Agent Skills

A collection of skills for AI coding agents. Skills are packaged instructions that extend agent capabilities with multi-agent workflows.

Skills follow the Agent Skills format.

Available Skills

adversarial-review

Multi-agent adversarial code review that dispatches three specialist agents in parallel to find bugs, security vulnerabilities, and design problems. Findings are consolidated by severity into a single actionable report.

Use when:

• Reviewing code changes, diffs, or PRs
• Before merging significant changes
• When you want thorough, adversarial code analysis

Review agents:

• Agent A — Bugs & Logic: correctness, races, null handling, off-by-one, state management
• Agent B — Security & Edge Cases: injection, auth, input validation, DoS, cryptographic issues
• Agent C — Design & Performance: architecture, coupling, N+1 queries, API design, testability

verified-planning

Drafts implementation plans then dispatches two independent review agents to adversarially verify them before presenting to the user. No plan reaches the user without dual verification.

Use when:

• Planning any feature or multi-step task
• Before implementing significant changes
• When brainstorming transitions to planning

Review agents:

• Agent A — Adversarial Reviewer: finds logic flaws, edge cases, architectural risks, security concerns
• Agent B — Completeness Reviewer: verifies file accuracy, missing steps, dependency order, test coverage

Installation

npx skills add shrimalmadhur/agent-skills

Local Development

From the repo root, sync real skill directories into Codex so they survive restarts and are discoverable as normal Codex skills:

make sync-codex-skills

That target calls scripts/sync_codex_skills.sh, which copies every skill directory containing a SKILL.md from skills/ into ~/.codex/skills.

You can also run the script directly:

bash scripts/sync_codex_skills.sh ./skills ~/.codex/skills

If you still use Claude-compatible symlinks, keep using:

make link

make status shows both the Claude symlink state and whether each skill has been synced into Codex.

Usage

Skills are automatically available once installed. The agent will use them when relevant tasks are detected.

Examples:

Review this PR for issues

Plan the implementation for adding user authentication

Skill Structure

Each skill contains:

• SKILL.md - Instructions for the agent

License

MIT