How do I install the Bmh Review Work skill?

Run npx skillmds@latest add vinilana/bmh-review-work in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Bmh Review Work safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Bmh Review Work?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Bmh Review Work free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Bmh Review Work?

vinilana (@vinilana) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Bmh Review Work by vinilana — Agent Skill

Name: Bmh Review Work
Author: vinilana

Bmh Review Work

Always use when reviewing, validating, auditing, or summarizing completed Bench My Harness work, including code changes, tests, docs, ADRs, schemas, adapters, CLI behavior, benchmark runner behavior, or pull-request style review. Enforces TDD acceptance criteria, hexagonal architecture, v1 Codex/Claude Code scope, redaction, source confidence, and quality gates before work is considered done.

vinilana Updated Sep 17, 2026

File contents

BMH Review Work

Workflow

Review as a code reviewer first: findings before summary, ordered by severity, with file and line references.
Read references/review-gates.md before judging completeness.
Check whether the change was driven by tests. New behavior without tests is a finding.
Verify hexagonal boundaries: domain must not import adapters, CLI, filesystem, process APIs, or provider-specific schemas.
Verify v1 scope: only Codex and Claude Code adapters are production scope.
Verify observability integrity: tokens, context, and cost must carry source and confidence.
Verify security: secrets are redacted before reports; raw payload handling is explicit.
Run or report verification commands. If commands fail, include the exact failure category.

Review Rules

Lead with bugs, risks, missing tests, and spec violations.
Treat missing acceptance coverage as a blocker for new behavior.
Do not accept "works manually" as benchmark evidence.
Do not accept global user config mutation for hook installation.
Do not accept metrics without measurement_source, capture_source, and confidence.
Do not accept provider-specific data in domain models.
Do not accept Cursor/OpenCode/Pi implementation as v1 production scope without a new ADR.

Required Output

Use this structure:

Findings
Open questions or assumptions
Verification run
Change summary

If there are no findings, say that clearly and identify any residual risk or unrun verification.

Frequently asked questions