Repository navigation
Drop AGENTS.md skill pointers that agents do not need to load the skill (PROF-16586) - #986
Conversation
…gger (PROF-16586) The skills load from their descriptions, so the pointers cost every session's context without changing behavior. Agentic Code Review shrinks to one line: the pre-pr-review skill holds the procedure and the hooks enforce it. The Work Tracking paragraph drops GitHub issues for external contributors. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…OF-16586) The page now names Jira as the tracker, as AGENTS.md's Work Tracking paragraph does. The PR ticket check still accepts a GitHub issue #<n>. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Scan-Build Report
Bug Summary
Filter Results by FileReports
|
|||||||||||||||||||||||||||||||||||||||||||||||||
The pre-pr-review skill loads from its description and holds the whole procedure, and the hooks enforce it. The comment-budget rule now names the agentic pre-PR review directly. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e row (PROF-16586) AGENTS.md said that the obligations that trigger a procedure stay inline and that a hoisted section leaves a pointer behind, which this change no longer follows. A skill's description now carries the obligation to run it, and AGENTS.md points to a skill only when a test drive shows the description does not load it. The pre-pr-review gate row named the removed AGENTS.md review section; it now covers any text that says when a review is required or which reviewer runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CI Test ResultsRun: #43440850263 | Commit:
Status Overview
Legend: ✅ passed | ❌ failed | ⚪ skipped | 🚫 cancelled Summary: Total: 32 | Passed: 32 | Failed: 0 Updated: 2026-11-13 14:37:55 UTC |
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3999a644ea
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
There was a problem hiding this comment.
More details
Skill descriptions retain the removed procedure triggers, and reviewer selection still covers review gates in AGENTS.md and skills. Harnesses without skill discovery or hooks lose the explicit review obligation, an accepted limitation documented in the PR.
🤖 Bits Code Review · Commit 3999a64
Conflict in AGENTS.md: main added a sentence to the PR Descriptions section, which this branch removes. The sentence is dropped: the pre-push hook warning names the pr-writing skill, and the skill holds the re-verify rule. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The PR ticket check and the README still accept and invite GitHub-issue contributions, so the docs keep that path until all three change together. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
This comment will be updated automatically if new data arrives. 🔗 Commit SHA: f434515 | Docs | View more details | Give us feedback! |
What does this PR do?:
Removes every
AGENTS.mdpointer to a skill, because each skill loads from its owndescription:verify-profiler-change(Development Guidelines)ticket-writing,work-planning,ticket-triage(Work Tracking)pre-pr-review,pr-review(Agentic Code Review)pr-writing,ci-script-change,doc-writingdoc-writingalready states the CI naming and index check.It also changes the rules that
AGENTS.mdsets for itself, since they asked for these pointers:descriptioncarries the obligation to run it.AGENTS.mdpoints to a skill only when a test drive shows that the description alone does not load it.doc/still leaves a one-line pointer behind. A section that moves into a skill follows the skill-pointer rule.pre-pr-review(and its gloss inpr-review), the row named "the AGENTS.md review rules", which was the removed section. It now covers any text inAGENTS.mdor a skill that says when a review is required or which reviewer runs.AGENTS.mdgoes from 550 to 504 lines (wc -lonorigin/mainand on this branch).Motivation:
Every session loads
AGENTS.md. Both harnesses already show each skill'snameanddescriptionat session start, and load the skill when the request matches. A pointer that does not change behavior costs every context for nothing.The Agentic Code Review section repeated the
pre-pr-reviewskill: the fix-or-waive rule, the reviewer choice, the self-checks, the marker, the no-reviewer sign-off and the hooks. The skill keeps all of them, and the hooks enforce the review.Additional Notes:
AGENTS.md, and the PR changes its own layering rule.AGENTS.mdbut neither loads.agents/skills/nor runs our hooks no longer sees the review obligation.pre-pr-reviewdoes not install.githooks/pre-push; the skill's step 0 installs it./code-reviewhigh on the section removal: it found thatAGENTS.md's layering and hoisting rules still required the pointers, and that the gate row named the removed section. The rule changes above fix both.AGENTS.mdanddoc/WorkTracking.mdstill name GitHub issues as the tracker for external contributors. ThePR ticket referencecheck and the README still accept and invite that route, so the docs change only together with them, in PROF-16620.How to test the change?:
Headless skill-load drives, each in a fresh clone of the candidate, without MCP servers and with read-only tools (plus edit tools for
ci-script-change, andgh pr view/gh pr diffforpr-review). Each skill has one direct prompt and one borderline prompt that does not name the artifact. Each prompt ran twice in each harness. A load counts only from a tool call: aSkillcall in Claude Code, or a read of the skill'sSKILL.mdin Pi.3999a644eClaude Code ran
claude-opus-5-5; Pi ranglm-5-3.After
3999a644e, the branch mergesmain(#980), then restores the GitHub-issue clause. No skill pointer and no skilldescriptionchanged after3999a644e. The merge's only skill change is #980's new rule in the body ofpr-writing, and the head'sAGENTS.mddiffers from3999a644eonly in the restored clause.The merge had one conflict. #980 added "After pushing commits to a branch with an open PR, re-verify the PR description…" to the PR Descriptions section, which this PR removes. The sentence is dropped. The
pre-pushwarning names thepr-writingskill, and the skill holds the rule./code-reviewmedium found no issue in the resolution.Local checks on the head (
f434515ea):utils/tests/test_quality_check_scripts.sh:PASS: 231 test groupsutils/tests/test_review_marker.sh:268 tests, 0 failed:ddprof-lib:qualityChecks:BUILD SUCCESSFUL,skills: OK (9 skills),doc index: OK (30 docs)For Datadog employees:
credentials of any kind, I've requested a security review (run the
dd:platform-security-reviewskill, or file a request via the PSEC review form).
bewairealso runs automatically on every PR.🤖 Generated with Claude Code