From review-cycle
Run reviewers in parallel and report findings without applying fixes. Use for sanity checks, mid-implementation inspection, or to see what reviewers think before committing to a full cycle. Does NOT modify code, does NOT update the review sentinel, does NOT loop.
How this skill is triggered — by the user, by Claude, or both
Slash command
/review-cycle:inspect [--base <ref>][--base <ref>]This skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Read-only review pass. Runs the same reviewers as `/review-cycle:review` but does not apply fixes, does not loop, and does not update the review sentinel.
Read-only review pass. Runs the same reviewers as /review-cycle:review but does not apply fixes, does not loop, and does not update the review sentinel.
Use this when you want to see findings without committing to a fix cycle — for example, a mid-implementation sanity check, a pre-commit final inspection, or when you want to triage findings manually rather than letting the cycle decide.
$ARGUMENTS may contain:
--base <ref> — scope review to git diff <ref>..HEAD instead of git diff HEADVerify in a git repo:
git rev-parse --is-inside-work-tree
PROJECT_ROOT=$(git rev-parse --show-toplevel)
Verify uncommitted changes exist:
git status --porcelain --untracked-files=all
If empty, report "nothing to review" and stop.
Verify Codex CLI is available by running codex --version. If it fails, surface the error and continue with pr-review-toolkit only — note in the summary that Codex was skipped.
In a single conversation turn, invoke:
Codex review (background) — direct CLI invocation:
Bash({
command: "cd \"$PROJECT_ROOT\" && codex review --uncommitted",
description: "Codex review",
run_in_background: true
})
No dependency on the codex Claude plugin; uses the codex CLI directly. Multi-agent parallelism comes from multi_agent = true in ~/.codex/config.toml.
Bundled review subagents (parallel) — spawn each applicable agent via the Agent tool with run_in_background: true. Conditional dispatch based on diff scope:
review-cycle:code-reviewer — alwaysreview-cycle:pr-test-analyzer — if diff touches test filesreview-cycle:silent-failure-hunter — if diff touches error-handling codereview-cycle:type-design-analyzer — if diff adds or modifies type declarationsWait for the codex bash output AND every spawned subagent to complete.
Collect findings from both reviewers. Attribute sources. Group by file.
Print findings as a structured summary:
Review inspect complete.
Critical (N):
- file:line — issue (source)
- ...
Important (N):
- file:line — issue (source)
- ...
Suggestions (N):
- file:line — issue (source)
- ...
Strengths:
- what's well-done in these changes
Then a "Next steps" block:
Next steps:
- Address findings automatically: /review-cycle:review
- Triage manually: edit files directly
- Bypass the gate for this project: touch .claude/.no-review-gate
Do NOT apply any fixes. Do NOT update the sentinel. Do NOT create tickets. Do NOT commit.
The Stop hook will still gate on unreviewed changes — /review-cycle:inspect does not satisfy the gate (because no review-cycle has run, no sentinel was updated). To clear the gate, either run /review-cycle:review or use the per-project opt-out marker.
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub oakoss/claude-plugins --plugin review-cycle