35 tested, documented config files — production subagents, guardrail hooks, slash commands, MCP recipes, CLAUDE.md templates, and the workflow playbooks that make agentic coding cheap and reviewable.
Get the pack — $10./install.sh · works with any Claude Code install · plain Markdown/shell/JSON, nothing obfuscatedEvery file follows the same bar: it must do something a fresh Claude Code install doesn't, and it must have survived actual use — the hook scripts ship with their test cases.
code-reviewer · debugger · test-writer · security-auditor · refactorer · explorer · perf-profiler · dependency-auditor · commit-splitter · pr-describer · doc-writer. Each with the right tool scope and a disciplined system prompt — not "You are a helpful reviewer."
Block rm -rf / and curl|sh before they run · keep Claude out of .env and keys · block commits containing secrets · auto-format edits · run tests on stop · desktop ping when done · full command audit log.
/scope · /review · /tidy · /commit · /pr · /test · /explain — the scope→build→review→tidy→commit loop that produces small reviewable diffs instead of 2,000-line surprises.
GitHub, Postgres (with the read-only role setup), Playwright, Sentry, scoped filesystem — copy-paste configs plus honest "when NOT to use this" notes, because every server taxes your context.
Context management (why sessions get expensive and dumb) · reviewable diffs · TDD with agents · parallel git worktrees · subagent patterns. The distilled versions of a hundred scattered threads.
Next.js/TS · Python/FastAPI · Go · monorepo · a cross-tool AGENTS.md. Pre-structured so every line pays rent — commands, conventions, diff discipline, gotchas.
Two unedited excerpts from the pack — judge for yourself.
## Method (follow it in order) 1. Reproduce. Run the failing test and capture the exact error. If you can't reproduce it, say so — do not speculate. 2. Localize. Read the stack from the top of YOUR code. Find the exact line. 3. Form one hypothesis, stated as a falsifiable claim. 4. Test it cheaply before touching more code. 5. Fix the cause, minimally. No defensive try/except that swallows the real signal. 6. Verify. Re-run the failing case AND a nearby passing case. ## Anti-patterns - Changing code randomly to see if the error moves. - "Fixing" the test instead of the code.
# PreToolUse · Bash — when Claude runs `git commit`, # scan the STAGED diff for secrets first. # Exit 2 blocks the commit with the reason. staged="$(git diff --staged -U0 | grep '^+')" hits="$(... grep -nE '(AKIA[0-9A-Z]{16} |ghp_[A-Za-z0-9]{36}|github_pat_ |sk-[A-Za-z0-9_-]{20,}|xox[baprs]- |BEGIN .* PRIVATE KEY |(password|secret|api_?key)\s*[:=]...)')" if [ -n "$hits" ]; then echo "BLOCKED: staged diff contains secrets:" >&2 ... exit 2 # Claude sees why, and fixes it fi
⬇ Free sample — 3 unmodified files from the pack (an agent, a hook, a playbook — no email required)
54 files · instant download · use in personal & commercial work
free updates to the pack, forever
The pack uses stable, documented extension points: .claude/agents/, .claude/commands/, hooks in settings.json, and .mcp.json. Anything current works. If something breaks in a future version, email your receipt and you'll get the fixed pack.
Every file is human-readable Markdown, shell, or JSON — read it all before installing. The installer only copies files into .claude/ directories and asks before overwriting anything. Nothing phones home.
Fine — the license covers use across your own team's repos. Commit the installed files to your project's .claude/ so the whole team gets them.
If it's not useful to you, reply to your receipt email within 14 days and you'll get your money back. No forms.