Personal Claude Code plugin accumulated over ~6 months of recreational use. Most of it is simple and likely part of hundreds of thousands of similar plugins; a few less common parts called out below.
Enable Claude Code to independently and quickly deliver high-quality code, so the user can spend as little time as possible.
For features and bug fixes alike, define desired outcomes clearly so Claude can (a) author tests verifying those outcomes and (b) iterate on implementation until verified.
Define → Test → Implement ordering is enforced in /design, /implement, and /bug. Tests are written against interfaces before implementation exists — this prevents bias toward verifying "how it was written" rather than "what it should do." Test cases are defined during design (Phase 2 of /design), carried into the implementation plan, and executed test-first by /implement.
Stash-based bug fix proof (/bug, /prep-commit --bugfix): stash the fix → run tests → they must fail (proving the bug is reproducible) → pop the stash → run tests → they must pass (proving the fix works). If the fail step doesn't fail, the test doesn't actually catch the bug — stop and revise.
Use agents to (a) reduce context usage of the main session and (b) limit confirmation bias.
Clean-slate reviews. Code reviews in /prep-commit and /prep-merge-pr run in separate agent sessions with no exposure to the implementation reasoning. The reviewer sees git diff, not the conversation that produced it.
Three-reviewer pattern (/prep-merge-pr): correctness, quality, and security — each reviewer goes deep on one concern instead of shallow on all three. They run in parallel (wall-clock time equals one review), produce independent perspectives, and findings are deduplicated before a single fixer agent addresses them in one pass.
Reduce wall-clock time by running independent workstreams simultaneously.
/prep-commitspawns up to 6 parallel agents: backend tests, frontend tests, backend quality, frontend quality, code reviewer, and Codex reviewer/prep-merge-prspawns up to 8 parallel agents: 2 test runners + 2 quality checkers + 3 specialized reviewers + Codex reviewer/implementspawns up to 3 parallel agents (DB, backend, frontend) for features with independent layers
Conditional skipping avoids wasted work: quality agents are skipped if /prep-commit already ran them with no code changes since; the security reviewer only spawns when changes touch auth, API, DB, or user input handling; test/quality agents are skipped entirely for unchanged sides.
Agents that spawn sub-agents need tool permissions pre-configured in .claude/settings.local.json — otherwise they block waiting for user approval. The permissions skill handles this: /setup bootstraps permissions during onboarding and preflight checks run before every agent-heavy command. /learn reflects on the full session — including scanning for patterns approved at runtime — and proposes improvements to permissions, conventions, and workflows.
| Command | Purpose |
|---|---|
/setup |
Onboard a project: configure permissions, bootstrap settings |
/brainstorm |
Explore problem space, challenge assumptions, recommend direction |
/design |
Architect layer-by-layer (storage → backend → frontend), define test cases |
/implement |
Execute plan — sequential or parallel agents, define → test → implement order |
/bug |
Investigate, write regression test, fix, prove fix works |
/prep-commit |
Parallel agents: scoped tests, quality checks, code review |
/commit |
Stage and commit per git conventions |
/prep-merge-pr |
Full test suites + 3 specialized reviews (correctness, quality, security) |
/pr |
Push branch, create PR via gh with structured body |
/merge |
Rebase on main, run /prep-merge-pr, merge, clean up |
/sync-main |
Pull latest main, delete merged local branches |
/push |
Push main to all configured remotes |
/learn |
Session reflection: review permissions, corrections, patterns; propose improvements |
| Skill | Purpose |
|---|---|
| coding-standards | Python/FastAPI and React/TypeScript patterns, testing standards |
| db-conventions | Migration-first workflow, RPC functions, Supabase patterns |
| git-conventions | Commit format, branch naming, PR structure |
| deployment | Local dev setup and Digital Ocean production deployment |
| permissions | Preflight checks before agents, post-session review of runtime approvals |
Clone this repository and add it as a Claude Code plugin:
claude mcp add-json ccupa '{"type":"local","path":"/path/to/ccupa"}'Or add the plugin path in your Claude Code settings.