メニュー
Audit whether changed code conforms to the project's own rules in .claude/rules/ and the conventions in CLAUDE.md. Read-only; produces a conformance report and never edits code.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Run a full code review of the current change — fan out every relevant review concern across independent subagents (the local review skills plus the ably-skills code-review-* plugin) and synthesise one consolidated report. Read-only; never edits code.
Audit whether the demos under demo/ are still current with a change — they compile against and use the changed public API correctly, showcase new or renamed capabilities, and keep their e2e test coverage current. Read-only; produces a conformance report and never edits code.
Review documentation for accuracy, completeness, and conformance to the project's doc standards — whether a code change left docs/ stale or incomplete, or as the review half of the docs writing loop. Read-only; produces a conformance report and never edits code or docs.
Audit whether the .claude/rules/* guidance still matches the code after a change — stale conventions, drifted examples, rules describing removed behaviour, or new conventions the change introduces that a rule should capture. Read-only; produces a conformance report and never edits code or rules.
Write or improve markdown documentation for the AI Transport SDK. Follows the project's doc principles (atomic pages, problem-first framing, one path then depth) and the established style from the docs that received positive feedback.
Generate a commit message for the current staged changes and commit
| name | code-review-rules-conformance |
| description | Audit whether changed code conforms to the project's own rules in .claude/rules/ and the conventions in CLAUDE.md. Read-only; produces a conformance report and never edits code. |
| license | proprietary |
| metadata | {"team":"engineering","tags":"code-review, rules, conventions, conformance, quality","mcp":false} |
| allowed-tools | Read, Grep, Glob, Skill, Bash(git diff *), Bash(git status *) |
Audit whether the changed code obeys the project's rules.
Invoke the ably-skills:code-review-protocol skill (via the Skill tool). It
defines the report-only rule, scope resolution (honouring any $ARGUMENTS
override), the severity scale, and the exact report format. Use
concern = rules conformance. Apply the criteria below, then emit the report
it specifies.
Read every rule file under .claude/rules/ and the conventions in CLAUDE.md
before judging — they are the authority. This concern checks conformance to the
project's own rules; generic logging and correctness hygiene belong to the
ably-skills:code-review-logging and ably-skills:code-review-correctness
concerns.
For each changed file, check it against each rule file you read in Step 1 and
the conventions in CLAUDE.md. Work from the rules as written in those files
today — .claude/rules/ is the source of truth and its set changes over time, so
do not rely on a remembered or hardcoded list. Read each rule, then hold the
change to what it actually says, covering whatever it addresses (type safety,
error handling, logging, promises, abstractions, comments, framework specifics,
and the CLAUDE.md conventions).
Cite the specific rule file — and the clause within it — in every finding, and
state the rule the change breaks. Per CLAUDE.md, the rules state principles and
point to the authoritative source in code; apply the principle, not a stale
summary of it.
A rule violation is at least major; a violation the rule itself calls out as a
hard contract (non-negotiable) is a blocker.
