toolkit

Multi-agent adversarial code review (5 lenses, 3 refuters per finding, critic). TRIGGER ONLY on an explicit ask — the user types /adversarial-review, or says "adversarial review" / "adversarially review" / "run the adversarial review" naming a PR, branch, or the working tree. DO NOT trigger for "review this", "code review", "thorough review", "look over my changes", or CodeRabbit follow-ups — those are /code-review (or the coderabbit skill). Never self-trigger on PR size or risk; at most SUGGEST it in one line when a diff touches auth, migrations, concurrency, or exceeds 500 lines.

Thorough multi-agent code review — 5 lenses, refuters, critic, plus style nitpicks and committable suggestions — for a diff, branch, or PR. Produces a categorized report only; never touches GitHub. TRIGGER ONLY on /deep-review, or an explicit ask naming "deep review" / "thorough categorized review" for a specific target. DO NOT trigger for bare "review this" / "code review" (the official code-review plugin handles that), for "adversarially review" (that's /adversarial-review directly), or for CodeRabbit follow-ups (the coderabbit skill).

End-to-end issue workflow composed of separate workflows with human gates between them — issue-triage (verify the premise) → decision gate → fix-plan (opus) → plan gate → fix-small → optional adversarial-review → PR gate → pr-open → pr-watch (review loop) → report. TRIGGER ONLY when the user types /issue-run, or literally says "run the issue workflow" / "issue-run" for a specific issue. DO NOT trigger for "fix issue 12", "look at this issue", "what does issue 12 say" — handle those directly. ~15 agents per run (~45 with --review).

Set up a new GitHub repository the way Cedric usually does — repo settings, CI, semantic PR titles, release-please, Dependabot, CLAUDE.md, and the Claude plugin pin with its auto-update workflow and cloud-session install hook. TRIGGER ONLY on /new-repository <name>, or an explicit ask to create/bootstrap/set up a new repo. DO NOT trigger for editing an existing repo, or for a bare "create a project" that doesn't ask for the repo setup.

Review a GitHub PR with the same multi-agent engine as deep-review, then submit a real GitHub review — a decision (approve/request changes/comment), a summary, and inline line comments with committable suggestions. TRIGGER ONLY on /pr-review <PR>, or an explicit ask naming a specific PR to review and submit a verdict on. DO NOT trigger for bare "review this" / "code review" without a PR, for "adversarially review" (use /adversarial-review directly, which never posts a formal review), for a report-only request (use deep-review), or for CodeRabbit follow-ups (the coderabbit skill).

Instrument any application with OpenTelemetry and ship its traces, logs, and metrics to the user's SignalDB deployment — provisioning a tenant and API key, configuring OTLP export with SignalDB's auth headers, and verifying data arrives via the query APIs. Use this whenever the user wants to "observe" an app, add telemetry/tracing/logging/metrics to a project, send OpenTelemetry data to SignalDB, set up observability for a service, or connect an app to their SignalDB instance — even if they don't name SignalDB explicitly (it is their default observability backend). Also use when querying SignalDB (TraceQL/LogQL/PromQL), managing SignalDB tenants or API keys, or debugging why telemetry isn't showing up.