=b"")
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(Tier3AgentNotFound) as exc:
+                await patch_agent(
+                    client, "ratatoskr:ghost", system_prompt="x"
+                )
+        assert exc.value.agent_id == "ratatoskr:ghost"
+
+    @respx.mock
+    async def test_no_fields_assert(self) -> None:
+        """no_fields_assert [adversarial]: both None → AssertionError, no HTTP."""
+        route = respx.patch("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(200, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(AssertionError):
+                await patch_agent(client, "ratatoskr:wizard")
+        assert route.call_count == 0
+
+    @respx.mock
+    async def test_non_tier3_id_assert(self) -> None:
+        """non_tier3_id_assert [adversarial]: agent_id without `:` → AssertionError."""
+        route = respx.patch("https://w.example/agents/mimir").mock(
+            return_value=httpx.Response(200, json=_FULL_AGENT_RESP)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(AssertionError):
+                await patch_agent(client, "mimir", system_prompt="x")
+        assert route.call_count == 0
+
+
+class TestDeleteAgent:
+    @respx.mock
+    async def test_happy_delete(self) -> None:
+        """happy_delete [happy,tracer]: 204 → returns None."""
+        respx.delete("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(204)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            result = await delete_agent(client, "ratatoskr:wizard")
+        assert result is None
+
+    @respx.mock
+    async def test_404(self) -> None:
+        """404 [error]: DELETE on non-existent agent → Tier3AgentNotFound."""
+        respx.delete("https://w.example/agents/ratatoskr:ghost").mock(
+            return_value=httpx.Response(404)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(Tier3AgentNotFound) as exc:
+                await delete_agent(client, "ratatoskr:ghost")
+        assert exc.value.agent_id == "ratatoskr:ghost"
+
+    @respx.mock
+    async def test_non_tier3_id_assert(self) -> None:
+        """non_tier3_id_assert [adversarial]: agent_id without `:` → AssertionError."""
+        route = respx.delete("https://w.example/agents/mimir").mock(
+            return_value=httpx.Response(204)
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(AssertionError):
+                await delete_agent(client, "mimir")
+        assert route.call_count == 0
+
+    @respx.mock
+    async def test_other_5xx(self) -> None:
+        """other_5xx [error]: 500 → SessionApiFailed."""
+        respx.delete("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(500, content=b"oops")
+        )
+        async with httpx.AsyncClient(base_url="https://w.example") as client:
+            with pytest.raises(SessionApiFailed) as exc:
+                await delete_agent(client, "ratatoskr:wizard")
+        assert exc.value.status == 500
+
+
+@pytest.fixture
+def _isolated_local_agents(
+    tmp_path: "Path", monkeypatch: pytest.MonkeyPatch
+) -> "Path":
+    """Isolate the v0.8.0 local-tier-3 index from the operator's real file."""
+    path = tmp_path / "local_agents.json"
+    monkeypatch.setenv("RATATOSKR_LOCAL_AGENTS", str(path))
+    return path
+
+
+class TestCli:
+    @respx.mock
+    def test_cli_define_happy(
+        self,
+        capsys: pytest.CaptureFixture[str],
+        monkeypatch: pytest.MonkeyPatch,
+        _isolated_local_agents: "Path",
+    ) -> None:
+        """cli_define_happy [happy]: argv → 201 mock → stdout confirmation;
+        local index updated with the new entry (v0.8.0 hook).
+        """
+        from ratatoskr.local_agents import load_local_agents
+
+        monkeypatch.setenv("WORLDTREE_API_URL", "https://w.example")
+        monkeypatch.setenv("WORLDTREE_API_KEY", "k")
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(201, json=_FULL_AGENT_RESP)
+        )
+        rc = main([
+            "define",
+            "--name", "wizard",
+            "--system-prompt", "You are a wizard.",
+            "--model", "qwen3.6-35-a3b",
+        ])
+        out = capsys.readouterr()
+        assert rc == 0
+        assert out.out.strip() == "defined ratatoskr:wizard (qwen3.6-35-a3b)"
+        # v0.8.0: local index now has the new entry.
+        entries = load_local_agents()
+        assert len(entries) == 1
+        assert entries[0].agent_id == "ratatoskr:wizard"
+        assert entries[0].model == "qwen3.6-35-a3b"
+
+    @respx.mock
+    def test_cli_patch_happy(
+        self,
+        capsys: pytest.CaptureFixture[str],
+        monkeypatch: pytest.MonkeyPatch,
+        _isolated_local_agents: "Path",
+    ) -> None:
+        """cli_patch_happy [happy]: argv → 200 mock → stdout confirmation;
+        local index refreshed with the post-patch state.
+        """
+        from ratatoskr.local_agents import load_local_agents
+
+        monkeypatch.setenv("WORLDTREE_API_URL", "https://w.example")
+        monkeypatch.setenv("WORLDTREE_API_KEY", "k")
+        respx.patch("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(200, json=_FULL_AGENT_RESP)
+        )
+        rc = main(["patch", "ratatoskr:wizard", "--system-prompt", "new"])
+        out = capsys.readouterr()
+        assert rc == 0
+        assert out.out.strip() == "patched ratatoskr:wizard"
+        entries = load_local_agents()
+        assert len(entries) == 1
+        assert entries[0].agent_id == "ratatoskr:wizard"
+
+    @respx.mock
+    def test_cli_delete_happy(
+        self,
+        capsys: pytest.CaptureFixture[str],
+        monkeypatch: pytest.MonkeyPatch,
+        _isolated_local_agents: "Path",
+    ) -> None:
+        """cli_delete_happy [happy]: argv → 204 mock → stdout confirmation;
+        local index entry removed (v0.8.0 hook).
+        """
+        from ratatoskr.local_agents import (
+            LocalAgentEntry,
+            add_local_agent,
+            load_local_agents,
+        )
+
+        # Pre-populate so we can verify removal.
+        add_local_agent(LocalAgentEntry(
+            agent_id="ratatoskr:wizard",
+            agent_name="wizard",
+            model="m",
+            description="d",
+            defined_at="t",
+        ))
+        monkeypatch.setenv("WORLDTREE_API_URL", "https://w.example")
+        monkeypatch.setenv("WORLDTREE_API_KEY", "k")
+        respx.delete("https://w.example/agents/ratatoskr:wizard").mock(
+            return_value=httpx.Response(204)
+        )
+        rc = main(["delete", "ratatoskr:wizard"])
+        out = capsys.readouterr()
+        assert rc == 0
+        assert out.out.strip() == "deleted ratatoskr:wizard"
+        assert load_local_agents() == []
+
+    def test_cli_missing_auth(
+        self, capsys: pytest.CaptureFixture[str], monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """cli_missing_auth [error]: no api-key → stderr [auth_error] + exit 11."""
+        monkeypatch.delenv("WORLDTREE_API_KEY", raising=False)
+        rc = main([
+            "define",
+            "--name", "wizard",
+            "--system-prompt", "x",
+            "--model", "m",
+        ])
+        err = capsys.readouterr().err
+        assert rc == 11
+        assert "[auth_error]" in err
+
+    @respx.mock
+    def test_cli_api_failed(
+        self, capsys: pytest.CaptureFixture[str], monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """cli_api_failed [error]: 500 → stderr [api_failed] + exit 20."""
+        monkeypatch.setenv("WORLDTREE_API_URL", "https://w.example")
+        monkeypatch.setenv("WORLDTREE_API_KEY", "k")
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(500, content=b"upstream out")
+        )
+        rc = main([
+            "define",
+            "--name", "wizard",
+            "--system-prompt", "x",
+            "--model", "m",
+        ])
+        err = capsys.readouterr().err
+        assert rc == 20
+        assert "[api_failed]" in err
+
+    @respx.mock
+    def test_cli_quota_exceeded(
+        self, capsys: pytest.CaptureFixture[str], monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """cli_quota_exceeded [error]: 429 → stderr [quota_exceeded] + exit 20."""
+        monkeypatch.setenv("WORLDTREE_API_URL", "https://w.example")
+        monkeypatch.setenv("WORLDTREE_API_KEY", "k")
+        respx.post("https://w.example/agents/define").mock(
+            return_value=httpx.Response(
+                429,
+                headers={"Retry-After": "0"},
+                json={"detail": {"error_code": "agent_quota_exceeded"}},
+            )
+        )
+        rc = main([
+            "define",
+            "--name", "wizard",
+            "--system-prompt", "x",
+            "--model", "m",
+        ])
+        err = capsys.readouterr().err
+        assert rc == 20
+        assert "[quota_exceeded]" in err
+
+    def test_cli_patch_no_fields(
+        self, capsys: pytest.CaptureFixture[str], monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """cli_patch_no_fields [error]: patch with no flags → [usage_error] + exit 10."""
+        monkeypatch.setenv("WORLDTREE_API_URL", "https://w.example")
+        monkeypatch.setenv("WORLDTREE_API_KEY", "k")
+        rc = main(["patch", "ratatoskr:wizard"])
+        err = capsys.readouterr().err
+        assert rc == 10
+        assert "[usage_error]" in err

tests/test_tui.py

@@ -1,5 +1,6 @@
 """Tests for ratatoskr.tui per docs/contracts/issues/4.contract.md."""
 
+from pathlib import Path
 from unittest.mock import MagicMock
 
 import httpx
@@ -60,20 +61,43 @@ def _args_existing(session_id: str = "s-1existing", **overrides) -> ParsedArgs:
 
 
 def _spy_writes(monkeypatch) -> list:
-    """Patch RichLog.write to record every arg into a list (returned).
+    """Patch RichLog.write AND VerticalScroll.mount to record every renderable
+    or mounted-widget content into a single list (returned).
 
-    Accepts *args/**kwargs so Textual's internal deferred-render path
-    (which calls write positionally with width/expand/shrink/scroll_end)
-    still works after a write-during-mount + Resize sequence.
+    v0.9.0: transcript content is mounted into a VerticalScroll, not written
+    to a RichLog. The spy captures both shapes — for each mounted Static, the
+    Static's `renderable` (Markdown / RichText / str) lands in the list,
+    indistinguishably from RichLog.write entries. Integration tests assert
+    on substrings or types in `writes` so the merged shape is the right
+    abstraction.
+
+    Accepts *args/**kwargs so Textual's internal deferred-render paths still
+    work after a write-during-mount + Resize sequence.
     """
+    from textual.containers import VerticalScroll
+    from textual.widgets import Static
+
     writes: list = []
-    original = RichLog.write
 
-    def spy(self, content, *args, **kw):
+    original_write = RichLog.write
+
+    def spy_write(self, content, *args, **kw):
         writes.append(content)
-        return original(self, content, *args, **kw)
+        return original_write(self, content, *args, **kw)
 
-    monkeypatch.setattr(RichLog, "write", spy)
+    monkeypatch.setattr(RichLog, "write", spy_write)
+
+    original_mount = VerticalScroll.mount
+
+    def spy_mount(self, *children, **kw):
+        for child in children:
+            if isinstance(child, Static):
+                writes.append(child.content)
+            else:
+                writes.append(child)
+        return original_mount(self, *children, **kw)
+
+    monkeypatch.setattr(VerticalScroll, "mount", spy_mount)
     return writes
 
 
@@ -115,97 +139,116 @@ SID = SseId(42, 5)
 class TestTuiPresenterState:
     """Tests for the new TuiPresenterState — per issue #12 contract."""
 
-    def test_thinking_streams_into_thinking_log(self) -> None:
-        """thinking_streams_into_thinking_log [happy,tracer, v0.6.5]:
-        3 Thinking deltas → thinking_log gets Rule(start) + 3 delta lines.
-        Transcript untouched; no thinking-current Static involved.
+    def test_thinking_coalesces_until_newline(self) -> None:
+        """thinking_coalesces_until_newline [happy,tracer, v0.7.1]:
+        Per-token deltas accumulate in the buffer; flush only on `\\n`.
+        Three short token-shaped deltas without `\\n` → thinking_log gets
+        ONLY Rule(start); content stays buffered.
         """
         from rich.rule import Rule
 
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         thinking_log = MagicMock()
         state = TuiPresenterState()
-        for chunk in ("a", "b", "c"):
+        for chunk in ("Let", " me", " think"):
             state.render(
                 Thinking(sse_id=SID, content=chunk),
-                log=log,
+                transcript=transcript,
                 tools_log=MagicMock(),
                 debug_log=MagicMock(),
-                current_text=MagicMock(),
                 thinking_log=thinking_log,
                 raw=False,
             )
         writes = [c[0][0] for c in thinking_log.write.call_args_list]
-        # 1 Rule(start) + 3 content lines = 4 writes
-        assert len(writes) == 4
+        # Only Rule(start) — content stays buffered (no `\n` seen).
+        assert len(writes) == 1
         assert isinstance(writes[0], Rule)
-        assert writes[1] == "a"
-        assert writes[2] == "b"
-        assert writes[3] == "c"
-        # Transcript untouched during thinking streaming.
-        assert log.write.call_count == 0
+        assert state.thinking_chunk_buffer == "Let me think"
+        assert transcript.mount.call_count == 0
+
+    def test_thinking_flushes_on_newline(self) -> None:
+        """thinking_flushes_on_newline [happy, v0.7.1]:
+        Delta carrying `\\n` flushes the accumulated buffer as ONE line.
+        """
+        from ratatoskr.tui import TuiPresenterState
+
+        thinking_log = MagicMock()
+        state = TuiPresenterState()
+        for chunk in ("Hello", " world", "\n"):
+            state.render(
+                Thinking(sse_id=SID, content=chunk),
+                transcript=MagicMock(),
+                tools_log=MagicMock(),
+                debug_log=MagicMock(),
+                thinking_log=thinking_log,
+                raw=False,
+            )
+        writes = [c[0][0] for c in thinking_log.write.call_args_list]
+        # Rule(start) + "Hello world" (one coalesced line) = 2 writes
+        assert len(writes) == 2
+        assert writes[1] == "Hello world"
+        assert state.thinking_chunk_buffer == ""
 
     def test_thinking_closes_to_thinking_log(self) -> None:
-        """thinking_closes_to_thinking_log [happy, v0.6.5]: 2x Thinking + WorkerPhase →
-        thinking_log gets Rule(start) + 2 delta lines + Rule(end); debug_log gets
-        the worker_phase line; transcript untouched.
+        """thinking_closes_to_thinking_log [happy, v0.7.1]: 2x Thinking + WorkerPhase →
+        v0.7.1 coalesces "a"+"b" into one buffered string; the close flushes
+        "ab" as a single line before Rule(end). Result: Rule(start) + "ab" +
+        Rule(end) = 3 writes. debug_log gets worker_phase; transcript untouched.
         """
         from rich.rule import Rule
 
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         debug_log = MagicMock()
         thinking_log = MagicMock()
         state = TuiPresenterState()
         for content in ("a", "b"):
             state.render(
                 Thinking(sse_id=SID, content=content),
-                log=log,
+                transcript=transcript,
                 tools_log=MagicMock(),
                 debug_log=debug_log,
-                current_text=MagicMock(),
                 thinking_log=thinking_log,
                 raw=False,
             )
         state.render(
             WorkerPhase(sse_id=SID, phase="streaming", turn_id=42),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
             debug_log=debug_log,
-            current_text=MagicMock(),
             thinking_log=thinking_log,
             raw=False,
         )
         thinking_writes = [c[0][0] for c in thinking_log.write.call_args_list]
-        # 1 Rule(start) + 2 delta lines + 1 Rule(end) = 4 writes
-        assert len(thinking_writes) == 4
+        # v0.7.1: 1 Rule(start) + 1 coalesced "ab" tail-flush + 1 Rule(end) = 3 writes
+        assert len(thinking_writes) == 3
         assert isinstance(thinking_writes[0], Rule)
-        assert thinking_writes[1] == "a"
-        assert thinking_writes[2] == "b"
-        assert isinstance(thinking_writes[3], Rule)
+        assert thinking_writes[1] == "ab"
+        assert isinstance(thinking_writes[2], Rule)
         # worker_phase still goes to debug_log; transcript untouched.
         assert "· worker_phase:" in _text_of(debug_log.write.call_args_list[-1][0][0])
-        assert not log.write.called
+        assert not transcript.mount.called
 
     # v0.6.5: thinking-current Static removed; test_thinking_widget_truncation
     # and test_thinking_widget_visibility_lifecycle deleted (no longer apply).
 
     def test_multiple_thinking_runs_each_get_thinking_log_section(self) -> None:
-        """multiple_thinking_runs_each_get_section [scenario, v0.6.5]:
+        """multiple_thinking_runs_each_get_section [scenario, v0.8.1]:
         Thinking → Text → Thinking → Done → TWO start/end Rule pairs in
-        thinking_log, each wrapping their delta lines. Text goes to
-        current_text (buffered). Transcript: [done] + Markdown body.
+        thinking_log (deltas coalesced into tail-flushes per run).
+        Text deltas now stream into the transcript via coalesce-on-newline
+        (no current-text Static); "hi" with no `\\n` stays buffered until
+        Done's tail-flush.
         """
         from rich.rule import Rule
 
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         thinking_log = MagicMock()
-        current_text = MagicMock()
         state = TuiPresenterState()
         for evt in (
             Thinking(sse_id=SID, content="first"),
@@ -213,28 +256,27 @@ class TestTuiPresenterState:
             Thinking(sse_id=SID, content="second"),
         ):
             state.render(
-                evt, log=log,
+                evt, transcript=transcript,
                 tools_log=MagicMock(), debug_log=MagicMock(),
-                current_text=current_text, thinking_log=thinking_log, raw=False,
+                thinking_log=thinking_log, raw=False,
             )
         state.render(
             _make_tui_done(),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(), debug_log=MagicMock(),
-            current_text=current_text, thinking_log=thinking_log, raw=False,
+            thinking_log=thinking_log, raw=False,
         )
-        # v0.6.5: thinking_log holds 4 Rules (start + end per run) + 2 delta lines.
+        # thinking_log: 4 Rules (start+end per run) + 2 tail-flush strings.
         thinking_writes = [c[0][0] for c in thinking_log.write.call_args_list]
         rules = [w for w in thinking_writes if isinstance(w, Rule)]
         delta_strs = [w for w in thinking_writes if isinstance(w, str)]
         assert len(rules) == 4, f"expected 4 Rules (2 start + 2 end), got {len(rules)}"
         assert "first" in delta_strs
         assert "second" in delta_strs
-        # Text "hi" went to current_text (buffered), not the transcript directly.
-        current_text.update.assert_any_call("hi")
-        # Transcript: [done] label + Markdown(response) (raw=False).
-        log_writes = [_text_of(c[0][0]) for c in log.write.call_args_list]
-        assert any(w.startswith("[done]") for w in log_writes if isinstance(w, str))
+        # v0.8.1: Text "hi" flushes as a line in transcript on Done.
+        transcript_renderables = [_text_of(r) for r in _mounted_renderables(transcript)]
+        assert "hi" in transcript_renderables
+        assert any(w.startswith("[done]") for w in transcript_renderables if isinstance(w, str))
 
     def test_render_exception_fallback(self) -> None:
         """render_exception_fallback [adversarial, v0.6.5]:
@@ -244,7 +286,7 @@ class TestTuiPresenterState:
         """
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         thinking_log = MagicMock()
         # First call (Rule write) raises; subsequent calls succeed for fallback.
         thinking_log.write.side_effect = [
@@ -255,10 +297,9 @@ class TestTuiPresenterState:
         state = TuiPresenterState()
         state.render(
             Thinking(sse_id=SID, content="x"),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
             debug_log=MagicMock(),
-            current_text=MagicMock(),
             thinking_log=thinking_log,
             raw=False,
         )
@@ -266,7 +307,7 @@ class TestTuiPresenterState:
         assert any(w.startswith("[thinking]") for w in writes), writes
         assert any(w == "[render_error] AttributeError" for w in writes), writes
         assert not any("rule write failed" in w for w in writes), writes
-        assert not log.write.called
+        assert not transcript.mount.called
 
     def test_state_reset_per_worker(self) -> None:
         """state_reset_per_worker [trace]: fresh TuiPresenterState() starts no thinking open."""
@@ -275,10 +316,9 @@ class TestTuiPresenterState:
         s1 = TuiPresenterState()
         s1.render(
             Thinking(sse_id=SID, content="x"),
-            log=MagicMock(),
+            transcript=MagicMock(),
             tools_log=MagicMock(),
             debug_log=MagicMock(),
-            current_text=MagicMock(),
             thinking_log=MagicMock(),
             raw=False,
         )
@@ -293,98 +333,101 @@ class TestTuiPresenterState:
         """
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         thinking_log = MagicMock()
         state = TuiPresenterState()
         state.render(
             Thinking(sse_id=SID, content="partial"),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=thinking_log, raw=False,
+            debug_log=MagicMock(), thinking_log=thinking_log, raw=False,
         )
         state.render(
             Cancelled(
                 sse_id=SID, phase="cancelled", turn_id=42, reason="user", partial_message_id=None
             ),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=thinking_log, raw=False,
+            debug_log=MagicMock(), thinking_log=thinking_log, raw=False,
         )
         # v0.6.5: streamed thinking + Rule(end) in thinking_log; [cancelled] in transcript.
-        log_writes = [_text_of(c[0][0]) for c in log.write.call_args_list]
-        assert any(w.startswith("[cancelled]") for w in log_writes)
+        transcript_renderables = [_text_of(r) for r in _mounted_renderables(transcript)]
+        assert any(w.startswith("[cancelled]") for w in transcript_renderables)
         # thinking_log got at least Rule(start) + "partial" delta + Rule(end)
         assert thinking_log.write.call_count >= 3
 
-    def test_done_renders_markdown_after_label(self) -> None:
-        """done_renders_markdown_after_label [happy, v0.6.0]:
-        Text("hi") accumulates into current_text Static (buffered streaming);
-        Done(response="hi") with raw=False → [done] label + Rule + Markdown
-        in transcript. current_text cleared on terminal.
+    def test_text_then_done_mounts_widget_and_finalizes(self) -> None:
+        """text_then_done_mounts_widget_and_finalizes [happy, v0.9.0]:
+        First Text delta mounts a Static(Markdown(buffer)) into the transcript;
+        Done finalizes the widget reference and mounts a styled [done] label.
+        No duplicate content (v0.9.0 replaces v0.8.x's flush-on-Done with
+        live in-place Markdown updates).
         """
         from rich.markdown import Markdown
-        from rich.rule import Rule
 
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
-        current_text = MagicMock()
+        transcript = MagicMock()
         state = TuiPresenterState()
         state.render(
             Text(sse_id=SID, content="hi"),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
             debug_log=MagicMock(),
-            current_text=current_text,
             thinking_log=MagicMock(),
             raw=False,
         )
-        # Text accumulated to current_text, NOT written to log.
-        current_text.update.assert_any_call("hi")
+        # v0.9.0: response widget mounted on first Text delta with Markdown wrapper.
+        assert transcript.mount.called
+        first_widget = transcript.mount.call_args_list[0][0][0]
+        assert isinstance(first_widget.content, Markdown)
+        assert first_widget.content.markup == "hi"
+        assert state.text_chunk_buffer == "hi"
+        # Done finalizes: text_chunk_buffer cleared, widget ref released, label mounted.
         state.render(
             _make_tui_done(),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
             debug_log=MagicMock(),
-            current_text=current_text,
             thinking_log=MagicMock(),
             raw=False,
         )
-        # Done cleared current_text and wrote [done] label + Rule + Markdown.
-        current_text.update.assert_any_call("")
-        writes = [c[0][0] for c in log.write.call_args_list]
+        writes = _mounted_renderables(transcript)
         assert any(_text_of(w).startswith("[done]") for w in writes)
-        assert any(isinstance(w, Rule) for w in writes)
-        assert any(isinstance(w, Markdown) for w in writes)
+        # v0.9.0: response Markdown rendered live during stream — only ONE
+        # Markdown renderable lands in the transcript (no post-Done re-render).
+        markdowns = [w for w in writes if isinstance(w, Markdown)]
+        assert len(markdowns) == 1
+        assert state.text_chunk_buffer == ""
+        assert state.current_response_widget is None
 
     def test_raw_flag_skips_markdown(self) -> None:
-        """raw_flag_skips_markdown [trace]: raw=True → no Rule, no Markdown."""
+        """raw_flag_skips_markdown [v0.9.0]: raw=True → response widget holds
+        plain str instead of Markdown. Live in-place update still happens;
+        only the wrapper differs.
+        """
         from rich.markdown import Markdown
-        from rich.rule import Rule
 
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         state = TuiPresenterState()
         state.render(
             Text(sse_id=SID, content="hi"),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=True,
+            debug_log=MagicMock(), thinking_log=MagicMock(), raw=True,
         )
         state.render(
             _make_tui_done(),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=True,
+            debug_log=MagicMock(), thinking_log=MagicMock(), raw=True,
         )
-        writes = [c[0][0] for c in log.write.call_args_list]
-        assert not any(isinstance(w, Rule) for w in writes)
+        writes = _mounted_renderables(transcript)
+        # Raw mode bypasses Markdown entirely — content lives as plain str.
         assert not any(isinstance(w, Markdown) for w in writes)
+        assert "hi" in writes
 
     def test_worker_phase_demoted_to_debug_log(self) -> None:
         """worker_phase_demoted_to_debug_log [trace, v0.5.0]: WorkerPhase → debug_log
@@ -395,18 +438,17 @@ class TestTuiPresenterState:
 
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         debug_log = MagicMock()
         state = TuiPresenterState()
         state.render(
             WorkerPhase(sse_id=SID, phase="streaming", turn_id=42),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=debug_log,
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=False,
+            debug_log=debug_log, thinking_log=MagicMock(), raw=False,
         )
         # v0.5.0: WorkerPhase routes to debug_log, NOT transcript.
-        assert not log.write.called
+        assert not transcript.mount.called
         renderable = debug_log.write.call_args[0][0]
         # INV-003: must be a styled Rich Text renderable, not a plain str.
         # v0.4.1 retheme: style is now Australis Sea dark-60 ("#86929d") instead
@@ -431,101 +473,108 @@ class TestTuiPresenterState:
         """
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         tools_log = MagicMock()
         state = TuiPresenterState()
         state.render(
             ToolStart(sse_id=SID, name="read_file", arguments={"path": "/x"}),
-            log=log,
+            transcript=transcript,
             tools_log=tools_log,
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=False,
+            debug_log=MagicMock(), thinking_log=MagicMock(), raw=False,
         )
         # INV-014: write went to tools_log
         assert tools_log.write.called
         assert _text_of(tools_log.write.call_args[0][0]).startswith("· tool_start:")
         # INV-014: transcript was NOT written to
-        assert not log.write.called
+        assert not transcript.mount.called
 
     def test_tool_result_routes_to_tools_log(self) -> None:
         """tool_result_routes_to_tools_log [INV-014]: ToolResult → tools_log, NOT transcript."""
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         tools_log = MagicMock()
         state = TuiPresenterState()
         state.render(
             ToolResult(sse_id=SID, name="read_file", result="ok", duration_ms=12),
-            log=log,
+            transcript=transcript,
             tools_log=tools_log,
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=False,
+            debug_log=MagicMock(), thinking_log=MagicMock(), raw=False,
         )
         assert tools_log.write.called
         assert _text_of(tools_log.write.call_args[0][0]).startswith("· tool_result:")
-        assert not log.write.called
+        assert not transcript.mount.called
 
-    def test_text_event_buffers_into_current_text(self) -> None:
-        """text_event_buffers_into_current_text [v0.6.0]: Text → current_text Static
-        (accumulated), NOT log or tools_log. Streaming UX fix — no per-token spam.
+    def test_text_first_delta_mounts_response_widget(self) -> None:
+        """text_first_delta_mounts_response_widget [v0.9.0]: first Text delta
+        mounts a Static carrying Markdown(buffer) into the transcript. The
+        text_chunk_buffer holds the accumulated content for the next delta's
+        in-place update.
         """
+        from rich.markdown import Markdown
+
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         tools_log = MagicMock()
-        current_text = MagicMock()
         state = TuiPresenterState()
         state.render(
             Text(sse_id=SID, content="hello"),
-            log=log,
+            transcript=transcript,
             tools_log=tools_log,
             debug_log=MagicMock(),
-            current_text=current_text,
             thinking_log=MagicMock(),
             raw=False,
         )
-        current_text.update.assert_called_once_with("hello")
-        assert not log.write.called
+        assert state.text_chunk_buffer == "hello"
+        assert transcript.mount.call_count == 1
+        widget = transcript.mount.call_args[0][0]
+        assert isinstance(widget.content, Markdown)
+        assert widget.content.markup == "hello"
+        assert state.current_response_widget is widget
         assert not tools_log.write.called
 
-    def test_text_deltas_accumulate(self) -> None:
-        """text_deltas_accumulate [v0.6.0]: multiple Text deltas → current_text shows
-        concatenated content, NOT separate per-delta lines.
+    def test_text_subsequent_deltas_update_in_place(self) -> None:
+        """text_subsequent_deltas_update_in_place [v0.9.0]: deltas after the
+        first do NOT mount a new widget — they update the existing widget's
+        Markdown content in place. The text_chunk_buffer accumulates.
         """
         from ratatoskr.tui import TuiPresenterState
 
-        current_text = MagicMock()
+        transcript = MagicMock()
         state = TuiPresenterState()
         for tok in ("Hel", "lo", " ", "world"):
             state.render(
                 Text(sse_id=SID, content=tok),
-                log=MagicMock(),
+                transcript=transcript,
                 tools_log=MagicMock(),
                 debug_log=MagicMock(),
-                current_text=current_text,
                 thinking_log=MagicMock(),
                 raw=False,
             )
-        # Final update reflects the full concatenation.
-        assert current_text.update.call_args_list[-1][0][0] == "Hello world"
+        # Exactly ONE mount (the first delta); subsequent deltas update.
+        assert transcript.mount.call_count == 1
+        assert state.text_chunk_buffer == "Hello world"
+        # Widget reference held; buffer is the source of truth re-rendered
+        # into Markdown(...) for each Static.update call.
+        assert state.current_response_widget is not None
 
     def test_duration_format_seconds(self) -> None:
         """duration_format_seconds [trace]: Done(duration_ms=5467) → label has "duration=5.5s"."""
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         state = TuiPresenterState()
         state.render(
             _make_tui_done(duration_ms=5467),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=True,
+            debug_log=MagicMock(), thinking_log=MagicMock(), raw=True,
         )
         done_line = next(
-            _text_of(c[0][0])
-            for c in log.write.call_args_list
-            if _text_of(c[0][0]).startswith("[done]")
+            _text_of(r)
+            for r in _mounted_renderables(transcript)
+            if _text_of(r).startswith("[done]")
         )
         assert "duration=5.5s" in done_line
         assert "duration_ms=5467" not in done_line
@@ -534,7 +583,7 @@ class TestTuiPresenterState:
         """usage_format_unicode_arrow [trace]: TUI Done label uses → (Unicode), not -> (ASCII)."""
         from ratatoskr.tui import TuiPresenterState
 
-        log = MagicMock()
+        transcript = MagicMock()
         state = TuiPresenterState()
         usage = {
             "prompt_tokens": 6756,
@@ -544,15 +593,14 @@ class TestTuiPresenterState:
         }
         state.render(
             _make_tui_done(usage=usage),
-            log=log,
+            transcript=transcript,
             tools_log=MagicMock(),
-            debug_log=MagicMock(),
-            current_text=MagicMock(), thinking_log=MagicMock(), raw=True,
+            debug_log=MagicMock(), thinking_log=MagicMock(), raw=True,
         )
         done_line = next(
-            _text_of(c[0][0])
-            for c in log.write.call_args_list
-            if _text_of(c[0][0]).startswith("[done]")
+            _text_of(r)
+            for r in _mounted_renderables(transcript)
+            if _text_of(r).startswith("[done]")
         )
         assert "usage 6756 in → 126 out (6882 total, 0 cached)" in done_line
 
@@ -563,14 +611,38 @@ def _text_of(write_arg: object) -> str:
     Issue #12 wraps demoted-telemetry entries in `rich.text.Text(..., style="dim")`
     so the RichLog can apply dim styling; non-demoted writes stay as plain str.
     Tests that want to assert against content need both shapes flattened.
+
+    v0.9.0: also extracts plain text from Markdown wrappers (the streaming-text
+    response path uses Markdown(buffer) now; tests assert against the source
+    markup, which lives in `Markdown.markup`).
     """
+    from rich.markdown import Markdown
     from rich.text import Text as RichText
 
     if isinstance(write_arg, RichText):
         return write_arg.plain
+    if isinstance(write_arg, Markdown):
+        return write_arg.markup
     if isinstance(write_arg, str):
         return write_arg
-    return ""  # Markdown / Rule / etc. — not text content
+    return ""  # Rule / etc. — not text content
+
+
+def _mounted_renderables(transcript_mock: MagicMock) -> list:
+    """v0.9.0: TuiPresenterState now mounts Static widgets into the transcript
+    VerticalScroll instead of writing renderables to a RichLog. Tests using a
+    MagicMock transcript inspect `transcript.mount.call_args_list`; each call's
+    first positional arg is the Static child whose `.content` carries the
+    Markdown / RichText / str that pre-v0.9.0 would have been the write arg.
+    Returns those renderables in mount-call order so tests can assert on them
+    with the same shape they used for `log.write.call_args_list` previously.
+    """
+    out: list = []
+    for call in transcript_mock.mount.call_args_list:
+        for child in call.args:
+            renderable = getattr(child, "content", child)
+            out.append(renderable)
+    return out
 
 
 def _make_tui_done(*, duration_ms: int = 1, usage: dict[str, int] | None = None) -> Done:
@@ -594,15 +666,15 @@ def _make_tui_done(*, duration_ms: int = 1, usage: dict[str, int] | None = None)
 class TestCancelViaSse:
     @respx.mock
     async def test_happy_cancel(self) -> None:
-        """happy_cancel [happy,tracer]: 200 OK → returns None; log has no [cancel_failed]."""
+        """happy_cancel [happy,tracer]: 200 OK → returns None; transcript has no [cancel_failed]."""
         respx.post("https://w.example/sessions/s-1/turns/42/cancel").mock(
             return_value=httpx.Response(200, json=_CANCEL_OK_RESP)
         )
-        log = MagicMock()
+        transcript = MagicMock()
         async with httpx.AsyncClient(base_url="https://w.example") as client:
-            result = await _cancel_via_sse(client, "s-1", 42, log=log)
+            result = await _cancel_via_sse(client, "s-1", 42, transcript=transcript)
         assert result is None
-        log.write.assert_not_called()
+        transcript.mount.assert_not_called()
 
     @respx.mock
     async def test_cancel_failed_500(self) -> None:
@@ -610,10 +682,10 @@ class TestCancelViaSse:
         respx.post("https://w.example/sessions/s-1/turns/42/cancel").mock(
             return_value=httpx.Response(500, content=b"boom")
         )
-        log = MagicMock()
+        transcript = MagicMock()
         async with httpx.AsyncClient(base_url="https://w.example") as client:
-            await _cancel_via_sse(client, "s-1", 42, log=log)
-        line = log.write.call_args[0][0]
+            await _cancel_via_sse(client, "s-1", 42, transcript=transcript)
+        line = transcript.mount.call_args[0][0].content
         assert "[cancel_failed]" in line
         assert "CancelFailed" in line
 
@@ -623,10 +695,10 @@ class TestCancelViaSse:
         respx.post("https://w.example/sessions/s-1/turns/42/cancel").mock(
             return_value=httpx.Response(409)
         )
-        log = MagicMock()
+        transcript = MagicMock()
         async with httpx.AsyncClient(base_url="https://w.example") as client:
-            await _cancel_via_sse(client, "s-1", 42, log=log)
-        line = log.write.call_args[0][0]
+            await _cancel_via_sse(client, "s-1", 42, transcript=transcript)
+        line = transcript.mount.call_args[0][0].content
         assert "[cancel_failed]" in line
         assert "CancelAlreadyCompleted" in line
 
@@ -636,10 +708,10 @@ class TestCancelViaSse:
         respx.post("https://w.example/sessions/s-1/turns/42/cancel").mock(
             side_effect=httpx.ConnectError("network down")
         )
-        log = MagicMock()
+        transcript = MagicMock()
         async with httpx.AsyncClient(base_url="https://w.example") as client:
-            await _cancel_via_sse(client, "s-1", 42, log=log)
-        line = log.write.call_args[0][0]
+            await _cancel_via_sse(client, "s-1", 42, transcript=transcript)
+        line = transcript.mount.call_args[0][0].content
         assert "[cancel_failed]" in line
         assert "ConnectError" in line
 
@@ -706,18 +778,18 @@ class TestLayoutShape:
             assert row is not None
 
     async def test_left_column_content_only(self) -> None:
-        """left_column_content_only [v0.6.5]: left column = transcript + prompt
-        + current-text (streaming text Static). thinking-current Static
-        removed entirely as of v0.6.5.
+        """left_column_content_only [v0.9.0]: left column = transcript-scroll
+        VerticalScroll + prompt Input. thinking-current Static removed in
+        v0.6.5; transcript RichLog replaced by VerticalScroll in v0.9.0.
         """
-        from textual.containers import Vertical
-        from textual.widgets import Input, RichLog
+        from textual.containers import Vertical, VerticalScroll
+        from textual.widgets import Input
 
         app = _resolved_app(_args_new(), session_id="s-new12345", agent_id="mimir")
         async with app.run_test() as pilot:
             await pilot.pause()
             left = app.query_one("#left-column", Vertical)
-            transcript = app.query_one("#transcript", RichLog)
+            transcript = app.query_one("#transcript-scroll", VerticalScroll)
             prompt = app.query_one("#prompt", Input)
             assert transcript in left.walk_children()
             assert prompt in left.walk_children()
@@ -742,7 +814,7 @@ class TestLayoutShape:
             assert tools_tab is not None
 
     async def test_tools_log_inside_tools_tab(self) -> None:
-        """tools_log_inside_tools_tab: tools-log RichLog is a descendant of tools-tab TabPane."""
+        """tools_log_inside_tools_tab: tools-transcript RichLog is a descendant of tools-tab TabPane."""
         from textual.widgets import RichLog, TabPane
 
         app = _resolved_app(_args_new(), session_id="s-new12345", agent_id="mimir")
@@ -793,7 +865,7 @@ class TestLayoutShape:
             )
 
     async def test_debug_tab_exists(self) -> None:
-        """debug_tab_exists [v0.5.0]: right column has Debug TabPane + #debug-log RichLog."""
+        """debug_tab_exists [v0.5.0]: right column has Debug TabPane + #debug-transcript RichLog."""
         from textual.widgets import RichLog, TabPane
 
         app = _resolved_app(_args_new(), session_id="s-new12345", agent_id="mimir")
@@ -815,35 +887,45 @@ class TestLayoutShape:
             assert app.query_one("#side-panes", TabbedContent).active == "debug-tab"
 
     async def test_done_label_styled_success(self) -> None:
-        """done_label_styled_success [v0.5.1]: [done] label renders in Aurora green."""
+        """done_label_styled_success [v0.9.0]: [done] label mounts as Static
+        carrying a RichText with Aurora green style. Inspect the mounted
+        Static's `.content`.
+        """
         from rich.text import Text as RichText
+        from textual.containers import VerticalScroll
         from textual.widgets import RichLog
 
         app = _resolved_app(_args_new(), session_id="s-new12345", agent_id="mimir")
         async with app.run_test() as pilot:
             await pilot.pause()
-            # Probe the presenter directly — write a Done via state.render.
             from ratatoskr.tui import TuiPresenterState
-            log = app.query_one("#transcript", RichLog)
+            transcript = app.query_one("#transcript-scroll", VerticalScroll)
             state = TuiPresenterState()
-            seen: list = []
-            orig = log.write
-            log.write = lambda c, *a, **kw: (seen.append(c), orig(c, *a, **kw))[1]
+            mounted: list = []
+            orig_mount = transcript.mount
+
+            def spy_mount(*ch, **kw):
+                mounted.extend(ch)
+                return orig_mount(*ch, **kw)
+
+            transcript.mount = spy_mount  # type: ignore[method-assign]
             state.render(
                 _make_tui_done(),
-                log=log,
+                transcript=transcript,
                 tools_log=app.query_one("#tools-log", RichLog),
                 debug_log=app.query_one("#debug-log", RichLog),
-                current_text=MagicMock(), thinking_log=MagicMock(), raw=True,
+                thinking_log=MagicMock(),
+                raw=True,
             )
             done = next(
-                c for c in seen
-                if isinstance(c, RichText) and _text_of(c).startswith("[done]")
+                w.content for w in mounted
+                if isinstance(getattr(w, "content", None), RichText)
+                and _text_of(w.content).startswith("[done]")
             )
             assert done.style == "#16B866"  # Aurora green
 
     async def test_empty_state_placeholders_present(self) -> None:
-        """empty_state_placeholders_present [v0.5.1]: tools-log + debug-log show
+        """empty_state_placeholders_present [v0.5.1]: tools-transcript + debug-transcript show
         placeholder lines before any turn fires."""
         from textual.widgets import RichLog
 
@@ -1051,10 +1133,14 @@ async def _submit_and_wait(app: RatatoskrApp, pilot, content: str) -> None:
 
 class TestStreamTurnWorker:
     @respx.mock
-    async def test_happy_text_done_renders_markdown(self, monkeypatch: pytest.MonkeyPatch) -> None:
-        """happy_text_done_renders_markdown [happy,tracer, v0.6.0]:
-        Text deltas go to current_text (not transcript); on Done, transcript
-        gets turn-header Rule, [done] label, post-Done Rule + Markdown body.
+    async def test_happy_text_done_no_double_print(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        """happy_text_done_no_double_print [happy,tracer, v0.9.0]:
+        Text("hello") mounts a Static(Markdown("hello")) into the transcript;
+        Done mounts a [done] label Static. The Markdown is rendered live (one
+        widget for the whole stream, updated in place), so there is NO
+        post-Done re-render — exactly ONE Markdown renderable lands in the
+        transcript for the response body. v0.9.0 supersedes v0.8.2's
+        drop-Markdown patch with proper live rendering.
         """
         stream = _sse_chunk("42:1", {"type": "text", "content": "hello"}) + _sse_chunk(
             "42:2", _DONE_BODY
@@ -1070,24 +1156,27 @@ class TestStreamTurnWorker:
             await pilot.pause()
             await _submit_and_wait(app, pilot, "hi")
             assert app.state == "idle"
-            # v0.6.0: Text("hello") goes to current_text Static, NOT log.
-            # writes spy captures RichLog.write only, so "hello" SHOULD NOT appear.
             from rich.markdown import Markdown
-            from rich.rule import Rule
 
-            assert not any(w == "hello" for w in writes)
-            assert any("[done]" in str(w) for w in writes)
-            # Post-Done: Markdown body + Rule + turn-header Rule all present.
-            assert any(isinstance(w, Markdown) for w in writes)
-            assert any(isinstance(w, Rule) for w in writes)
+            # The response body lives as ONE Markdown renderable mounted into
+            # the transcript; live updates happen via Static.update, not via
+            # re-mount, so there's exactly one Markdown in the spy stream.
+            markdowns = [w for w in writes if isinstance(w, Markdown)]
+            assert len(markdowns) == 1, (
+                f"v0.9.0: expected exactly ONE Markdown mounted, got {len(markdowns)}"
+            )
+            assert markdowns[0].markup == "hello"
+            # [done] label fires too.
+            assert any("[done]" in _text_of(w) for w in writes)
 
     @respx.mock
     async def test_raw_flag_skips_markdown_render(self, monkeypatch: pytest.MonkeyPatch) -> None:
-        """raw_flag_skips_markdown_render [trace, v0.6.0]:
-        With --raw, no Markdown render. A turn-header Rule IS still written
-        (v0.6.0 INV — turn correlation lives in every pane). The post-Done
-        Rule(separator) is suppressed; accumulated streamed text is written
-        as a plain string instead.
+        """raw_flag_skips_markdown_render [trace, v0.9.0]:
+        With --raw, the response widget holds plain str instead of Markdown.
+        Turn-header markers still appear in every pane: 3 RichLog panes
+        receive a Rule, the transcript-scroll receives a Static-wrapped
+        RichText (mounted, not written), giving 3 Rules in the captured
+        writes list.
         """
         stream = _sse_chunk("42:1", {"type": "text", "content": "hi"}) + _sse_chunk(
             "42:2", _DONE_BODY
@@ -1105,11 +1194,11 @@ class TestStreamTurnWorker:
 
             # No Markdown in raw mode.
             assert not any(isinstance(w, Markdown) for w in writes)
-            # Only turn-header Rules — one per pane (transcript + tools +
-            # debug + thinking = 4). No post-Done separator Rule.
+            # 3 Rules — one per RichLog pane (tools / debug / thinking).
+            # Transcript-scroll uses a Static turn-header Markdown alternative.
             rules = [w for w in writes if isinstance(w, Rule)]
-            assert len(rules) == 4, f"expected 4 turn-header Rules, got {len(rules)}"
-            # Accumulated text "hi" written as plain string post-Done.
+            assert len(rules) == 3, f"expected 3 turn-header Rules, got {len(rules)}"
+            # Accumulated text "hi" mounted as plain str into transcript.
             assert "hi" in writes
 
     @respx.mock
@@ -1483,8 +1572,14 @@ class TestActionInterrupt:
                 await pilot.pause(0.02)
             # Give _cancel_via_sse time to write the [cancel_failed] line
             await pilot.pause(0.05)
-            log = app.query_one("#transcript", RichLog)
-            rendered = "\n".join(str(strip.text) for strip in log.lines)
+            from textual.containers import VerticalScroll
+            from textual.widgets import Static
+            transcript = app.query_one("#transcript-scroll", VerticalScroll)
+            rendered = "\n".join(
+                str(child.content)
+                for child in transcript.children
+                if isinstance(child, Static)
+            )
             assert "[cancel_failed]" in rendered
             assert app.state == "cancelling"
             stream_gate.set()  # let stream finish for teardown
@@ -2099,6 +2194,72 @@ class TestResolveThenRunWithPicker:
         assert agents_route.call_count == 0
         assert sessions_route.call_count == 1
 
+    @respx.mock
+    def test_picker_merges_local_tier3_agents(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+    ) -> None:
+        """picker_merges_local_tier3_agents [v0.8.0]: local index entries
+        appear in the picker's agent list alongside remote agents."""
+        from ratatoskr.local_agents import LocalAgentEntry, add_local_agent
+
+        # Isolate the local index in a tmp file.
+        monkeypatch.setenv(
+            "RATATOSKR_LOCAL_AGENTS", str(tmp_path / "local_agents.json")
+        )
+        add_local_agent(LocalAgentEntry(
+            agent_id="ratatoskr:wizard",
+            agent_name="wizard",
+            model="qwen3.6-35-a3b",
+            description="(tier 3) test wizard",
+            defined_at="2026-05-25T00:00:00+00:00",
+        ))
+        respx.get("https://w.example/agents").mock(
+            return_value=httpx.Response(200, json=_AGENTS_RESP)
+        )
+        respx.post("https://w.example/sessions").mock(
+            return_value=httpx.Response(201, json=_CREATE_OK_RESP)
+        )
+
+        from ratatoskr.tui import AgentPickerApp
+
+        captured: list = []
+
+        async def capture_picker_init(self, *a, **kw):
+            captured.append(list(self.agents))
+            return "mimir"  # auto-pick something so the rest succeeds
+
+        # Patch __init__ to capture the agent list passed to the picker.
+        orig_init = AgentPickerApp.__init__
+
+        def init_spy(self, agents):
+            captured.append(list(agents))
+            orig_init(self, agents)
+
+        monkeypatch.setattr(AgentPickerApp, "__init__", init_spy)
+
+        async def picker_returns_mimir(self):
+            return "mimir"
+
+        monkeypatch.setattr(AgentPickerApp, "run_async", picker_returns_mimir)
+
+        async def fake_main(self, *a, **kw):
+            return 0
+
+        monkeypatch.setattr(RatatoskrApp, "run_async", fake_main)
+        from ratatoskr.tui import run_tui
+
+        rc = run_tui(_args_new_no_agent())
+        assert rc == 0
+        # The local tier-3 agent should appear in the picker's agents list.
+        assert captured, "AgentPickerApp.__init__ was never called"
+        agent_ids = {a.agent_id for a in captured[0]}
+        assert "ratatoskr:wizard" in agent_ids
+        # Plus the remote agents.
+        assert "mimir" in agent_ids
+        assert "lofn" in agent_ids
+
     @respx.mock
     def test_picker_skipped_when_session_mode(self, monkeypatch: pytest.MonkeyPatch) -> None:
         """picker_skipped_when_session_mode: --session s-1 → no list_agents, no create_session."""
@@ -2165,8 +2326,16 @@ class TestResolveThenRunWithPicker:
         self,
         monkeypatch: pytest.MonkeyPatch,
         capsys: pytest.CaptureFixture[str],
+        tmp_path: Path,
     ) -> None:
-        """list_agents returns [] → stderr [no_agents]; exit 13; picker NOT opened."""
+        """list_agents returns [] AND no local tier-3 entries → stderr
+        [no_agents]; exit 13; picker NOT opened. Isolate
+        $RATATOSKR_LOCAL_AGENTS so the operator's real local index
+        doesn't merge in and turn this into a non-empty list."""
+        # v0.8.0 isolation: point local agents at an empty tmp file.
+        monkeypatch.setenv(
+            "RATATOSKR_LOCAL_AGENTS", str(tmp_path / "empty_local_agents.json")
+        )
         respx.get("https://w.example/agents").mock(return_value=httpx.Response(200, json=[]))
 
         from ratatoskr.tui import AgentPickerApp

uv.lock

@@ -968,7 +968,7 @@ wheels = [
 
 [[package]]
 name = "ratatoskr"
-version = "0.6.5"
+version = "0.9.0"
 source = { editable = "." }
 dependencies = [
     { name = "httpx" },