Implementation hub — TDD, debug, map, or architecture audit. Infers mode from context. Checks for spec before starting.
Control pakka compression. Switch output intensity (lite|strict|ultra|super-ultra), re-compress input files, restore originals. Default level is `super-ultra`.
Show pakka status — what's on, what you can run.
Design hub — spec, probe, challenge, or slice. Infers mode from context. Writes spec to docs/specs/.
Search pakka's audit trail across sessions via FTS5 full-text index.
Parallel reviewer for architecture drift — coupling, shallow abstractions, module bloat. Returns findings with confidence 0-100.
Parallel reviewer for performance regressions — N+1 queries, quadratic hot paths, allocation in tight loops. Returns findings with confidence 0-100.
Parallel reviewer for correctness and maintainability. Returns findings with confidence 0-100.
Security reviewer. Finds injection, auth bypass, secret leaks, crypto misuse. Returns findings with confidence 0-100.
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Uses Bash, Write, or Edit tools
Uses Bash, Write, or Edit tools
Claude Code harness — fewer tokens, fewer bugs, audit-ready. Apache-2.0.
/plugin marketplace add amargautam/pakka-marketplace
/plugin install pakka@pakka-marketplace
Zero-config. Uses your existing Claude Code auth. No API key required.
/plugin marketplace update
/plugin install pakka@pakka-marketplace
/reload-plugins
/plugin marketplace update must run first — it pulls the latest catalog ref. Without it, install resolves to a stale cached version.
Pakka injects discipline into every session. Commands infer what you need from context — no mode flags, no guessing.
| Command | Infers from | What it does |
|---|---|---|
/pakka:plan | "build X", "design", "challenge this", "probe me", "break into tickets" | Design hub. Writes spec to docs/specs/. Routes to spec · challenge · probe · slice based on context. Never auto-chains to build. |
/pakka:build | "implement", "write tests", "broken", "how does X work", "hard to test" | Implementation hub. Checks for spec approval first. Routes to TDD · debug · map · audit based on context. Blocks completion claims without exit-code evidence. |
/pakka:review | "done?", "ship?", "they said...", "merge?" | Quality hub. Verifies first (exit codes), then runs reviewer + security + architect + performance agents in parallel. Handles incoming feedback and branch landing. |
/pakka:triage | "triage", "look at issue #N", "what needs attention" | Issue queue. Routes bugs and features through classification state machine. Produces agent-ready briefs. |
/pakka:setup | one-time setup | Detects stack, writes permissions overlay. setup guard installs git guard hook. |
/pakka:compress | — | Compression control. [lite|strict|ultra|super-ultra|status]. Default: super-ultra. Hook-handled — instant, no LLM round-trip. |
/pakka:recall [query] | — | search audit trail across sessions via FTS5 index |
/pakka:help | — | Show pakka status — active level, gate config, hooks. |
Verification: before any "done", "working", "passing" claim — pakka requires actual exit-code evidence. Injected at session start.
Skill-check: before each response — pakka checks whether the message calls for /pakka:plan, /pakka:build, or /pakka:review. Catches cases that explicit invocation misses.
4-vector compression: output tokens · input context · tool results · subagent returns — all compressed independently.
Review gate: reviewer + security + architect + performance subagents run in parallel on every Claude-authored commit. Confidence threshold ≥ 80. Blocks on severity=error findings.
Deny-by-default permissions: secrets, destructive git, shell-fetched-then-executed commands blocked at the permission layer. Guard covers secret reads and writes — the hook matches Read|Write|Edit|MultiEdit|Bash (was Read|Bash before v0.11.0). Repeated overrides teach a per-repo allowlist (.pakka/guard-allowlist.json) with override-count decay; secret categories are never allowlistable.
Audit trail: every tool call appended to ~/.pakka/audit/<session>.jsonl. No dial-home.
recall: /pakka:recall searches your audit trail. cross-session memory backed by local FTS5 index (SQLite). no remote storage.
skill-check: UserPromptSubmit hook keyword-scans every message. if a build/plan/review signal matches, targeted alert fires before the model responds. no more relying on model memory.
Status line: pakka [super-ultra] · ~$72.47 saved · 21 bugs caught — compression level, token savings, and bugs caught, always visible. Savings priced per model — Fable 5, Mythos 5, and Opus 4.x covered; dated model IDs resolve by prefix.
Kill-switch: PAKKA_DISABLED=1 turns off every pakka hook for the session. Used by make bench to isolate the raw arm; works anywhere.
Benchmarks: make bench A/Bs pakka vs raw Claude Code via claude -p on your existing OAuth session. No API key.
Three absolute numbers, each verifiable from artifacts in this repo.
Bug catch rate: 9/10. Combined reviewer + security + architect agents caught 9 of 10 seeded bugs on the Pass 5b in-session corpus.
Bytes saved: 307,451 cumulative since 2026-04-24. Estimated tokens: 88,073 (bytes ÷ 3.5). Total estimated savings: ~$72.47. Source: RECEIPTS.md, regenerated via make self-report.
Gate enforcement: every Claude-authored commit path. Architectural claim — gate runs and blocks on findings. Verify: git log --format='%H' | while read sha; do git show -s --format='%(trailers:key=Reviewed-by-pakka,valueonly=true)' "$sha" | grep -q . && echo "$sha"; done | wc -l.
Every commit pakka reviews carries:
npx claudepluginhub amargautam/pakka-marketplace --plugin pakkaMulti-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Curate auto-memory, promote learnings to CLAUDE.md and rules, extract proven patterns into reusable skills.