feat(chat): always-on system context manual injected into every agent dispatch#295
Merged
Conversation
…ispatch Adds agent_manual.py with build_manual() — a pure function that returns a per-agent operating manual (≤500 tokens) covering @-mention routing, project task verbs, and lead/non-lead status. The router prepends it as role:system before the conversation history on every chat dispatch so agents always know the channel primitives without per-session briefs.
📝 WalkthroughWalkthroughA new Changes
Sequence DiagramsequenceDiagram
actor User
participant Router as AgentChatRouter
participant ManualBuilder as build_manual()
participant Context as Message Context
participant Queue as bridge.enqueue_user_message
User->>Router: sends message in channel
Router->>ManualBuilder: invoke for each recipient<br/>(channel, agent_name, leads)
ManualBuilder->>ManualBuilder: analyze channel type & role
ManualBuilder-->>Router: return agent-specific manual text
Router->>Context: prepend {"role": "system",<br/>"content": manual} to context
Router->>Queue: enqueue message with<br/>per-recipient agent_context
Queue-->>User: message routed to recipient<br/>with context-injected instructions
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Review rate limit: 9/10 reviews remaining, refill in 6 minutes. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@tinyagentos/agent_chat_router.py`:
- Around line 171-172: The current code unconditionally prepends
build_manual(channel, agent_name, leads) into agent_context which injects
channel-routing and visibility rules into threaded deliveries; modify the logic
where agent_context is built (the lines that set manual_text and agent_context)
to detect when thread_id is set and either (a) call a new thread-specific manual
generator (e.g., build_thread_manual(channel, agent_name, leads, thread_id))
that omits or replaces routing/visibility sections, or (b) call build_manual
with a flag (e.g., build_manual(..., for_thread=True)) so it suppresses
routing/visibility content for threads; ensure resolve_thread_recipients()
remains the single source of truth for visibility when thread_id is present and
use that branch to produce the appropriate manual_text or no routing text at all
before composing agent_context.
In `@tinyagentos/agent_manual.py`:
- Around line 97-122: build_manual currently ignores settings["response_mode"]
and always injects _SECTION_MENTIONS for group/topic channels; update
build_manual to read response_mode = settings.get("response_mode", "lively") (or
similar) and branch when channel_type in ("group","topic") to append the correct
mention/visibility copy based on response_mode (e.g., append
_SECTION_MENTIONS_QUIET when response_mode == "quiet" and append the existing
_SECTION_MENTIONS or a new _SECTION_MENTIONS_LIVELY when response_mode ==
"lively"), leaving DM and A2A logic (project_id/is_a2a, _SECTION_TASK_VERBS,
_SECTION_LEAD_IS/_SECTION_NON_LEAD, _SECTION_QUICK_REF) unchanged; add any new
constants for the alternative text if needed and ensure references use settings
and response_mode so the manual matches the router behavior.
🪄 Autofix (Beta)
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: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 9cb34d8f-86ed-4fb5-8c84-4da8cf98e60c
📒 Files selected for processing (4)
tests/test_agent_chat_router.pytests/test_agent_manual.pytinyagentos/agent_chat_router.pytinyagentos/agent_manual.py
Comment on lines
+171
to
+172
| manual_text = build_manual(channel, agent_name, leads) | ||
| agent_context = [{"role": "system", "content": manual_text}, *context] |
There was a problem hiding this comment.
Avoid injecting the channel-routing manual into thread deliveries as-is.
When thread_id is set, Lines 54-58 route through resolve_thread_recipients() instead of the normal channel quiet/lively rules. Prepending the same manual here means a threaded recipient can be told they receive every channel message or only @-mentions, even though visibility is now governed by the thread resolver. Please add a thread-specific manual branch, or suppress the routing/visibility sections for threaded dispatches.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@tinyagentos/agent_chat_router.py` around lines 171 - 172, The current code
unconditionally prepends build_manual(channel, agent_name, leads) into
agent_context which injects channel-routing and visibility rules into threaded
deliveries; modify the logic where agent_context is built (the lines that set
manual_text and agent_context) to detect when thread_id is set and either (a)
call a new thread-specific manual generator (e.g., build_thread_manual(channel,
agent_name, leads, thread_id)) that omits or replaces routing/visibility
sections, or (b) call build_manual with a flag (e.g., build_manual(...,
for_thread=True)) so it suppresses routing/visibility content for threads;
ensure resolve_thread_recipients() remains the single source of truth for
visibility when thread_id is present and use that branch to produce the
appropriate manual_text or no routing text at all before composing
agent_context.
Comment on lines
+97
to
+122
| channel_type = channel.get("type") or "" | ||
| project_id = channel.get("project_id") | ||
| settings = channel.get("settings") or {} | ||
| is_a2a = settings.get("kind") == "a2a" | ||
|
|
||
| # DM: header + stub only. | ||
| if channel_type == "dm": | ||
| return _HEADER + _DM_STUB | ||
|
|
||
| # Group / topic channels. | ||
| parts = [_HEADER] | ||
|
|
||
| # Section 1: @-mention routing (group and topic channels). | ||
| if channel_type in ("group", "topic"): | ||
| parts.append(_SECTION_MENTIONS) | ||
|
|
||
| # Sections 2 + 3: project a2a only. | ||
| if project_id and is_a2a: | ||
| parts.append(_SECTION_TASK_VERBS) | ||
| if agent_name in leads: | ||
| parts.append(_SECTION_LEAD_IS) | ||
| else: | ||
| parts.append(_SECTION_NON_LEAD) | ||
|
|
||
| # Section 4: always for non-DM. | ||
| parts.append(_SECTION_QUICK_REF) |
There was a problem hiding this comment.
Make the manual respect response_mode.
build_manual() never looks at settings["response_mode"], but Line 67 and Lines 96-99 in tinyagentos/agent_chat_router.py still fan lively channels out to every non-muted member. In a lively group/A2A channel this manual can tell agents “A reply with no @-tag reaches NO ONE” and “You only see messages where you're explicitly @-tagged” even though the router does the opposite. Please branch the mention/visibility copy on quiet vs lively mode before injecting it.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@tinyagentos/agent_manual.py` around lines 97 - 122, build_manual currently
ignores settings["response_mode"] and always injects _SECTION_MENTIONS for
group/topic channels; update build_manual to read response_mode =
settings.get("response_mode", "lively") (or similar) and branch when
channel_type in ("group","topic") to append the correct mention/visibility copy
based on response_mode (e.g., append _SECTION_MENTIONS_QUIET when response_mode
== "quiet" and append the existing _SECTION_MENTIONS or a new
_SECTION_MENTIONS_LIVELY when response_mode == "lively"), leaving DM and A2A
logic (project_id/is_a2a, _SECTION_TASK_VERBS,
_SECTION_LEAD_IS/_SECTION_NON_LEAD, _SECTION_QUICK_REF) unchanged; add any new
constants for the alternative text if needed and ensure references use settings
and response_mode so the manual matches the router behavior.
3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
tinyagentos/agent_manual.pywithbuild_manual(channel, agent_name, leads) -> str— a pure function, no IO, no globals beyond constant strings. Returns a markdown operating manual covering @-mention routing, project kanban verbs, and lead/non-lead status, branched per channel type and agent role.agent_name in leads.role: "system"atcontext[0]inagent_chat_router.py(afterbuild_context_window, before the per-recipient loop). Built per-agent so lead vs non-lead text is correct for each recipient.tests/test_agent_manual.pyand 3 new end-to-end tests intests/test_agent_chat_router.py. All 49 tests in both files pass. 199 tests pass across the wider suite (1 unrelated disk-full environmental error intest_routes_beads.py).Token count
Longest branch (project a2a, lead agent): 417 estimated tokens (words × 1.3 heuristic; 1869 chars, 321 words). Well within the 500-token budget.
Hook point
tinyagentos/agent_chat_router.pylines 145–170 (aftercontext = build_context_window(...), inside the per-recipientfor agent_name in recipients:loop, beforebridge.enqueue_user_message). Theleadslist is extracted fromsettings.get("leads")once before the loop;build_manualis called per-agent so each agent gets the correct lead/non-lead branch.No conflicts with in-flight PRs
feat/taos-assistant) — touchesdesktop/src/,tinyagentos/routes/taos_agent.py, and docs. No overlap.feat/image-gen-lifecycle-fixes) — touchestinyagentos/routes/models.pyandtinyagentos/services/rknn_sd_server.py. No overlap.Test plan
pytest tests/test_agent_manual.py— 24 tests covering all channel type branches, empty leads list, token budgetpytest tests/test_agent_chat_router.py— all pre-existing tests pass; 3 new e2e tests verify manual injected at context[0] with correct lead/non-lead textbuild_manualis a pure function (no IO, no side effects)role: "system"in routerSummary by CodeRabbit