claude

review

Two review skills in one plugin. review:deep is a two-phase convergence harness: iterative multi-reviewer fix and re-review until a round finds nothing actionable, then a multi-round cross-examination with an opposing model (Codex, else Gemini). review:adversarial is the single-pass version: Claude and the opposing model find issues independently, cross-examine each other, and only findings the other side confirms are reported. Both save an audit trail on the PR (or a local file with no PR) and degrade loudly to Claude-only when no adversary is available.

1 on GitHubv1.5.0MITupdated 7 days ago
/

Plugin review from marketplace claude-code-skills in abhattacherjee/claude-code-skills. Commands read from the repo's manifest — 6 days ago.

What it installs

Skills (2)

  • adversarial

    Runs an adversarial code review of a PR diff or working-tree diff between Claude and an opposing model (Codex when installed and logged in, else Gemini), surfacing only findings both models independently confirm (high-precision, both-confirm rule). Use when: (1) reviewing a PR or working-tree diff w

  • deep

    Use when the user wants a thorough, high-assurance review of code changes — phrases like \"review this until it's clean\", \"converge to zero issues\", \"adversarial review\", \"have Codex or Gemini and Claude review\", \"deep review this PR\", or \"make this change ironclad\". Runs TWO phases on a

Subagents (3)

  • bug-hunter

    Performs R1 adversarial bug-hunting on a code diff, grounded in actual source files. Finds bugs, security vulnerabilities, performance issues, and correctness defects. NOT user-invocable — spawned by the review:adversarial and review:deep skills.

  • convention-reviewer

    Performs R1 adversarial convention and maintainability review on a code diff, grounded in actual source files and project conventions (CLAUDE.md, linting rules, established patterns). NOT user-invocable — spawned by the review:adversarial and review:deep skills.

  • cross-examiner

    Performs symmetric cross-examination in the adversarial review: judges the adversary's (Codex or Gemini) independent findings against actual source, returning confirm/refute verdicts. NOT user-invocable — spawned by the review:adversarial and review:deep skills.

View source on GitHub

More from abhattacherjee/claude-code-skills

All 9 plugins →

Record and produce demo videos. demo-video:record turns a macOS screen recording into a narrated demo with AI-written zoom scripts and voiceover. demo-video:produce builds a narrated product video in code with Remotion, from real app screenshots, TTS voiceover and background music. Ships the nine agents they use.

2 skills · 9 agents

1

GitHub workflow skills in one install: create-board, triage-issues, plan-milestones, plan-week, move-card, promote-shipped and prune-branches, plus the four agents create-board dispatches. Per-user settings live in ~/.config/github-board/config.json.

7 skills · 4 agents

1

context

claude

Two skills for working with large content and past sessions, plus the agents they use. context:shield keeps the context window from overflowing by delegating token-heavy reads (URLs, Figma frames, wiki pages, code directories) to isolated sub-agents that return short summaries, and hands multi-batch work to ralph-loop. context:search finds past Claude Code conversations by topic, date, branch or project, shows them verbatim, and can summarize them.

2 skills · 2 agents

1

skill-kit

claude

Three skills for building and sharing Claude Code skills in one plugin. skill-kit:author writes and optimizes skills to Anthropic's best practices, with parallel sub-agents and script-first determinism. skill-kit:publish packages skills as plugins and syncs them to a GitHub monorepo. skill-kit:extract turns what a work session just taught you into a new or updated skill.

3 skills

1

spec

claude

Three skills for story specs in one plugin. spec:create writes a template-compliant story spec from a plan, requirements file, prompt or GitHub issue, with TDD implementation steps, success metrics, a Figma mockup gate for UI stories and vertical splitting for large stories. spec:review checks a spec against the real codebase and architecture and adds verified sub-tasks, an API test plan and design simplification notes. spec:implement builds a reviewed spec end-to-end: feature branch, sub-tasks, acceptance checks and a PR.

3 skills

1

Claude Code statusline in one install: install the 3-tier adaptive statusline, create your own from 20 composable items, or check context usage with context-bar. Never replaces a statusline script it did not write unless you pass --force, and backs up every file it replaces.

3 skills

1

Maintain review?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins9 plugins