Home / Claude Code / Subagents & Skills
Subagents give Claude Code focused specialists with their own instructions, tool limits and context window: a reviewer that can only read, a test writer that can run tests, a security auditor that never edits. Skills give it reusable know-how packaged with scripts and reference files that load only when relevant. This pack ships both, ready to copy.
The 12 subagents - code reviewer, test writer, debugger, security auditor, docs writer, refactorer, performance profiler, migration planner, dependency auditor, accessibility auditor, CI triager and codebase explorer - use the real .claude/agents/*.md format with name, description (written so Claude delegates at the right moments), least-privilege tools, and a tight system prompt with an output contract. The 6 skills use SKILL.md folders: repo health check, changelog from git, ADR writer, test-failure triage (JUnit), TODO triage and a PR review checklist. Every script is dependency-free Python, cross-platform, and was run before shipping.
Includes a short guide on choosing between subagents, skills and slash commands, and on tuning descriptions so delegation works. Claude Code evolves; items note what to verify in the current docs.
Highlights
- 12 subagents with least-privilege tool lists and output contracts
- 6 skills as SKILL.md folders with working, tested Python scripts
- Descriptions written to trigger delegation correctly
- Read-only agents for review/audit; editing agents only where needed
- Cross-platform scripts, standard library only
- CC0: edit, rename, and ship in your own team kit
mammoth templates


