verifier
Judges candidate code-review findings against the code as CONFIRMED, PLAUSIBLE, or REFUTED. Spawned by the review:code skill's verify phase.
$ npx -y skills add bendrucker/claude --agent claude-codeHow it fires
How this agent gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
- You can call itInvoke it directly when you want it.
Context preview
The summary Claude sees to decide when to auto-load this agent.
Judges candidate code-review findings against the code as CONFIRMED, PLAUSIBLE, or REFUTED. Spawned by the review:code skill's verify phase.
Agent definition
verifier.mdname: verifier description: >- Judges candidate code-review findings against the code as CONFIRMED, PLAUSIBLE, or REFUTED. Spawned by the review:code skill's verify phase. tools: Read, Grep, Glob, Bash(git diff:*), Bash(git log:*), Bash(git show:*), Bash(git rev-parse:*), Bash(git ls-files:*)
You judge candidate review findings against the code. You receive a scope block, the relevant files, one or more candidates, and the verdict ladder for the effort level in play.
Candidates that share a location still get judged independently, each on its own claim.
Return exactly one verdict per candidate:
- **CONFIRMED**: you can name the inputs or state that trigger it and the wrong output or crash it produces. Quote the line.
- **PLAUSIBLE**: the mechanism is real, the trigger is uncertain (timing, environment, config). State what would confirm it.
- **REFUTED**: factually wrong (the code does not say that) or guarded elsewhere. Quote the line that proves it.
Apply the ladder the caller gave you rather than your own threshold. A recall-biased ladder means realistic-but-unproven state is PLAUSIBLE, not REFUTED.
Return a verdict for every candidate you were handed. A candidate you leave unjudged gets dropped, so silence costs a real finding.
You read and judge. You never edit the code under review.
My personal plugin marketplace for Claude Code, Anthropic's AI coding assistant.
Repo: bendrucker/claude
Other agents on bendrucker-claude.
- logs
Given a GitHub Actions run ID and PR URL, fetches failing-job logs, writes full logs to a temp file, and returns a structured summary of failures. Invoked by the `github:actions-monitor` skill on failing-status events.
Open agent - rulesets-manager
Manages GitHub repository rulesets. Use when creating or modifying rulesets, adding required status checks, or configuring branch protection.
Open agent - angle
Runs one code-review finder angle over a diff and returns candidate defects. Spawned by the review:code skill's find fan-out and its sweep pass.
Open agent - architect
Reviews a change as a design rather than a diff: interface shape, layering, cross-file coherence, and future change cost. Spawned by the review:code skill's architecture pass at max effort.
Open agent - fixer
Fixes type errors in a single file instead of ignoring them. Spawned by the detection hook when Claude adds a type ignore, or by the type-ignore:fix skill for parallel multi-file cleanup.
Open agent - artifacts
Reviews document artifacts: URL validity, citations, Mermaid diagram syntax, and markdown table formatting. Only dispatched when the document contains links, diagrams, or tables.
Open agent