docs(amos-next): mandate Opus for staleness + flag-and-verify review gate - #45
Merged
Merged
Conversation
…rotocol for review gate Two changes to `.claude/skills/amos-next/SKILL.md`: **Step 2.5 (staleness check)** — make Opus mandatory and explicit: - Reasoning-heavy step; never falls back to Sonnet. Whenever a subagent is spawned for verification, pass `model: "opus"`. - Use max effort: full reads, not skim. Treat AI Agent Notes as hard-won research the user wants preserved, not suggestions to override. - Default direction on contradictions is "the body is right, current code is the drift" — flip only when evidence is unambiguous. - For non-trivial verification spawn parallel Opus general-purpose subagents (one per concern, in a single message). `subagent_type=Explore` is for narrow lookups only. - When body research disagrees with fresh analysis on architecture specifically (not file paths or method names), pause and surface to the user before locking the plan. **Step 7 (pre-PR review)** — advisory verdict, session-agent owns the final judgment: - The pr-review-gate skill (spawns an Opus reviewer) flags concerns; the session agent verifies each one against actual code and decides what to do. - PASS / FIX / DISCUSS are reviewer recommendations, not binding. - Even on PASS, independently spot-check issue-body fidelity, behavior preservation, and test quality before opening the PR. - The session agent should leave Step 7 with complete understanding of the issue and current code, and high confidence the body of work is correct. If not, dig further or surface to the user. Why: Sonnet defaults applied CLAUDE.md naming rules mechanically in a recent review and flagged names that came verbatim from an issue body — wasted attention. Worse, on architecture-flavored issues, Sonnet risks overriding hard-won research the user deliberately captured. Opus everywhere reasoning is needed, with the session agent as the final decision-maker, fixes both.
tato123
added a commit
that referenced
this pull request
Jul 29, 2026
It lived only in ~/.claude as an untracked copy — usable on one machine, lost on reinstall, and invisible to the installer that provisions every other skill. amos-next has invoked it as a mandatory step since #45, so a skill the workflow depends on was the one skill not shipped with the workflow. Carries the P0 priority gate added in 011d6c1: a reviewer flagging something is agent-initiated by definition — it is a tool the agent invoked, not the user asking for a ticket. Verifying a finding is real settles whether it is true, not whether it is worth tracking. Hard rule 6 scopes the skill's autonomy to the open-PR decision and nothing else. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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
Two updates to
.claude/skills/amos-next/SKILL.md:Why
Sonnet defaults applied CLAUDE.md naming rules mechanically in a recent review and flagged names that came verbatim from an issue body — wasted attention. On architecture-flavored issues, Sonnet risks overriding hard-won research the user deliberately captured in AI Agent Notes. Opus everywhere reasoning is needed, with the session agent as the final decision-maker, fixes both failure modes.
Test plan
🤖 Generated with Claude Code