release: promote dev to master (v1.0.0-beta.48)

[jaylfc]

Promotes dev to master for v1.0.0-beta.48 (release PR #2364, 27 PRs since beta.47). Closes out the tsk-laezfg / #2362 release-time verification: the secret-ignores-gate runs on this PR's merge result and the master push, proving the promoted .gitignore keeps every secret-material rule. Post-merge checks: tree identity (diff dev..master empty) before tagging v1.0.0-beta.48.

Summary by CodeRabbit

• New Features

  • Added five Hailo-10H model options, increasing the catalog to 120 manifests.
  • Added editable agent handles for owners and administrators.
  • Added wallpaper fitting options per device: cover, contain, stretch, centered, and tiled.
  • Added task quarantine and lead-authorized unquarantine controls with strike tracking.
  • Improved agent chat responsiveness with queued messages, background work, progress, and status reporting.
  • Added external access for supported OpenAI-compatible model endpoints.

• Bug Fixes

  • Improved security around session binding, memory paths, and unauthorized resource visibility.
  • Fixed project reorder rollback and task state handling.

• Documentation

  • Updated release, model catalog, API, agent, and deployment documentation.

[qodo-code-review]

ⓘ Qodo reviews are paused because the subscription is no longer active. Ask your workspace admin to reactivate the subscription to resume reviews. Manage billing

[github-actions]

👋 Thanks for the PR! This one targets master, which is our
stable branch (it's what live installs track). Please retarget it to
dev — click Edit next to the PR title and change the base
branch dropdown from master to dev. Your commits and any review
carry over, nothing is lost.

See CONTRIBUTING.md for the branch model.

[coderabbitai]

📝 Walkthrough

Walkthrough

Beta 48 adds task quarantine controls, AgentLoop-based chat coordination, Hailo model manifests, wallpaper-fit preferences, registry handle editing, authorization changes, chat export support, CI validation gates, tests, documentation, and release metadata.

Changes

Validation gates and release controls

Layer / File(s)	Summary
Store and secret validation gates .github/workflows/*, scripts/check_*, tests/test_check_*	Added Git-aware checks for new store wiring and secret-ignore protections.
CI runtime and release metadata .github/workflows/ci.yml, CHANGELOG.md, pyproject.toml, tinyagentos/__init__.py, desktop/package.json	Updated Node.js to 22 and versioned the release as beta.48.

Hailo model catalog

Layer / File(s)	Summary
Hailo manifests and resolver support app-catalog/models/*hef, tinyagentos/cluster/capabilities.py, tinyagentos/routes/models.py, tests/catalog/*	Added five Hailo HEF manifests and Hailo hardware resolution coverage.
Manifest metadata and integrity validation tinyagentos/registry.py, tests/test_model_manifest_integrity.py, tests/test_registry.py	Loaded context windows and validated manifest targets, hashes, URLs, and sizes.

Desktop presentation controls

Layer / File(s)	Summary
Wallpaper-fit persistence and rendering desktop/src/stores/theme-store.ts, desktop/src/components/*, desktop/src/apps/SettingsApp.tsx, desktop/src/theme/tokens.css, desktop/src/stores/__tests__/*	Added per-device wallpaper-fit persistence, settings controls, CSS modes, and tests.
Registry handle editing and row behavior desktop/src/apps/agents/RegistryPanel.*	Added authorized handle editing and status-section behavior tests.
GuidesApp coverage desktop/src/apps/GuidesApp.test.tsx	Added tests for loading, recommendations, errors, and empty results.

AgentLoop coordination

Layer / File(s)	Summary
AgentLoop lifecycle and subagents tinyagentos/agent_loop.py, docs/design/agent-loop-subagents.md, tests/test_agent_loop.py	Added loop states, queued messages, safe points, supervised subagents, cancellation, status, and wait operations.
ACP and desktop chat integration tinyagentos/agent_chat_router.py, tinyagentos/routes/taos_agent.py, tests/test_agent_chat_router.py, tests/test_taos_agent_chat.py	Serialized per-agent turns, queued concurrent messages, drained queued frames, and exposed loop status.

Task quarantine and persistence

Layer / File(s)	Summary
Strike storage and task lifecycle tinyagentos/projects/strike_store.py, tinyagentos/projects/task_store.py, tinyagentos/app.py, tinyagentos/routes/projects.py, tests/projects/test_strike_wiring.py, tests/test_task_store.py	Added strike persistence, quarantine transitions, unquarantine routing, cleanup, auditing, and integration tests.
Project list transaction handling tinyagentos/projects/lists_store.py, tests/projects/test_lists_store.py	Fixed cursor access and rollback behavior for reorder failures.

Authentication and authorization boundaries

Layer / File(s)	Summary
Middleware and session validation tinyagentos/auth_middleware.py, tinyagentos/routes/auth.py, tinyagentos/routes/{canvas,channel_hub,chat,terminal}.py, tests/test_auth.py, tests/test_routes_agent_model_api.py	Added exact model-route exemptions and consistent User-Agent session validation.
Existence-hiding authorization tinyagentos/routes/agent_*.py, tests/test_agent_registry.py, tests/test_agent_scope_requests.py, related route tests	Unauthorized registry and scope-request targets now return indistinguishable 404 responses.
Path and invite validation tinyagentos/routes/memory.py, tests/test_routes_memory.py, tests/test_routes_project_invites.py	Rejected unsafe memory paths and added invite scope and rollback coverage.

Chat export and route coverage

Layer / File(s)	Summary
A2A chat export tinyagentos/chat/chat_exporter.py, tests/test_chat_exporter.py	Added deterministic multi-channel export with content handling, reply filtering, and size limits.
Route and context regression coverage tests/test_chat_context_window.py, tests/test_routes_mcp.py, tests/conftest.py	Added context-window, MCP endpoint, and fixture coverage.

Documentation and installation coverage

Layer / File(s)	Summary
Agent and Radio Studio designs docs/design/*, docs/agent-manual/*, docs/taos-agent-manual.md, docs/mirror-policy.md	Added AgentLoop and Radio Studio specifications and revised agent guidance.
Generic LXC installation checks tests/test_lxc_installer.py	Added generic installation fixtures and strict acceptance tests.

Estimated code review effort: 5 (Critical) | ~120 minutes

Possibly related PRs

• jaylfc/taOS#2333: Adds the same StrikeStore, task lifecycle wiring, unquarantine route, and tests.
• jaylfc/taOS#2335: Adds the same AgentLoop implementation, design documentation, and tests.
• jaylfc/taOS#2334: Adds the same store-wiring gate, checker, workflow, documentation, and tests.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 37.38% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the release promotion from dev to master and names version v1.0.0-beta.48.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch dev

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[gitar-bot]

Important

You are using the Gitar free plan. Upgrade to unlock code review, CI analysis, auto-apply, custom automations, and more.

_Gitar

[coderabbitai]

Actionable comments posted: 17

Note

Due to the large number of review comments, Critical, Major severity comments were prioritized as inline comments.

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

tests/test_agent_scope_requests.py (1)

440-462: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Update the stale expected-status comment.

The comment says authorization returns 403, but Line 462 correctly asserts 404. State that authorization runs before scope validation and returns 404.

Proposed fix

-        # Authz runs first -> 403, NOT a 400 vocab error confirming the bad scope.
+        # Authz runs first -> 404, NOT a 400 vocab error confirming the bad scope.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_agent_scope_requests.py` around lines 440 - 462, Update the
assertion comment in test_create_authorizes_before_scope_vocab to state that
authorization runs before scope validation and returns 404, matching the
existing resp.status_code assertion.

🟡 Minor comments (15)

tests/test_lxc_installer.py-791-797 (1)

791-797: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Assert a state-path creation operation.

The assertion passes when any command only references /var/lib/taosr1. A generic installer can then XPASS without creating the directory. Assert that a directory-creation operation targets the declared path.

Proposed test change

-        assert any(
-            "/var/lib/taosr1" in cmd for cmd in full_cmds
-        ), "state_paths must be created inside the container (/var/lib/taosr1)"
+        assert any(
+            "/var/lib/taosr1" in cmd and ("mkdir" in cmd or "install -d" in cmd)
+            for cmd in full_cmds
+        ), "state_paths must be created inside the container (/var/lib/taosr1)"

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_lxc_installer.py` around lines 791 - 797, Update
test_state_paths_created_in_container to assert that captured commands include a
directory-creation operation targeting /var/lib/taosr1, rather than merely
checking for the path text. Match the installer’s actual mkdir command structure
while preserving the requirement that creation occurs inside the container.

tests/test_routes_mcp.py-44-44 (1)

44-44: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Remove the unused app bindings.

Ruff reports RUF059 on Lines 44 and 92. Bind the unused tuple value to _.

Proposed fix

-        client, app = app_client
+        client, _ = app_client

Also applies to: 92-92

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_routes_mcp.py` at line 44, Update the tuple unpacking in the
affected test setup statements to bind the unused second value to `_` instead of
`app`, including both occurrences around the `app_client` assignments. Keep the
`client` binding unchanged.

Source: Linters/SAST tools

tests/test_chat_context_window.py-47-55 (1)

47-55: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Use distinct message content to verify suffix retention.

All messages have the same content on Line 48. The assertion on Line 55 passes for any retained subset and does not verify that build_context_window kept the newest contiguous suffix.

Proposed test correction

-    message = "x" * 400  # 100 tokens per message
-    msgs = [_msg("user", message) for _ in range(20)]  # 2000 tokens total
+    msgs = [_msg("user", f"{index:04d}" + "x" * 396) for index in range(20)]
     ctx = build_context_window(msgs, limit=20, max_tokens=budget)
 ...
-    assert [m["content"] for m in ctx] == [message] * len(ctx)
+    assert [m["content"] for m in ctx] == [
+        m["content"] for m in msgs[-len(ctx):]
+    ]

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_chat_context_window.py` around lines 47 - 55, Update the test
around build_context_window to assign distinct content to each message, such as
sequence-labeled values, while preserving the token-budget setup. Change the
final assertion to compare the retained messages against the expected newest
contiguous suffix, ensuring older messages are dropped first rather than merely
checking identical content.

tests/test_model_manifest_integrity.py-201-205 (1)

201-205: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Parse download_url before accepting it.

startswith("https://") accepts malformed values such as https://. It also accepts URLs without a hostname. Require a string URL with scheme == "https" and a non-empty netloc.

Proposed fix

+from urllib.parse import urlparse
+
 ...
-            if not url or not url.startswith("https://"):
+            parsed = urlparse(url) if isinstance(url, str) else None
+            if parsed is None or parsed.scheme != "https" or not parsed.netloc:
                 errors.append(
                     f"{mid}/{vid}: download_url must be a non-empty https URL (got {url!r})"
                 )

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_model_manifest_integrity.py` around lines 201 - 205, Update the
download_url validation in the model manifest integrity check to parse the value
as a URL before accepting it. Require a string value whose parsed scheme is
exactly "https" and whose netloc is non-empty, while preserving the existing
error reporting through errors.append for invalid URLs.

tests/test_auth.py-1256-1259 (1)

1256-1259: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Assert the expected successful response.

Line 1259 accepts a 404 or 500 response. The matching User-Agent case must return 200.

Proposed fix

-        assert ok.status_code != 401
+        assert ok.status_code == 200

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_auth.py` around lines 1256 - 1259, Update the assertion in the
auth client test around the /auth/me request to require status code 200 for the
matching old-browser User-Agent, rather than merely accepting any non-401
response.

scripts/check_store_wiring.py-218-221 (1)

218-221: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Exclude BaseStore from the subclass result.

_inherits_base_store("BaseStore", ...) returns True. Line 218 therefore reports a newly added BaseStore as an unwired subclass. The gate must only inspect subclasses.

Proposed fix

     return {
         name for name in classes_in_file
-        if _inherits_base_store(name, all_classes)
+        if name != "BaseStore" and _inherits_base_store(name, all_classes)
     }

Add a regression test for a file that defines only BaseStore.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@scripts/check_store_wiring.py` around lines 218 - 221, Exclude the
`BaseStore` symbol from the subclass set returned by the surrounding
store-wiring discovery function, while preserving detection of all other classes
that inherit from it. Add a regression test covering a file defining only
`BaseStore` and assert that no unwired subclass is reported.

.github/workflows/secret-ignores-gate.yml-30-30 (1)

30-30: 🔒 Security & Privacy | 🟡 Minor | ⚡ Quick win

Disable persisted checkout credentials in both gate workflows.

Both workflows run repository-controlled scripts after checkout. Set persist-credentials: false in both actions/checkout@v7 steps. The store-wiring workflow can still fetch the public origin without the token.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.github/workflows/secret-ignores-gate.yml at line 30, Disable persisted
checkout credentials by setting persist-credentials to false on the
actions/checkout@v7 step in .github/workflows/secret-ignores-gate.yml at lines
30-30 and .github/workflows/store-wiring-gate.yml at lines 32-34; apply the same
change to both workflow checkout steps.

Source: Linters/SAST tools

desktop/src/apps/agents/RegistryPanel.test.tsx-800-803 (1)

800-803: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Update RegistryStatus to accept unknown server statuses.

RegistryStatus is a closed union, so the "frozen" fixture does not type-check and the test cannot compile.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@desktop/src/apps/agents/RegistryPanel.test.tsx` around lines 800 - 803,
Update the RegistryStatus type used by RegistryEntry and RegistryStatus handling
to accept arbitrary server-provided status strings, while preserving existing
recognized-status behavior. Ensure the "frozen" fixture in the
unrecognised-status test type-checks and follows the existing Other-section
rendering path.

docs/agent-manual/01-rules.md-26-26 (1)

26-26: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Use afterward in both manual files.

The same American-English locale violation appears in the source and generated manual. Fix the source, then regenerate the generated output.

• docs/agent-manual/01-rules.md#L26-L26: replace afterwards with afterward.
• docs/taos-agent-manual.md#L41-L41: regenerate this file from the corrected source.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/agent-manual/01-rules.md` at line 26, Replace “afterwards” with
“afterward” in docs/agent-manual/01-rules.md:26, then regenerate
docs/taos-agent-manual.md so docs/taos-agent-manual.md:41 reflects the corrected
source; do not edit the generated file independently.

Source: Linters/SAST tools

docs/design/radio-studio.md-100-100 (1)

100-100: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Add language identifiers to all fenced blocks.

Static analysis flags these seven fences. Use bash for command examples and text for diagrams, routes, and file maps.

Also applies to: 106-106, 265-265, 281-281, 334-334, 423-423, 451-451

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/radio-studio.md` at line 100, Add language identifiers to all
seven flagged fenced code blocks in the document: use bash for command examples
and text for diagrams, routes, and file maps.

Source: Linters/SAST tools

docs/taos-agent-manual.md-45-45 (1)

45-45: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Fix the generated section separators.

The new prose is immediately followed by ---. Markdown parses the preceding prose as a setext heading, which triggers MD003. Add a blank line before the separator in scripts/build-agent-manual.py or emit an explicit thematic break, then regenerate this file.

Also applies to: 265-267

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/taos-agent-manual.md` at line 45, Update the generated-section
formatting in scripts/build-agent-manual.py so newly emitted prose is separated
from the following `---` by a blank line, or emit an explicit thematic break
instead. Regenerate docs/taos-agent-manual.md and ensure the affected separators
no longer parse the preceding prose as setext headings.

Source: Linters/SAST tools

tinyagentos/projects/strike_store.py-73-87 (1)

73-87: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Make strike ordering deterministic.

created_at can be equal for consecutive inserts. In that case, SQLite can return either row as the latest strike. This can make the response and tests/projects/test_strike_wiring.py flaky.

Add an insertion-order tie breaker to both queries.

Proposed fix

- "SELECT * FROM task_strikes WHERE task_id = ? ORDER BY created_at ASC",
+ "SELECT * FROM task_strikes WHERE task_id = ? ORDER BY created_at ASC, rowid ASC",

- "SELECT * FROM task_strikes WHERE task_id = ? ORDER BY created_at DESC LIMIT 1",
+ "SELECT * FROM task_strikes WHERE task_id = ? ORDER BY created_at DESC, rowid DESC LIMIT 1",

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/projects/strike_store.py` around lines 73 - 87, The list and
latest queries in the strike store must use deterministic ordering when
created_at values tie. Update both ORDER BY clauses in the methods that fetch
strikes and latest to add the insertion-order column as a secondary ascending or
descending tie breaker matching each query’s direction.

tests/test_chat_exporter.py-649-670 (1)

649-670: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Extend this test with characters that JSON escaping expands.

The body is only "x" repeated. JSON serialization does not expand it, so this test cannot detect the size-budget gap in tinyagentos/chat/chat_exporter.py. Add a case whose body consists of quotes, backslashes, or control characters. The serialized envelope then grows well beyond the raw byte count.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_chat_exporter.py` around lines 649 - 670, Extend
test_oversize_envelope_never_exceeds_limit_serialized with JSON-escaping-heavy
content, such as quotes, backslashes, or control characters, instead of relying
only on repeated "x" characters. Keep the existing serialized-envelope length
assertion and ensure the payload exercises expansion between raw content size
and json.dumps output.

tinyagentos/chat/chat_exporter.py-38-56 (1)

38-56: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Handle blocks that are neither dict nor str.

A block of any other type (for example a number or null in the stored JSON) contributes nothing. The body can then become empty while content_blocks is non-empty. Line 188 then raises ChatExportError and aborts the whole channel export. This contradicts the docstring claim that a non-text block never disappears.

🛠️ Proposed fallback placeholder

         elif isinstance(block, str):
             if block:
                 parts.append(block)
+        elif block is not None:
+            parts.append(f"[{type(block).__name__} block]")
     return "\n".join(parts)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/chat/chat_exporter.py` around lines 38 - 56, Update the
block-processing function around the existing dict/string branches to handle
every other block type by appending a non-empty fallback placeholder, rather
than silently skipping it. Preserve current formatting for dict and string
blocks, ensuring non-text entries such as numbers or null remain represented so
content_blocks cannot produce an empty body solely because unsupported types
were discarded.

tinyagentos/chat/chat_exporter.py-146-152 (1)

146-152: 🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win

Handle unknown content_blocks values without aborting export. flatten_body skips non-dict, non-string elements such as 123. A list like [123] produces an empty body, so line 188 raises ChatExportError and aborts the channel export. Add a fallback placeholder for unsupported block values.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/chat/chat_exporter.py` around lines 146 - 152, Update the export
flow around flatten_body to handle unsupported content_blocks values without
raising ChatExportError. Ensure flatten_body emits a fallback placeholder for
non-dict, non-string blocks such as numeric values, so channels with content
like [123] continue exporting instead of producing an empty body.

🧹 Nitpick comments (9)

tinyagentos/routes/taos_agent.py (1)

585-589: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Prefer an explicit GeneratorExit handler over sys.exc_info().

sys.exc_info()[1] inside finally works only while the exception propagates. An explicit handler states the intent and removes the sys import dependency for this check.

♻️ Proposed alternative

         except Exception as exc:
             logger.exception("taos-agent: generator error")
             yield json.dumps({"error": str(exc)}) + "\n"
             content_frame_yielded = True
+        except GeneratorExit:
+            client_gone = True
+            raise

Set client_gone = False next to leftovers, then log on if leftovers and client_gone.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/routes/taos_agent.py` around lines 585 - 589, Update the
surrounding generator flow to catch GeneratorExit explicitly and set a
client-disconnected flag, initialized alongside leftovers; replace the
sys.exc_info() check in the warning condition with that flag, and remove the
now-unused sys import.

tests/test_taos_agent_chat.py (1)

332-339: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Move the assertion out of the fake adapter.

prompt runs inside the supervised _drive task. _drive catches Exception, so a failing assert action is LoopAction.QUEUED becomes an error frame in the stream instead of a test failure. Record the action on a nonlocal and assert it after the response.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_taos_agent_chat.py` around lines 332 - 339, Update the fake
adapter’s prompt method to store the result of loop.handle_message in a nonlocal
test variable instead of asserting there; after the supervised response/stream
completes, assert that recorded action is LoopAction.QUEUED from the outer test
context so assertion failures propagate to the test.

tinyagentos/agent_loop.py (2)

182-190: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

_prune_subagents discards every completed handle once the cap is passed.

When len(self._subagents) exceeds _max_subagents, the method deletes all done entries, not just the oldest ones. A caller that later calls await_subagent on one of those ids gets KeyError, and get_subagent returns None. Consider removing only the oldest done entries until the map is back at the cap.

♻️ Proposed change

     def _prune_subagents(self) -> None:
         if len(self._subagents) <= self._max_subagents:
             return
-        to_remove = [
+        done_ids = [
             sid for sid, entry in self._subagents.items()
             if entry.task_obj is not None and entry.task_obj.done()
         ]
-        for sid in to_remove:
+        excess = len(self._subagents) - self._max_subagents
+        for sid in done_ids[:excess]:
             del self._subagents[sid]

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/agent_loop.py` around lines 182 - 190, Update _prune_subagents to
remove only the oldest completed entries, stopping once len(self._subagents) is
at or below _max_subagents. Preserve completed handles until pruning requires
their removal, and retain newer done entries so await_subagent and get_subagent
can still access them.

---

432-443: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Collapse the duplicated result branches.

Both branches return entry.handle.result. The state == "cancelled" check has no effect in either location.

♻️ Proposed simplification

         if entry.task_obj.done():
-            if entry.handle.state == "cancelled":
-                return entry.handle.result
             return entry.handle.result
-        done, pending = await asyncio.wait([entry.task_obj], timeout=timeout)
+        _done, pending = await asyncio.wait([entry.task_obj], timeout=timeout)
         if entry.task_obj in pending:
             raise asyncio.TimeoutError(
                 f"subagent {sub_id} did not finish within {timeout}s"
             )
-        if entry.handle.state == "cancelled":
-            return entry.handle.result
         return entry.handle.result

The _done rename also clears the Ruff RUF059 hint at Line 436. Line 481 has the same hint.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/agent_loop.py` around lines 432 - 443, In the result-waiting
logic around the relevant agent-loop method, remove the redundant
cancelled-state condition and return entry.handle.result through a single path
for both already-completed and newly-completed tasks. Rename the unused done
value from asyncio.wait to _done, and apply the same unused-result rename at the
corresponding wait site around line 481.

Source: Linters/SAST tools

docs/design/agent-loop-subagents.md (1)

64-66: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Move the language tag onto the fence.

The fence opens without a language and the first content line is the literal word text. That word renders inside the diagram. markdownlint reports MD040 for the same fence.

📝 Proposed fix

-```
-text
+```text
 IDLE

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/agent-loop-subagents.md` around lines 64 - 66, Update the diagram
code fence containing the IDLE state so the language tag is attached to the
opening fence as text, and remove the literal text line from the diagram
content. Keep IDLE as the first diagram line and ensure the fence is properly
closed.

Source: Linters/SAST tools

tests/test_agent_loop.py (1)

379-418: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider covering the retention caps.

_max_subagents and _max_delivered control pruning in tinyagentos/agent_loop.py. No test exercises either path. A small test that spawns past _max_subagents would pin the retention behavior, including which completed handles remain reachable through get_subagent.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_agent_loop.py` around lines 379 - 418, Add tests covering the
retention limits configured by AgentLoop._max_subagents and _max_delivered.
Exercise spawning beyond _max_subagents and verify pruning plus which completed
handles remain accessible through get_subagent; also cover delivered-message
pruning beyond _max_delivered, preserving existing status behavior.

tests/test_agent_chat_router.py (1)

820-841: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add a cancellation case next to the failure case.

The suite covers a raising drive_turn. It does not cover cancellation of the _run_acp_turn task, which is the path AgentChatRouter.close() takes at shutdown. Add a test that cancels the task mid-turn and then asserts the loop state, so the drain invariant in tinyagentos/agent_chat_router.py Lines 89-93 is pinned by a test.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_agent_chat_router.py` around lines 820 - 841, Add an async
cancellation test adjacent to
test_run_acp_turn_drive_failure_does_not_wedge_loop, using a blocking fake
drive_turn and cancelling the _run_acp_turn task while it is mid-turn. Await the
cancellation, then assert the agent loop state is IDLE, covering the shutdown
path used by AgentChatRouter.close() and the drain invariant in _run_acp_turn.

tests/projects/test_lists_store.py (1)

400-437: 🩺 Stability & Availability | 🔵 Trivial | ⚡ Quick win

The test simulates a raised CancelledError, not real task cancellation.

cancelling_execute raises asyncio.CancelledError() directly. The enclosing task is never cancelled, so await self._db.rollback() in the production handler completes normally.

Under real cancellation the behavior differs. If asyncio.Task.cancel() cancels the task, the await inside the except BaseException handler can raise CancelledError again before rollback() finishes. The rollback is then lost, which is the exact failure the guard prevents.

Consider covering real cancellation and shielding the rollback.

♻️ Proposed additional test for real task cancellation

`@pytest.mark.asyncio`
async def test_reorder_entries_rolls_back_on_task_cancel(entries_store, monkeypatch):
    a = await entries_store.add_entry(
        list_id="lst-1", project_id="prj-1", text="A", original_text="A",
        author_kind="agent", author_id="agent-1",
    )
    b = await entries_store.add_entry(
        list_id="lst-1", project_id="prj-1", text="B", original_text="B",
        author_kind="agent", author_id="agent-1",
    )

    real_execute = entries_store._db.execute
    started = asyncio.Event()
    update_calls = 0

    async def slow_execute(sql, params=()):
        nonlocal update_calls
        if sql.startswith("UPDATE project_list_entries SET position"):
            update_calls += 1
            if update_calls == 2:
                started.set()
                await asyncio.sleep(3600)
        return await real_execute(sql, params)

    monkeypatch.setattr(entries_store._db, "execute", slow_execute)

    task = asyncio.create_task(
        entries_store.reorder_entries(
            "prj-1", "lst-1",
            [{"id": a["id"], "position": 1}, {"id": b["id"], "position": 0}],
        )
    )
    await started.wait()
    task.cancel()
    with pytest.raises(asyncio.CancelledError):
        await task
    monkeypatch.undo()

    await entries_store.add_entry(
        list_id="lst-2", project_id="prj-1", text="Unrelated",
        original_text="Unrelated", author_kind="agent", author_id="agent-1",
    )
    assert (await entries_store.get_entry(a["id"]))["position"] == 0
    assert (await entries_store.get_entry(b["id"]))["position"] == 1

🛡️ Proposed production change in tinyagentos/projects/lists_store.py to survive real cancellation

         except BaseException:
             # Without this, the UPDATEs already issued stay pending on the
             # shared connection and the next unrelated commit() flushes a
             # half-applied reorder. BaseException, not Exception: task
             # cancellation (CancelledError) must also roll back, and commit()
             # itself is inside the guard for the same reason.
-            await self._db.rollback()
+            # shield(): under real task cancellation a bare await in the
+            # handler can be cancelled again before the rollback completes.
+            await asyncio.shield(self._db.rollback())
             raise

This requires import asyncio in tinyagentos/projects/lists_store.py.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/projects/test_lists_store.py` around lines 400 - 437, Update
test_reorder_entries_rolls_back_on_cancellation to exercise real task
cancellation: run reorder_entries in an asyncio task, pause the second position
update with an event, cancel the task, and assert CancelledError while verifying
the entries were rolled back. Also update the production rollback handler in
reorder_entries to shield the rollback await from cancellation, adding the
required asyncio import.

tests/test_chat_exporter.py (1)

235-239: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Target the updated row by its id.

The UPDATE selects the row with ORDER BY created_at DESC LIMIT 1. It works only because ts is captured before the inserts and is therefore older than every new row. A later refactor that moves the ts assignment can make this statement rewrite the same row three times, and the test would still pass while covering nothing. send_message returns the message, so use its id.

♻️ Proposed test refactor

-    for i in range(3):
-        await store.send_message(
+    for i in range(3):
+        msg = await store.send_message(
             channel_id="ch1",
             author_id="user1",
             author_type="user",
             content=f"msg{i}",
             content_blocks=[{"type": "paragraph", "text": f"msg{i}"}],
         )
         await store._db.execute(
-            "UPDATE chat_messages SET created_at = ? WHERE id = (SELECT id FROM chat_messages ORDER BY created_at DESC LIMIT 1)",
-            (ts,),
+            "UPDATE chat_messages SET created_at = ? WHERE id = ?",
+            (ts, msg["id"]),
         )
         await store._db.commit()

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/test_chat_exporter.py` around lines 235 - 239, Update the test setup
around the send_message calls to retain the message returned by send_message,
then use that message’s id directly in the UPDATE statement instead of selecting
by the latest created_at. Preserve the existing timestamp update and commit
behavior while ensuring the intended inserted row is targeted explicitly.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/design/radio-studio.md`:
- Around line 325-330: Update the “Agent Chat” section to specify that Radio
Studio routes chat through the existing per-agent AgentChatRouter and AgentLoop,
including safe-point queue draining; do not describe it as an independent
standard chat path, and preserve the listed slash commands and rich response
behavior.
- Around line 184-202: Constrain hackrf_spectrum_survey inputs in the
tool_schema and mirror those checks in the API and skill implementation: require
positive start_hz and stop_hz with start_hz < stop_hz, enforce supported gain
bounds, require a positive bin_width_hz, and cap frequency span and resulting
output size. Reject invalid or resource-heavy requests before dispatch using
shared limit definitions where available.
- Around line 285-288: Update the Phase 1 UI documentation around the Spectrum
View and Signal Library layout to remove “Demod actions” or clearly mark it
disabled and explicitly Phase 2, and apply the same treatment to the chat’s
/stare and /capture controls. Ensure targeted sweeps, audio capture, and
demodulation are not presented as available Phase 1 functionality.
- Around line 334-343: Expand the radio-studio API design around POST
/api/radio-studio/survey to define the survey job contract: response payload and
job identifier, states exposed by GET /api/radio-studio/survey/{id}, failure
representation, retry behavior, and idempotency rules. Ensure the contract
supports safe client polling and clarifies how duplicate submissions are
handled.
- Around line 74-76: Keep HardwareProfile.profile_id stable and independent of
SDR detection; remove the -sdr suffix behavior while retaining SDR data through
the separate HardwareProfile.sdr capability. Ensure HardwareProfile.save() and
exact profile_id cache/tier lookups continue using the unchanged base
identifier.
- Around line 349-357: Update the Radio Studio design to define deny-by-default
permissions for skills, specify device policy enforcement that blocks direct USB
TX endpoint access, and add negative tests covering transmit requests and TX
writes. Retain the existing UI/API rejection and receive-only CLI behavior while
documenting concrete container or host restrictions that enforce the policy.
- Around line 110-113: Update the SDR device deployment flow described around
create_container() to pass the detector’s serial value through usb_devices,
using Incus serial selection to identify the intended device. Retain
vendorid/productid as matching metadata, and document busnum/devnum only as
fallback selectors rather than the primary device-selection mechanism.

In `@scripts/check_store_wiring.py`:
- Around line 279-285: Update the new-class detection loop around
_class_def_in_added_lines so renamed files (status starting with “R”) use
added-definition-line detection instead of marking every class as new; retain
unconditional detection for added/copied files and existing modified-file
behavior. Add a regression test for a rename-only change that preserves the
existing store class and confirms no subclass is reported.

In `@tests/conftest.py`:
- Around line 359-362: Update the client fixture teardown to close task_strikes
after project_task_store.close() and before releasing dependent resources. Reuse
the existing task_strikes cleanup pattern, ensuring the connection is closed
only when task_strikes._db is not None.

In `@tinyagentos/agent_chat_router.py`:
- Around line 89-93: Make the cleanup in _run_acp_turn cancellation-safe: ensure
cancellation cannot interrupt reach_safe_point before the AgentLoop state
returns to IDLE, such as by running the drain in a detached task or using a
synchronous reset on AgentLoop. Preserve queued-message draining and update the
finally path accordingly, then add a test that cancels _run_acp_turn during
drive_turn and asserts the a1 loop reaches LoopState.IDLE.

In `@tinyagentos/chat/chat_exporter.py`:
- Around line 229-246: The truncation in _serialized_size/envelope construction
must account for JSON escape expansion: after applying the initial byte budget,
repeatedly shrink emitted_body and re-check _serialized_size(envelope) until it
is within _max_message_bytes, preserving valid UTF-8 and the truncation marker.
In tests/test_chat_exporter.py lines 649-670, add coverage using quotes,
backslashes, or control characters and assert the serialized envelope remains
within the limit.

In `@tinyagentos/projects/lists_store.py`:
- Around line 263-281: Add a per-connection async lock for the shared database
connection and use it to serialize every write unit in ProjectListEntriesStore,
including each execute and its corresponding commit or rollback. Update
reorder_entries() and all other write methods to acquire the same lock across
their full transaction scope, including BaseException cancellation handling, so
writers cannot interleave or affect one another’s pending statements.

In `@tinyagentos/projects/task_store.py`:
- Around line 478-482: Update the unquarantine flow around clear_strikes so a
cleanup failure is not followed by a successful response while stale strikes
remain. When self._strikes.clear_strikes(task_id) fails, preserve or restore the
task’s quarantined status, or persist a retryable cleanup operation, and
propagate the failure through the route’s existing error path.
- Around line 470-473: Update the unquarantine SQL transition in the relevant
task-store method to set claimed_by and claimed_at to NULL alongside status =
'open'. Add a regression test covering a task claimed by one worker,
quarantined, unquarantined, and then successfully claimed by another worker.

In `@tinyagentos/routes/agent_auth_requests.py`:
- Around line 1211-1218: Update the denial flow around the record lookup and
ownership check to apply the same active-state predicate used by
approve_scope_request before allowing denial; inactive or missing identities
must return the existing 404 “agent not found” response. Add a regression test
that suspends an identity with a pending scope request and verifies denial
returns the unknown-target 404.

In `@tinyagentos/routes/auth.py`:
- Around line 657-664: Update AuthManager.validate_session so UA-bound sessions
reject both missing and mismatched User-Agent values, while sessions without
user_agent_hash remain valid. In tinyagentos/routes/auth.py:657-664 and 691-693
retain forwarding; in tinyagentos/routes/canvas.py:82-88,
channel_hub.py:268-274, chat.py:114-120, and terminal.py:26-30 ensure absent
headers are rejected before accepting or starting work. Add absent-User-Agent
coverage for /auth/status and /auth/me in tests/test_auth.py:1224-1263.

In `@tinyagentos/routes/models.py`:
- Line 44: Update delete_model() to locate and remove nested .hef model files
under the documented <backend>/<family>/<id>/<file> layout, rather than only
matching top-level paths with models_dir.glob(). Preserve deletion of the
registry entry and existing handling for other model file suffixes.

---

Outside diff comments:
In `@tests/test_agent_scope_requests.py`:
- Around line 440-462: Update the assertion comment in
test_create_authorizes_before_scope_vocab to state that authorization runs
before scope validation and returns 404, matching the existing resp.status_code
assertion.

---

Minor comments:
In @.github/workflows/secret-ignores-gate.yml:
- Line 30: Disable persisted checkout credentials by setting persist-credentials
to false on the actions/checkout@v7 step in
.github/workflows/secret-ignores-gate.yml at lines 30-30 and
.github/workflows/store-wiring-gate.yml at lines 32-34; apply the same change to
both workflow checkout steps.

In `@desktop/src/apps/agents/RegistryPanel.test.tsx`:
- Around line 800-803: Update the RegistryStatus type used by RegistryEntry and
RegistryStatus handling to accept arbitrary server-provided status strings,
while preserving existing recognized-status behavior. Ensure the "frozen"
fixture in the unrecognised-status test type-checks and follows the existing
Other-section rendering path.

In `@docs/agent-manual/01-rules.md`:
- Line 26: Replace “afterwards” with “afterward” in
docs/agent-manual/01-rules.md:26, then regenerate docs/taos-agent-manual.md so
docs/taos-agent-manual.md:41 reflects the corrected source; do not edit the
generated file independently.

In `@docs/design/radio-studio.md`:
- Line 100: Add language identifiers to all seven flagged fenced code blocks in
the document: use bash for command examples and text for diagrams, routes, and
file maps.

In `@docs/taos-agent-manual.md`:
- Line 45: Update the generated-section formatting in
scripts/build-agent-manual.py so newly emitted prose is separated from the
following `---` by a blank line, or emit an explicit thematic break instead.
Regenerate docs/taos-agent-manual.md and ensure the affected separators no
longer parse the preceding prose as setext headings.

In `@scripts/check_store_wiring.py`:
- Around line 218-221: Exclude the `BaseStore` symbol from the subclass set
returned by the surrounding store-wiring discovery function, while preserving
detection of all other classes that inherit from it. Add a regression test
covering a file defining only `BaseStore` and assert that no unwired subclass is
reported.

In `@tests/test_auth.py`:
- Around line 1256-1259: Update the assertion in the auth client test around the
/auth/me request to require status code 200 for the matching old-browser
User-Agent, rather than merely accepting any non-401 response.

In `@tests/test_chat_context_window.py`:
- Around line 47-55: Update the test around build_context_window to assign
distinct content to each message, such as sequence-labeled values, while
preserving the token-budget setup. Change the final assertion to compare the
retained messages against the expected newest contiguous suffix, ensuring older
messages are dropped first rather than merely checking identical content.

In `@tests/test_chat_exporter.py`:
- Around line 649-670: Extend
test_oversize_envelope_never_exceeds_limit_serialized with JSON-escaping-heavy
content, such as quotes, backslashes, or control characters, instead of relying
only on repeated "x" characters. Keep the existing serialized-envelope length
assertion and ensure the payload exercises expansion between raw content size
and json.dumps output.

In `@tests/test_lxc_installer.py`:
- Around line 791-797: Update test_state_paths_created_in_container to assert
that captured commands include a directory-creation operation targeting
/var/lib/taosr1, rather than merely checking for the path text. Match the
installer’s actual mkdir command structure while preserving the requirement that
creation occurs inside the container.

In `@tests/test_model_manifest_integrity.py`:
- Around line 201-205: Update the download_url validation in the model manifest
integrity check to parse the value as a URL before accepting it. Require a
string value whose parsed scheme is exactly "https" and whose netloc is
non-empty, while preserving the existing error reporting through errors.append
for invalid URLs.

In `@tests/test_routes_mcp.py`:
- Line 44: Update the tuple unpacking in the affected test setup statements to
bind the unused second value to `_` instead of `app`, including both occurrences
around the `app_client` assignments. Keep the `client` binding unchanged.

In `@tinyagentos/chat/chat_exporter.py`:
- Around line 38-56: Update the block-processing function around the existing
dict/string branches to handle every other block type by appending a non-empty
fallback placeholder, rather than silently skipping it. Preserve current
formatting for dict and string blocks, ensuring non-text entries such as numbers
or null remain represented so content_blocks cannot produce an empty body solely
because unsupported types were discarded.
- Around line 146-152: Update the export flow around flatten_body to handle
unsupported content_blocks values without raising ChatExportError. Ensure
flatten_body emits a fallback placeholder for non-dict, non-string blocks such
as numeric values, so channels with content like [123] continue exporting
instead of producing an empty body.

In `@tinyagentos/projects/strike_store.py`:
- Around line 73-87: The list and latest queries in the strike store must use
deterministic ordering when created_at values tie. Update both ORDER BY clauses
in the methods that fetch strikes and latest to add the insertion-order column
as a secondary ascending or descending tie breaker matching each query’s
direction.

---

Nitpick comments:
In `@docs/design/agent-loop-subagents.md`:
- Around line 64-66: Update the diagram code fence containing the IDLE state so
the language tag is attached to the opening fence as text, and remove the
literal text line from the diagram content. Keep IDLE as the first diagram line
and ensure the fence is properly closed.

In `@tests/projects/test_lists_store.py`:
- Around line 400-437: Update test_reorder_entries_rolls_back_on_cancellation to
exercise real task cancellation: run reorder_entries in an asyncio task, pause
the second position update with an event, cancel the task, and assert
CancelledError while verifying the entries were rolled back. Also update the
production rollback handler in reorder_entries to shield the rollback await from
cancellation, adding the required asyncio import.

In `@tests/test_agent_chat_router.py`:
- Around line 820-841: Add an async cancellation test adjacent to
test_run_acp_turn_drive_failure_does_not_wedge_loop, using a blocking fake
drive_turn and cancelling the _run_acp_turn task while it is mid-turn. Await the
cancellation, then assert the agent loop state is IDLE, covering the shutdown
path used by AgentChatRouter.close() and the drain invariant in _run_acp_turn.

In `@tests/test_agent_loop.py`:
- Around line 379-418: Add tests covering the retention limits configured by
AgentLoop._max_subagents and _max_delivered. Exercise spawning beyond
_max_subagents and verify pruning plus which completed handles remain accessible
through get_subagent; also cover delivered-message pruning beyond
_max_delivered, preserving existing status behavior.

In `@tests/test_chat_exporter.py`:
- Around line 235-239: Update the test setup around the send_message calls to
retain the message returned by send_message, then use that message’s id directly
in the UPDATE statement instead of selecting by the latest created_at. Preserve
the existing timestamp update and commit behavior while ensuring the intended
inserted row is targeted explicitly.

In `@tests/test_taos_agent_chat.py`:
- Around line 332-339: Update the fake adapter’s prompt method to store the
result of loop.handle_message in a nonlocal test variable instead of asserting
there; after the supervised response/stream completes, assert that recorded
action is LoopAction.QUEUED from the outer test context so assertion failures
propagate to the test.

In `@tinyagentos/agent_loop.py`:
- Around line 182-190: Update _prune_subagents to remove only the oldest
completed entries, stopping once len(self._subagents) is at or below
_max_subagents. Preserve completed handles until pruning requires their removal,
and retain newer done entries so await_subagent and get_subagent can still
access them.
- Around line 432-443: In the result-waiting logic around the relevant
agent-loop method, remove the redundant cancelled-state condition and return
entry.handle.result through a single path for both already-completed and
newly-completed tasks. Rename the unused done value from asyncio.wait to _done,
and apply the same unused-result rename at the corresponding wait site around
line 481.

In `@tinyagentos/routes/taos_agent.py`:
- Around line 585-589: Update the surrounding generator flow to catch
GeneratorExit explicitly and set a client-disconnected flag, initialized
alongside leftovers; replace the sys.exc_info() check in the warning condition
with that flag, and remove the now-unused sys import.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 9dd9c55b-b1e2-464f-9bcb-201b9222ea49

📥 Commits

Reviewing files that changed from the base of the PR and between f78ff3a and aaf1113.

⛔ Files ignored due to path filters (2)

• desktop/package-lock.json is excluded by !**/package-lock.json, !**/package-lock.json
• uv.lock is excluded by !**/*.lock, !**/uv.lock

📒 Files selected for processing (82)

• .claude/skills/taos-development-skill/SKILL.md
• .github/workflows/ci.yml
• .github/workflows/secret-ignores-gate.yml
• .github/workflows/store-wiring-gate.yml
• .gitignore
• CHANGELOG.md
• README.md
• app-catalog/models/deepseek-r1-distill-qwen-1.5b-hef/manifest.yaml
• app-catalog/models/llama-3.2-3b-instruct-hef/manifest.yaml
• app-catalog/models/qwen2-1.5b-instruct-hef/manifest.yaml
• app-catalog/models/qwen2.5-1.5b-instruct-hef/manifest.yaml
• app-catalog/models/qwen2.5-coder-1.5b-instruct-hef/manifest.yaml
• desktop/package.json
• desktop/src/App.tsx
• desktop/src/apps/GuidesApp.test.tsx
• desktop/src/apps/SettingsApp.tsx
• desktop/src/apps/agents/RegistryPanel.test.tsx
• desktop/src/apps/agents/RegistryPanel.tsx
• desktop/src/components/Desktop.tsx
• desktop/src/stores/__tests__/wallpaper-fit.test.ts
• desktop/src/stores/theme-store.ts
• desktop/src/theme/tokens.css
• docs/RELEASING.md
• docs/agent-coordination.md
• docs/agent-manual/01-rules.md
• docs/agent-manual/10-image-prompting.md
• docs/agent-manual/index.md
• docs/design/agent-loop-subagents.md
• docs/design/radio-studio.md
• docs/mirror-policy.md
• docs/taos-agent-manual.md
• pyproject.toml
• scripts/check_secret_ignores.py
• scripts/check_store_wiring.py
• tests/catalog/test_resolver_hailo.py
• tests/conftest.py
• tests/projects/test_lists_store.py
• tests/projects/test_strike_wiring.py
• tests/test_agent_chat_router.py
• tests/test_agent_loop.py
• tests/test_agent_registry.py
• tests/test_agent_scope_requests.py
• tests/test_auth.py
• tests/test_chat_context_window.py
• tests/test_chat_exporter.py
• tests/test_check_secret_ignores.py
• tests/test_check_store_wiring.py
• tests/test_lxc_installer.py
• tests/test_model_manifest_integrity.py
• tests/test_registry.py
• tests/test_registry_governance_lifecycle.py
• tests/test_routes_agent_model_api.py
• tests/test_routes_agent_org.py
• tests/test_routes_mcp.py
• tests/test_routes_memory.py
• tests/test_routes_project_invites.py
• tests/test_taos_agent_chat.py
• tests/test_task_store.py
• tests/test_token_rotation.py
• tinyagentos/__init__.py
• tinyagentos/agent_chat_router.py
• tinyagentos/agent_loop.py
• tinyagentos/app.py
• tinyagentos/auth_middleware.py
• tinyagentos/chat/chat_exporter.py
• tinyagentos/cluster/capabilities.py
• tinyagentos/projects/ids.py
• tinyagentos/projects/lists_store.py
• tinyagentos/projects/strike_store.py
• tinyagentos/projects/task_store.py
• tinyagentos/registry.py
• tinyagentos/routes/agent_auth_requests.py
• tinyagentos/routes/agent_registry.py
• tinyagentos/routes/auth.py
• tinyagentos/routes/canvas.py
• tinyagentos/routes/channel_hub.py
• tinyagentos/routes/chat.py
• tinyagentos/routes/memory.py
• tinyagentos/routes/models.py
• tinyagentos/routes/projects.py
• tinyagentos/routes/taos_agent.py
• tinyagentos/routes/terminal.py

[coderabbitai]

🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- candidate files ---'
git ls-files | rg '(^|/)(radio-studio\.md|.*hardware.*|.*profile.*)$' | head -80
printf '%s\n' '--- design document ---'
cat -n docs/design/radio-studio.md | sed -n '1,260p'
printf '%s\n' '--- HardwareProfile/profile_id references ---'
rg -n --glob '!docs/design/radio-studio.md' '\bHardwareProfile\b|\bprofile_id\b|SdrInfo|sdr' . | head -300

Repository: jaylfc/taOS

Length of output: 44832

---

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- implemented HardwareProfile and detection/cache flow ---'
cat -n tinyagentos/hardware.py | sed -n '1,145p'
cat -n tinyagentos/hardware.py | sed -n '580,675p'
printf '%s\n' '--- design profile contract and resolver consumers ---'
cat -n docs/design/plan-hardware-appregistry.md | sed -n '165,225p'
cat -n docs/design/plan-hardware-appregistry.md | sed -n '360,405p'
cat -n docs/design/plan-hardware-appregistry.md | sed -n '600,630p'
printf '%s\n' '--- direct profile/tier consumers in implementation ---'
rg -n -C 4 'hardware_tiers|get_hardware_profile|profile\.profile_id|hw\.profile_id|hardware_profile\.profile_id|detect_hardware\(' \
  tinyagentos app-catalog tests docs/design | head -500

Repository: jaylfc/taOS

Length of output: 46596

---

🏁 Script executed:

#!/bin/bash
set -eu
printf '%s\n' '--- exact compatibility and profile consumers ---'
cat -n tinyagentos/routes/store.py | sed -n '70,100p'
cat -n tinyagentos/routes/models.py | sed -n '720,765p'
cat -n tinyagentos/routes/dashboard.py | sed -n '118,138p'
cat -n tinyagentos/routes/memory_management.py | sed -n '42,60p'
printf '%s\n' '--- read-only source invariant probe ---'
python3 - <<'PY'
import ast
from pathlib import Path

hardware = Path("tinyagentos/hardware.py").read_text()
tree = ast.parse(hardware)

classes = {
    node.name: node for node in tree.body
    if isinstance(node, ast.ClassDef)
}
hp = classes["HardwareProfile"]
methods = {
    node.name: node for node in hp.body
    if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))
}
top_level = {
    node.name: node for node in tree.body
    if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))
}

def calls(func, name):
    return any(
        isinstance(n, ast.Call)
        and ((isinstance(n.func, ast.Name) and n.func.id == name)
             or (isinstance(n.func, ast.Attribute) and n.func.attr == name))
        for n in ast.walk(func)
    )

def has_profile_id_write(func):
    for n in ast.walk(func):
        if not isinstance(n, ast.Assign):
            continue
        for target in n.targets:
            if (isinstance(target, ast.Subscript)
                and isinstance(target.value, ast.Name)
                and target.value.id == "data"
                and isinstance(target.slice, ast.Constant)
                and target.slice.value == "profile_id"):
                return True
    return False

print("HardwareProfile.save writes data['profile_id']:", has_profile_id_write(methods["save"]))
print("HardwareProfile.load removes persisted profile_id:",
      "data.pop(\"profile_id\", None)" in hardware)
print("get_hardware_profile calls detect_hardware:", calls(top_level["get_hardware_profile"], "detect_hardware"))
print("get_hardware_profile calls profile.save:", calls(top_level["get_hardware_profile"], "save"))

radio = Path("docs/design/radio-studio.md").read_text().splitlines()
tier_lines = radio[211:215]
tier_keys = [line.split(":", 1)[0].strip() for line in tier_lines if ":" in line]
print("Radio Studio hardware tier keys:", tier_keys)
print("Radio Studio tier keys contain '-sdr':", any("-sdr" in key for key in tier_keys))
print("Radio Studio specifies re-detection on USB add/remove:",
      any("re-runs `detect_hardware()` on udev `add/remove`" in line for line in radio))
PY

Repository: jaylfc/taOS

Length of output: 5579

---

Keep SDR detection out of HardwareProfile.profile_id.

HardwareProfile.save() writes profile_id to the hardware cache, while USB re-detection can change the detected SDR state. Compatibility lookups use exact profile_id keys, and the listed hardware_tiers do not include -sdr keys. Keep profile_id stable and expose SDR through a separate capability. If the suffix remains, update all tier manifests, resolvers, cache consumers, and migrations together.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/radio-studio.md` around lines 74 - 76, Keep
HardwareProfile.profile_id stable and independent of SDR detection; remove the
-sdr suffix behavior while retaining SDR data through the separate
HardwareProfile.sdr capability. Ensure HardwareProfile.save() and exact
profile_id cache/tier lookups continue using the unchanged base identifier.

[coderabbitai]

🔒 Security & Privacy | 🟠 Major | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- target document ---'
sed -n '80,135p' docs/design/radio-studio.md

printf '%s\n' '--- relevant files ---'
git ls-files | rg '(^|/)(radio|sdr|incus|container|device|orchestr|agent)' | head -200

printf '%s\n' '--- symbols and USB fields ---'
rg -n -i --glob '!node_modules' --glob '!dist' --glob '!build' \
  'usb_devices|vendorid|productid|serial|rf-scanning|rf_scanning|create_container|usb' . | head -300

Repository: jaylfc/taOS

Length of output: 50368

---

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- target document ---'
sed -n '80,135p' docs/design/radio-studio.md

printf '%s\n' '--- relevant files ---'
git ls-files | rg -i '(^|/)(radio|sdr|incus|container|device|orchestr|agent)' | head -200

printf '%s\n' '--- USB-related references ---'
rg -n -i --glob '!node_modules' --glob '!dist' --glob '!build' \
  'usb_devices|vendorid|productid|serial|rf-scanning|rf_scanning|create_container|usb' . | head -300

Repository: jaylfc/taOS

Length of output: 50368

---

🌐 Web query:

Incus USB device type serial vendorid productid busnum devnum device selectors documentation

💡 Result:

In Incus, USB devices can be passed to instances (both containers and virtual machines) using the usb device type [1]. You can identify and select specific USB devices using several configuration keys, which act as selectors [2]. The following configuration keys are available to define or select a USB device [2]: vendorid: The vendor ID of the USB device (e.g., 0c88) [3][2]. productid: The product ID of the USB device (e.g., 163a) [3][2]. serial: The serial number of the USB device [4][2]. busnum: The bus number where the USB device is attached [4][2]. devnum: The device number of the USB device [4][2]. To add a USB device to an instance, use the incus config device add command [5][6]. For example: incus config device add <instance_name> <device_name> usb vendorid=<vendor_id> productid=<product_id> [3][7] Additional context: - For containers, the usb device type is primarily for libusb-compatible devices [1]. Devices requiring dedicated kernel drivers may be better handled via unix-char or unix-hotplug device types [1]. - For virtual machines, the entire USB device is passed through to the instance [1]. - You can also set a device as required (default false) to control whether the instance fails to start if the device is missing [1][2]. - The keys busnum, devnum, and serial were added to allow more granular or stable selection of USB devices [4][2].

Citations:

• 1: https://linuxcontainers.org/incus/docs/main/reference/devices_usb/
• 2: https://github.com/lxc/incus/blob/47b31414/internal/server/device/usb.go
• 3: https://discuss.linuxcontainers.org/t/using-incus-for-automatic-building-of-windows-application/18981
• 4: Add serial number option for type usb, and make busnum and devnum available as config keys for device type usb.  lxc/incus#629
• 5: https://linuxcontainers.org/incus/docs/main/reference/manpages/incus/config/device/add/
• 6: https://linuxcontainers.org/incus/docs/main/_sources/howto/instances_configure.md.txt
• 7: https://blog.simos.info/how-to-use-your-android-phone-with-adb-in-an-incus-container/

---

🏁 Script executed:

#!/bin/bash
set -eu

# Read-only checks for the exact claim: whether this design drops a device
# identity field and whether the repository documents Incus USB selectors.
rg -n -C 5 -i \
  'serial|vendorid|productid|busnum|devnum|usb_devices|create_container|rf-scanning' \
  docs/design/radio-studio.md README.md docs 2>/dev/null | head -400

printf '%s\n' '--- all tracked radio-studio references ---'
rg -n -i 'radio-studio|radio studio|SDR|software-defined radio|rf-scanning' \
  --glob '!docs/design/radio-studio.md' . | head -250

Repository: jaylfc/taOS

Length of output: 50368

---

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- design excerpt ---'
nl -ba docs/design/radio-studio.md | sed -n '100,120p'

printf '%s\n' '--- candidate tracked source files ---'
git ls-files | rg -i '(^|/)(sdr|radio|usb|incus|container|orchestr|agent)([^/]*)$|(^|/)[^/]*(sdr|radio|usb|incus|container|orchestr|agent)[^/]*\.(go|rs|py|ts|tsx|js|jsx|md|yaml|yml)$' | head -200

printf '%s\n' '--- exact tracked references outside generated/vendor content ---'
git grep -n -I -E 'usb_devices|vendorid|productid|serial|rf-scanning|rf_scanning|create_container' -- \
  ':!**/node_modules/**' ':!**/dist/**' ':!**/build/**' ':!**/*.min.js' \
  ':!**/*.map' ':!**/static/**' ':!**/public/**' | head -300

Repository: jaylfc/taOS

Length of output: 50367

---

Pass the SDR serial to create_container(). vendorid and productid can match multiple devices. A replacement or second matching device can attach to the agent. Use Incus serial selection and document busnum/devnum only as a fallback.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/radio-studio.md` around lines 110 - 113, Update the SDR device
deployment flow described around create_container() to pass the detector’s
serial value through usb_devices, using Incus serial selection to identify the
intended device. Retain vendorid/productid as matching metadata, and document
busnum/devnum only as fallback selectors rather than the primary
device-selection mechanism.

Source: MCP tools

[coderabbitai]

🩺 Stability & Availability | 🟠 Major | 🏗️ Heavy lift

Bound survey inputs before dispatch.

The schema accepts arbitrary integers. It does not constrain positive frequencies, start_hz < stop_hz, gain limits, bin width, maximum span, or output size. Add schema constraints and enforce the same limits in the API and skill implementation to prevent invalid or resource-heavy sweeps.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/radio-studio.md` around lines 184 - 202, Constrain
hackrf_spectrum_survey inputs in the tool_schema and mirror those checks in the
API and skill implementation: require positive start_hz and stop_hz with
start_hz < stop_hz, enforce supported gain bounds, require a positive
bin_width_hz, and cap frequency span and resulting output size. Reject invalid
or resource-heavy requests before dispatch using shared limit definitions where
available.

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Remove Phase 2 actions from the Phase 1 UI.

The layout advertises “Demod actions,” and the chat exposes /stare and /capture. The Phase 1 cut excludes targeted sweeps, audio capture, and demodulation. Remove these controls or mark them disabled and explicitly Phase 2.

Also applies to: 325-328

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/radio-studio.md` around lines 285 - 288, Update the Phase 1 UI
documentation around the Spectrum View and Signal Library layout to remove
“Demod actions” or clearly mark it disabled and explicitly Phase 2, and apply
the same treatment to the chat’s /stare and /capture controls. Ensure targeted
sweeps, audio capture, and demodulation are not presented as available Phase 1
functionality.

[coderabbitai]

🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

Route Radio Studio chat through the existing AgentLoop.

“Standard chat pane” does not identify the serialization owner. Specify that Radio Studio uses the existing per-agent AgentChatRouter and AgentLoop, including safe-point queue draining. A parallel chat path could race with ACP or desktop turns.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/radio-studio.md` around lines 325 - 330, Update the “Agent Chat”
section to specify that Radio Studio routes chat through the existing per-agent
AgentChatRouter and AgentLoop, including safe-point queue draining; do not
describe it as an independent standard chat path, and preserve the listed slash
commands and rich response behavior.

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Clear the prior claim when unquarantining.

If quarantine_task receives a claimed task, it preserves claimed_by. This update restores status = 'open' but also preserves that claim. claim_task requires claimed_by IS NULL, so the task cannot re-enter the claimable pool.

Clear claimed_by and claimed_at in this transition. Add a regression test for claimed → quarantined → unquarantined → claimed by another worker.

Proposed fix

- SET status = 'open', updated_at = ?
+ SET status = 'open', claimed_by = NULL, claimed_at = NULL, updated_at = ?

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	"""UPDATE project_tasks
	SET status = 'open', updated_at = ?
	WHERE id = ? AND status = 'quarantined'""",
	(now, task_id),
	"""UPDATE project_tasks
	SET status = 'open', claimed_by = NULL, claimed_at = NULL, updated_at = ?
	WHERE id = ? AND status = 'quarantined'""",
	(now, task_id),

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/projects/task_store.py` around lines 470 - 473, Update the
unquarantine SQL transition in the relevant task-store method to set claimed_by
and claimed_at to NULL alongside status = 'open'. Add a regression test covering
a task claimed by one worker, quarantined, unquarantined, and then successfully
claimed by another worker.

[coderabbitai]

🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

Do not report success when strike cleanup fails.

The task is committed as open before clear_strikes() runs. If cleanup fails, this code only logs the error. The route then returns success with stale strikes still attached. A later verifier can apply the threshold using those stale strikes.

Keep the task quarantined, compensate the status change, or persist a retryable cleanup operation when strike clearing fails.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/projects/task_store.py` around lines 478 - 482, Update the
unquarantine flow around clear_strikes so a cleanup failure is not followed by a
successful response while stale strikes remain. When
self._strikes.clear_strikes(task_id) fails, preserve or restore the task’s
quarantined status, or persist a retryable cleanup operation, and propagate the
failure through the route’s existing error path.

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Reject inactive identities before a denial.

Line 1210 checks only for a missing record. An owner or admin can therefore deny a pending request after the identity becomes suspended or revoked. approve_scope_request rejects inactive identities, and the existence-hiding contract requires inactive targets to return 404.

Use the same active-state predicate as approval. Add a regression test that suspends an identity with a pending request, then verifies that denial returns the unknown-target 404 response.

Proposed fix

-    if record is None:
-        logger.info("scope request deny 404-unknown for %s", canonical_id)
-        raise HTTPException(status_code=404, detail="agent not found")
+    if record is None or record.get("status") != "active":
+        logger.info("scope request deny 404-unknown-or-inactive for %s", canonical_id)
+        raise HTTPException(status_code=404, detail="agent not found or not active")

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/routes/agent_auth_requests.py` around lines 1211 - 1218, Update
the denial flow around the record lookup and ownership check to apply the same
active-state predicate used by approve_scope_request before allowing denial;
inactive or missing identities must return the existing 404 “agent not found”
response. Add a regression test that suspends an identity with a pending scope
request and verifies denial returns the unknown-target 404.

[coderabbitai]

🔒 Security & Privacy | 🟠 Major | ⚡ Quick win

Reject an absent User-Agent for UA-bound sessions.

Each route converts a missing User-Agent header to "". AuthManager.validate_session then skips comparison because it checks if stored_ua and user_agent. A caller with a stolen UA-bound cookie can omit the header and bypass the binding check.

Update the shared validator. If stored_ua exists, reject both a missing User-Agent and a mismatched User-Agent. Keep sessions without user_agent_hash valid for backward compatibility.

• tinyagentos/routes/auth.py#L657-L664: retain User-Agent forwarding after the shared validator rejects empty values.
• tinyagentos/routes/auth.py#L691-L693: retain User-Agent forwarding after the shared validator rejects empty values.
• tinyagentos/routes/canvas.py#L82-L88: ensure an absent User-Agent closes a UA-bound session before websocket.accept().
• tinyagentos/routes/channel_hub.py#L268-L274: ensure an absent User-Agent closes a UA-bound session before connector creation.
• tinyagentos/routes/chat.py#L114-L120: ensure an absent User-Agent closes a UA-bound session before websocket.accept().
• tinyagentos/routes/terminal.py#L26-L30: ensure an absent User-Agent rejects a UA-bound session before the PTY process starts.
• tests/test_auth.py#L1224-L1263: add absent-User-Agent cases for /auth/status and /auth/me.

Proposed shared-validator fix

-        if stored_ua and user_agent:
-            if not secrets.compare_digest(
-                stored_ua, hashlib.sha256(user_agent.encode()).hexdigest()
-            ):
-                return None
+        if stored_ua:
+            if not user_agent or not secrets.compare_digest(
+                stored_ua, hashlib.sha256(user_agent.encode()).hexdigest()
+            ):
+                return None

📍 Affects 6 files

• tinyagentos/routes/auth.py#L657-L664 (this comment)
• tinyagentos/routes/auth.py#L691-L693
• tinyagentos/routes/canvas.py#L82-L88
• tinyagentos/routes/channel_hub.py#L268-L274
• tinyagentos/routes/chat.py#L114-L120
• tinyagentos/routes/terminal.py#L26-L30
• tests/test_auth.py#L1224-L1263

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/routes/auth.py` around lines 657 - 664, Update
AuthManager.validate_session so UA-bound sessions reject both missing and
mismatched User-Agent values, while sessions without user_agent_hash remain
valid. In tinyagentos/routes/auth.py:657-664 and 691-693 retain forwarding; in
tinyagentos/routes/canvas.py:82-88, channel_hub.py:268-274, chat.py:114-120, and
terminal.py:26-30 ensure absent headers are rejected before accepting or
starting work. Add absent-User-Agent coverage for /auth/status and /auth/me in
tests/test_auth.py:1224-1263.

[coderabbitai]

🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

Delete nested HEF model files.

Line 44 makes nested .hef files visible in get_downloaded_models(). delete_model() only searches the top level with models_dir.glob(f"{model_id}*"). It does not find files in the documented <backend>/<family>/<id>/<file> layout. A user can delete the registry entry but retain the multi-gigabyte HEF download.

Proposed fix

-    for f in models_dir.glob(f"{model_id}*"):
+    for f in models_dir.rglob("*"):
         # Use the same canonical suffix set (case-insensitive) the scan uses,
         # so .safetensors/.onnx models aren't left orphaned on disk after a
         # "delete" — the narrower hardcoded list missed them.
-        if f.is_file() and f.suffix.lower() in _MODEL_FILE_SUFFIXES:
+        rel = f.relative_to(models_dir)
+        belongs_to_model = f.name.startswith(model_id) or model_id in rel.parts
+        if (
+            f.is_file()
+            and belongs_to_model
+            and f.suffix.lower() in _MODEL_FILE_SUFFIXES
+        ):
             f.unlink()
             deleted.append(f.name)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tinyagentos/routes/models.py` at line 44, Update delete_model() to locate and
remove nested .hef model files under the documented
<backend>/<family>/<id>/<file> layout, rather than only matching top-level paths
with models_dir.glob(). Preserve deletion of the registry entry and existing
handling for other model file suffixes.

[kilo-code-bot]

CRITICAL: Regex ^\+.* matches +++ diff headers

The pattern ^\+.*class\s+... matches the unified-diff header line +++ b/tinyagentos/<class_name>_store.py for any modified store file whose class name appears in its own filename. A pre-existing store class like MetricsStore in metrics_store.py would be falsely flagged as "newly added" on any PR that touches the file, producing a permanent false-positive CI failure.

Suggested change

	pattern = re.compile(rf"^\+.*class\s+{re.escape(class_name)}\s*\(", re.MULTILINE)
	pattern = re.compile(rf"^\+[^+].*class\s+{re.escape(class_name)}\s*\(", re.MULTILINE)

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

WARNING: Missing active-status guard — inconsistent with approve_scope_request

approve_scope_request (line 1084) rejects inactive agents with 404, but deny_scope_request only checks record is None and owner/admin. A scope request for a suspended agent can still be denied by its owner while approval is blocked, creating a behavioral gap between the two endpoints.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

WARNING: unquarinate_task response omits strike metadata

The get_task route enriches the task dict with strike_count and latest_strike. unquarantine_task returns store.get_task(task_id) directly without adding these fields, so clients expecting strike metadata on all task-detail responses see it missing here.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

WARNING: except Exception misses asyncio.CancelledError in unquarantine_task

The new unquarantine_task catches except Exception around clear_strikes(), but CancelledError is a BaseException, not an Exception. If the coroutine is cancelled during strike clearing, the exception propagates unhandled. The sibling reorder_entries in lists_store.py correctly uses except BaseException for the same cancellation-safety reason.

Suggested change

	except Exception:
	except BaseException:

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

WARNING: rollback() exception may mask the original exception in reorder_entries

The except BaseException handler calls await self._db.rollback() and then raise. If rollback() itself raises, the original exception (e.g. CancelledError or a database error) is lost, making debugging transaction failures significantly harder.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

WARNING: data.get("context_window", 0) returns None when the YAML manifest contains context_window: null

The field is typed as int, but dict.get returns the stored value (including None) when the key exists. A manifest author could legitimately write context_window: null, causing a TypeError in downstream consumers that perform arithmetic or string formatting.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

SUGGESTION: Exact-path matching does not exempt /v1/models/ (trailing slash)

The middleware exemption uses path == "/v1/models" and path == "/v1/chat/completions". A request to /v1/models/ with a trailing slash does not match and falls through to the session gate, returning 401 before the route's consent-key check runs.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

WARNING: data.get("context_window", 0) returns None when the YAML manifest contains context_window: null

The field is typed as int, but dict.get returns the stored value (including None) when the key exists. A manifest author could legitimately write context_window: null, causing a TypeError in downstream consumers that perform arithmetic or string formatting.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

SUGGESTION: Exact-path matching does not exempt /v1/models/ (trailing slash)

The middleware exemption uses path == "/v1/models" and path == "/v1/chat/completions". A request to /v1/models/ with a trailing slash does not match and falls through to the session gate, returning 401 before the route's consent-key check runs.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

Code Review Summary

Status: 7 Issues Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	1
WARNING	5
SUGGESTION	1

Issue Details (click to expand)

CRITICAL

File	Line	Issue
scripts/check_store_wiring.py	96	Regex ^\+.* matches +++ diff headers, causing false-positive CI failures for any store class whose filename contains its class name

WARNING

File	Line	Issue
tinyagentos/routes/agent_auth_requests.py	1210	deny_scope_request lacks active-status guard present in approve_scope_request
tinyagentos/routes/projects.py	1083	unquarantine_task response omits strike_count and latest_strike fields
tinyagentos/projects/task_store.py	481	except Exception misses asyncio.CancelledError in unquarantine_task
tinyagentos/projects/lists_store.py	280	rollback() exception in reorder_entries may mask the original exception
tinyagentos/registry.py	79	context_window=data.get("context_window", 0) returns None for explicit null in manifest

SUGGESTION

File	Line	Issue
tinyagentos/auth_middleware.py	420	Exact-path matching does not exempt /v1/models/ (trailing slash)

Files Reviewed (7 files)

• scripts/check_store_wiring.py - 1 issue
• tinyagentos/routes/agent_auth_requests.py - 1 issue
• tinyagentos/routes/projects.py - 1 issue
• tinyagentos/projects/task_store.py - 1 issue
• tinyagentos/projects/lists_store.py - 1 issue
• tinyagentos/registry.py - 1 issue
• tinyagentos/auth_middleware.py - 1 issue

Fix these issues in Kilo Cloud

---

_{Reviewed by step-3.7-flash · Input: 218.9K · Output: 30K · Cached: 7.4M}

[jaylfc]

Superseded by the sync-branch promotion PR - master's 3 direct Dependabot merges made dev BEHIND under the strict up-to-date rule; same resolution as beta.45/46 (sync branch = dev + merge of master).