docs: require synchronous dispatch in subagent prompts + add deepseek-v4-pro#1
Merged
Merged
Conversation
Adds deepseek-v4-pro to the reviewer registry. Routes via aichat -> OpenRouter (deepseek/deepseek-v4-pro). 1.6T-param MoE with 49B active, 1M context, released 2026-04-24, $0.435/$0.87 per M. Empirically: returned 1 finding (LOW) on a ~2900-line bundle at 255s during real review session 2026-05-16, vs 0 findings from opencode on same bundle. Useful breadth additive — not redundant with v3.2 or kimi.
Adds CRITICAL warning + subagent prompt template to SKILL.md §6a.
Failure mode this prevents: when a calling agent spawns subagents
without explicit synchronous-execution wording, the subagent commonly
runs dispatch.py via its own run_in_background, then ends its own
session before dispatch.py exits. $RUN_DIR/reviews/<name>.json is
never written; the parent sees a 'completed' subagent that reports a
plan ('I'll wait for the dispatch...') instead of results.
Observed 3/4 times on subagent dispatch in real session 2026-05-16
(argus-codex, argus-deepseek, argus-gemini all bailed; argus-opencode
+ argus-kimi correctly waited). Pattern is consistent enough to
treat as a default-behavior bug in subagent prompts that omit the
explicit foreground-wait directive.
Fix is documentation-only:
- SKILL.md §6a: adds 'CRITICAL: subagent execution semantics' subsection
with the failure-mode fingerprint and a minimal safe prompt template
- CLAUDE.md: cross-references SKILL.md §6a for the requirement
No code change — argus's dispatch.py is correct; the failure is in how
callers instruct their subagents. Catching this at the skill-doc layer
means every project using argus inherits the fix immediately.
There was a problem hiding this comment.
Pull request overview
This PR updates Argus documentation to prevent a known silent failure mode in subagent-per-reviewer dispatch, and extends the reviewer registry to include the new deepseek-v4-pro model.
Changes:
- Documented that subagents must run
dispatch.pysynchronously in their own session (norun_in_backgroundinside the subagent) and added a minimal safe prompt template (SKILL.md§6a). - Added a cross-reference in
CLAUDE.mdreinforcing the synchronous subagent prompt requirement. - Registered the
deepseek-v4-proreviewer inconfig.yaml(OpenRouter route, 1M ctx).
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| SKILL.md | Adds a “CRITICAL” subsection clarifying subagent execution semantics and provides a safe subagent prompt template. |
| config.yaml | Registers deepseek-v4-pro as a paid OpenRouter-backed reviewer with 1M context. |
| CLAUDE.md | Cross-references the new SKILL.md §6a requirement near the existing dispatch guidance. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
added 2 commits
June 10, 2026 07:37
Roster badge 14 -> 15, roster table row, and OPENROUTER_API_KEY coverage counts (8 of 14 -> 9 of 15) to match the new config.yaml entry. Also brings the branch up to date with master post-PR #12 (clean auto-merge). Confidence: high Scope-risk: narrow
Owner
Author
|
Post-#12 status check before merging:
Merging once CI is green. 🤖 Generated with Claude Code |
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
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
Two related changes from a live review session on 2026-05-16:
deepseek-v4-proinconfig.yaml— 1.6T MoE, 1M ctx, OpenRouter route.SKILL.md§6a + cross-reference inCLAUDE.md.Why the docs change matters
Subagent-per-reviewer dispatch (the default per
SKILL.md§6a) has a silent failure mode when subagent prompts omit explicit synchronous-execution wording: the subagent runsdispatch.pyvia its ownrun_in_background, then ends its session beforedispatch.pycompletes, and\$RUN_DIR/reviews/<name>.jsonis never written. The parent sees a "completed" subagent that reports a plan ("I'll wait for the dispatch...") rather than results.Observed 3 out of 4 times in a real subagent dispatch session (2026-05-16):
argus-opencodeargus-codexargus-deepseekargus-geminiargus-kimiThe 60% failure rate is high enough to treat as a default-behavior bug in subagent prompts that omit the directive, not a one-off.
Fix
Documentation-only.
dispatch.pyitself is correct — the failure is in how callers instruct their subagents. The fix is:SKILL.md§6a now has aCRITICAL: subagent execution semanticssubsection that explains the fingerprint and provides a minimal safe prompt templateCLAUDE.mdcross-references the SKILL.md §6a requirement near the existing dispatch-pattern guidanceEvery project using argus picks up the fix immediately via the skill load — no per-caller change required.
Test plan
SKILL.md§6a renders correctly (markdown headers consistent with existing structure)CLAUDE.mdcross-reference is correct🤖 Generated with Claude Code