A Claude Code plugin marketplace.
Enforce conventional commits, conventional comments, and engineering ownership for commits, PRs, and code reviews.
- Auto-invoking
commventionalskill activates on commit, PR, and review context without explicit commands - Sub-agent architecture: commit-crafter for diffs, review-formatter for feedback
- Three conventions: conventional commits, conventional comments, engineering ownership
- Three consumer-invoked skills for engineering-ownership wiring:
:strip-trailers(capability),:strip-pr-body(one-shot PR cleanup),:install-trailer-stripper(writes Claude Code or git-hook wirings into the consumer's surface on demand) - Reviews post as a single GitHub review submission with grouped inline comments at
path:line(v2.1) —review-formatteremits a locked JSON contract,bin/commventional-post-review.shis the deterministic poster - v2.0 ADR-006 conformance — no plugin-installed Claude Code hooks; trigger surface belongs to the consumer
Auto-invokes on: commits, pull requests, code reviews
Audit and optimize Claude Code configurations with dynamic best-practice research.
- Research-first architecture: subagents fetch official Anthropic docs before analysis
- Knowledge cache: research results cached at
~/.cache/claudit/with version-based + 7-day TTL invalidation - Knowledge skill:
/claudit:knowledgeexposes cached research to other plugins — auto-refreshes stale domains - Decision memory: stores audit decisions so future runs annotate recommendations with past context (team-shared, committable)
- Over-engineering detection as highest-weighted scoring category
- 6-category health scoring with interactive fix selection
- Persistent memory on research agents for faster subsequent runs
- Cross-tool synergy: cached knowledge speeds up skillet, smith, and hone
Commands: /claudit — run audit, /claudit:refresh — refresh cache, /claudit:status — show cache state, /claudit:knowledge — retrieve cached research
Build, audit, and improve Claude Code skills with research-first architecture and opinionated structure.
- Research-first: fetches latest Anthropic skill/agent docs before every action
- Claudit cache integration: uses claudit's cached ecosystem knowledge when available, falls back to own research
- Three workflows: build from scratch, audit existing, improve from findings
- Opinionated directory template enforcement for consistent skill structure
- 6-category scoring rubric specific to skill quality
Commands: /skillet:build <name>, /skillet:audit <path>, /skillet:improve <path>
Emit a structured JSON event for every Claude Code hook to a configurable transport.
- Registers all 26 documented hook events; each one is wrapped in a uniform envelope (
id,ts,type,host,session_id,pid,cwd,data) - Pluggable transport —
file:(default),fifo:, orhttp://— via~/.towncrier/config.jsonorTOWNCRIER_TRANSPORTenv var - Hard 2s timeout per emit with automatic fallback to the default file — Claude hooks never hang and events are never lost
- Strictly observational: pass-through
PermissionRequest, no stdout interference, no behavior changes skip_eventsconfig filter for muting noisy events without uninstalling- Producer only in v0.1.0 — write your own consumer against the documented envelope
Default output: ~/.towncrier/events.jsonl — tail -F and jq to start
Meta-orchestrator for Claude-Code-readiness — audits a repo against a rubric of readiness dimensions and delegates depth scoring to sibling plugins.
- Rubric-driven: scores Claude Code config, skills, commit hygiene, docs, lint posture, observability, AGENTS.md, and project records
- Delegation over re-implementation: folds
claudit,skillet,commventionalaudit output into a composite scorecard via a shared wire contract - Kernel scaffolds the minimum (AGENTS.md,
project/container,.pronto/tool state) and flags sibling-covered dimensions as "not configured" when a recommended plugin is missing - Roll-your-own references for every dimension — recommendations are registered, not required
- Machine-parseable
--jsonoutput alongside the human-readable markdown scorecard
Commands: /pronto:init, /pronto:audit, /pronto:status, /pronto:improve
The SDLC work layer — authors and maintains the records under project/ (plans, tickets, ADRs, pulse journal) and drives each record through its lifecycle.
- Three lifecycles: plans (draft → active → done), tickets (open → in-progress → closed), ADRs (proposed → accepted → superseded)
- Folder-as-primary — the folder a file lives in is the authoritative state; frontmatter
status:mirrors for machine-readability - Plan-scoped tickets — every ticket belongs to a plan; no standalone tickets
- Per-day pulse files (
project/pulse/YYYY-MM-DD.md) — append-only, merge-friendly - Declares pronto's
project-recordaudit dimension natively viaplugin.json— emits wire-contract JSON under--json - Templates ship portable;
/avanti:plan,/avanti:ticket,/avanti:adrcopy and fill into consumer repos
Commands: /avanti:plan, /avanti:ticket, /avanti:adr, /avanti:promote, /avanti:pulse, /avanti:status, /avanti:audit
Lint-posture toolkit for Claude Code consumer repos: configure, lint, format, fix. Pronto-sibling audit grades lint-posture against the rubric.
- Five skills:
/lintguini:configure(write rubric-aligned lint/format configs),/lintguini:lint(run linters, surface findings),/lintguini:format(run formatters,--checkfor diff-only),/lintguini:fix(auto-fix wrapper with--apply/--apply-semanticsplit),/lintguini:audit(Pronto-sibling lint-posture auditor) - Six first-class language paths: python (ruff), javascript / typescript (biome or eslint+prettier), rust (rustfmt + clippy), ruby (rubocop or standardrb), go (golangci-lint + gofmt)
- Per-language config templates derived from
plugins/pronto/references/roll-your-own/lint-posture.md— the rubric is the source of truth (ADR-008) - Self-describing configs — every generated config carries a
Generated by lintguiniprovenance comment pointing at the rubric - Wire-contract v2 envelope on
/lintguini:audit --json— five observations (four legacy + lint-pass-rate conditional on lintguini-managed config) - Declares
compatible_pronto: ">=0.2.0"per ADR-004 handshake
Commands: /lintguini:configure, /lintguini:lint, /lintguini:format, /lintguini:fix, /lintguini:audit
Documentation toolkit for Claude Code: write, search, query (with code corroboration), and tidy docs in a repo's docs/ tree. Pronto-sibling audit for code-documentation depth ships alongside.
- Five skills:
/inkwell:doc(Diátaxis-template scaffold/update),/inkwell:search(FTS5 overdocs/),/inkwell:query(RAG Q&A with citations + corroboration),/inkwell:tidy(drift-finder),/inkwell:audit(Pronto-sibling code-documentation auditor) - Diátaxis four-quadrant templates ship under
templates/—concept,how-to,reference,tutorial - FTS5 index at
docs/.inkwell.fts5.db(gitignored), rebuilt on-write bybin/inkwell-index.sh - Inference-time code corroboration — Tier 1 deterministic name-resolution, Tier 2 LLM-judged behavioural verification, Tier 3 annotated-only — architecture in ADR-007
- Audit stays deterministic: pure shell + grep + awk + jq, no LLM dispatch in scorers; conditional scorers (
score-template-compliance.sh,score-backlink-coverage.sh,score-duplicate-density.sh) gate on inkwell-frontmatter presence and emit empty-scope on non-inkwell consumers - Pronto sibling — depth auditor for the
code-documentationrubric dimension (weight 15); declarescompatible_pronto: ">=0.3.0"per ADR-004 handshake
Commands: /inkwell:doc, /inkwell:search, /inkwell:query, /inkwell:tidy, /inkwell:audit
Repo-level skills for plugin authors (not distributable plugins — these live in .claude/):
| Command | Purpose |
|---|---|
/smith <name> |
Scaffold a new plugin with correct structure and conventions |
/hone <name> |
Audit an existing plugin's quality (8-category scoring with interactive fixes) |
Both tools dispatch research agents to fetch the latest Anthropic plugin docs before operating, ensuring scaffolds and audits reflect the current spec.
If you're building a plugin or skill that needs Claude Code ecosystem knowledge (plugin specs, skill authoring, MCP, hooks), you can consume claudit's cached research instead of fetching docs yourself.
| Domain | Content |
|---|---|
ecosystem |
Plugin system, skills, agents, hooks, MCP servers |
core-config |
Settings, permissions, CLAUDE.md, memory system |
optimization |
Performance patterns, over-engineering detection |
In your skill's research phase, invoke the knowledge skill and fall back to your own research if claudit isn't installed:
### Step 1: Load Expert Context
Invoke `/claudit:knowledge ecosystem` to retrieve ecosystem knowledge.
**If the skill runs successfully** (outputs `=== CLAUDIT KNOWLEDGE: ecosystem ===` block):
- Use its output as Expert Context
- Also read your own domain-specific supplement for depth
- Skip research phase
**If the skill is not available** (claudit not installed — the invocation produces an error, is not recognized as a command, or produces no knowledge output):
- Fall back to your own research agentsThe knowledge skill checks cache freshness and auto-refreshes stale domains transparently. Your plugin doesn't need to understand the cache protocol — just invoke and use the output.
Output is wrapped in delimiters for easy identification:
=== CLAUDIT KNOWLEDGE: ecosystem ===
[cached research content]
=== END CLAUDIT KNOWLEDGE ===
Knowledge source: cache (fresh, fetched 2026-03-22) | Domains: ecosystem
Users can manually refresh with /claudit:refresh [domain|all] or check status with /claudit:status. The cache auto-refreshes on any /claudit or /claudit:knowledge invocation when stale.
Add quickstop as a plugin marketplace, then install:
/plugin marketplace add acostanzo/quickstop
/plugin install claudit@quickstopgit clone https://github.com/acostanzo/quickstop.git
claude --plugin-dir /path/to/quickstop/plugins/clauditSee the Claude Code plugin documentation for plugin authoring and marketplace details.
MIT