chore: give sub-agents Skill access and fix stale KB listings - #415
chore: give sub-agents Skill access and fix stale KB listings#415dpage wants to merge 2 commits into
Conversation
The agent definitions under .claude/agents/ had drifted from reality in three ways, and this change addresses all of them. None of the agents listed the Skill tool, so none of them could invoke the repository's own skills; the fix-issue, codacy and playwright-cli workflows had to be paraphrased into each prompt by hand, which goes stale the moment the underlying skill file changes. Every agent now carries Skill alongside its existing tools. The Knowledge Base sections had fallen behind their directories. The worst case was golang-expert, which listed only testing-strategy.md whilst .claude/golang-expert/ actually holds five files, so the database-scan, metrics-queries, partitioning and rbac-patterns notes were invisible to the agent that owns them. The documentation-writer and security-auditor sections each omitted their README.md. Every listed file now exists on disk and carries a description drawn from the file's own opening section. postgres-expert has no knowledge base directory at all, yet its section was titled "Knowledge Base" whilst pointing at source files; that section is now titled "Authoritative Source Files", which is what it always was. Finally, generic boilerplate that asserted capability without saying anything about this project has gone: the "Core Expertise Areas" lists in golang-expert and react-expert, postgres-expert's six-part catalogue of general PostgreSQL responsibilities together with its "Quality Assurance" section, and security-auditor's "Security Best Practices Reference" snippets. Everything project-specific is untouched, including the four-space indent and copyright header rules, gofmt, the 90% coverage floor and its commands, the OWASP checklist, the MUI and typography rules, and the documentation style guide.
|
Warning Review limit reachedYou’ve reached a temporary PR review limit under our Fair Usage Limits Policy. Next review available in: 49 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (4)
WalkthroughThe pull request updates five Claude agent definitions. It adds ChangesAgent prompt updates
Estimated code review effort: 2 (Simple) | ~10 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Up to standards ✅🟢 Issues
|
|
@coderabbitai review |
✅ Action performedReview finished.
|
![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 current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.claude/agents/documentation-writer.md:
- Around line 31-32: Remove the leading slash from both knowledge-base paths:
update `.claude/agents/documentation-writer.md` lines 31-32 to use
`.claude/documentation-writer/`, and `.claude/agents/security-auditor.md` lines
34-35 to use `.claude/security-auditor/`.
In @.claude/agents/postgres-expert.md:
- Line 4: Update the tool declaration for the advisory agent to remove Bash,
Edit, Write, and Skill, leaving only read-only tools such as Read, Grep, Glob,
WebFetch, WebSearch, and AskUserQuestion.
🪄 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
Run ID: b8729c2e-0b64-4e4a-b490-49d5611bc8fa
📒 Files selected for processing (5)
.claude/agents/documentation-writer.md.claude/agents/golang-expert.md.claude/agents/postgres-expert.md.claude/agents/react-expert.md.claude/agents/security-auditor.md
The agent prompts referred to their knowledge bases as `/.claude/<agent>/`, which literally denotes a directory at the filesystem root; the directories actually sit under `.claude/` in the repository. All four agents that carry a knowledge base now use the repository-relative form, so the convention stays consistent across the set.
1 participant
Summary
Three concrete defects in the sub-agent definitions under
.claude/agents/, none of which touch product code.1. No agent could invoke a skill. None of the five definitions
listed
Skillin theirtools:frontmatter, so the repository's ownfix-issue,codacyandplaywright-cliskills, along with theSuperpowers set, were unreachable from a sub-agent; the workflows had
to be paraphrased into each prompt by hand, which goes stale the moment
the skill file changes. Every agent now carries
Skillalongside itsexisting tools, and nothing else in the frontmatter has moved.
2. The Knowledge Base sections had drifted out of date. The worst
case was
golang-expert, which listed onlytesting-strategy.mdwhilst
.claude/golang-expert/actually holds five files, so thedatabase-scan,metrics-queries,partitioningandrbac-patternsnotes were invisible to the agent that owns them. The
documentation-writerandsecurity-auditorsections each omittedtheir
README.md. Every listed file now exists on disk and carries aone-line description taken from that file's own opening section, rather
than guessed from the filename.
react-expertwas already accurate andis unchanged in this respect.
postgres-experthas no knowledge basedirectory at all, and no directory has been invented for it; its
section was titled "Knowledge Base" whilst in fact pointing at
authoritative source files, so it is now titled "Authoritative Source
Files" and says so plainly.
3. Generic boilerplate trimmed. Prose that asserted capability
without conveying anything project-specific has gone: the "Core
Expertise Areas" lists in
golang-expertandreact-expert(a currentmodel does not need telling what a channel is),
postgres-expert'ssix-part catalogue of general PostgreSQL responsibilities along with
its "Quality Assurance" section and the "world-class ... 15 years"
opener, and
security-auditor's "Security Best Practices Reference"code snippets. This was deliberately conservative: everything
project-specific stays, including the four-space indentation and
copyright header rules,
gofmt, the 90% coverage floor and the exactcommands that measure it, the OWASP checklist and the security
requirements, the KB-update obligation, the MUI and typography rules,
and the whole documentation style guide.
docs/changelog.mdis not updated, because it records user-facingproduct changes and this is internal agent tooling with no effect on
the shipped software.
Test plan
git ls-files .claude/agents/confirms all five definitions aretracked;
.claude/plans/and.claude/specs/remain gitignored anduntouched.
parsed with PyYAML (with the pre-existing unquoted-colon
description:value excluded, since that predates this change andClaude Code's own parser is lenient about it). All five parse, all
five keep
tools:as a single comma-separated line, andname:,description:,model:andcolor:are byte-identical tomain.filesystem: 10 references, 10 files present, no dangling entries.
tools:frontmatter line itself, which has to stay on one line.
.claude/agents/are touched, so the Go and clientsuites are unaffected by this change.
Summary by CodeRabbit