| name | review-pr |
| description | Reviews the changes a branch introduces — the diff of the current branch (HEAD) against a base branch, `develop` by default — by orchestrating this repository's specialised review agents: `jwt-allauth-security-reviewer` (session and token invariants, and delegating to allauth), `jwt-allauth-efficiency-reviewer` (the stateless path, the rotation lock, simplicity) and `jwt-allauth-surface-reviewer` (backwards compatibility of settings and endpoints, documentation, release notes, and the claims that go stale). Computes the diff from the merge base (three-dot, the way GitHub shows a PR), picks the agents the diff actually concerns, launches them IN PARALLEL, and aggregates their findings into one report ordered by severity with an overall verdict. Optionally takes the base branch as an argument: `/review-pr main` compares against `main` instead of `develop`. Read-only: it audits and reports; it changes nothing. Use it before opening or merging a pull request, or whenever you want a full review of a branch. Triggers on: "review the PR", "review-pr", "review the branch changes", "review the diff", "review this branch against X".
|
| arguments | ["base_branch"] |
| argument-hint | [base-branch] — default develop, e.g. main |
| allowed-tools | Agent Read Grep Glob Bash(git diff *) Bash(git log *) Bash(git merge-base *) Bash(git rev-parse *) Bash(git branch *) Bash(git fetch *) Bash(git status *)
|
| model | opus |
Review PR — reviewing a branch through the repository's agents
Reviewing a change to this library covers three independent dimensions,
and each has its own read-only auditor in .claude/agents/. A single
prompt trying to check all three at once dilutes every one of them, so
this skill delegates each dimension to its specialist and aggregates
the results.
The skill orchestrates: it resolves the base branch, computes the PR
diff, decides which agents apply, launches them in parallel and
consolidates their reports. It does not review the code itself — that
work lives in the agents. It is read-only: neither the skill nor its
agents modify code or documentation; they propose the fix, they do not
apply it.
What this skill deliberately does not cover: general correctness,
naming, dead code and style. /code-review and /simplify already do
that, and duplicating them here would produce two reports arguing about
the same lines. The three agents carry the same rule and refer such
findings on rather than developing them.
Input
$ARGUMENTS optionally holds the base branch to compare against:
- (empty) → compares
HEAD against develop (the default).
main → compares HEAD against main.
The branch under review is always the current one (HEAD); the argument
only changes the base. Do not expect a second argument.
Execution
Step 1 — Resolve the base branch and the diff range
base = $ARGUMENTS after trim; if empty, base = develop.- Resolve the real reference, preferring the remote (it is closer to
what the PR will show):
- Refresh, best-effort and non-blocking:
git fetch origin <base>. If
the network fails, retry with backoff (2s, 4s, 8s, 16s) per this
environment's git rules, then carry on with what is local.
- Use
origin/<base> if it verifies (git rev-parse --verify origin/<base>); otherwise the local <base>.
- If neither exists → stop and ask for a valid base, listing the
candidates with
git branch -a. Do not invent one.
