Skip to content

About

My personal Claude Code plugin collection.

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

claude-code-plugins

My personal Claude Code plugin collection — five focused plugins in one marketplace, so you install only what you need.

Installation

# Add marketplace
claude plugin marketplace add cacack/claude-code-plugins

# Then install the plugins you want
claude plugin install delivery@cacack     # play → do → panel → ship cycle
claude plugin install panels@cacack       # repo-wide health panels
claude plugin install authoring@cacack    # create/audit Claude Code resources
claude plugin install principles@cacack   # the engineering canon installer
claude plugin install toolbox@cacack      # productivity utilities

Slash commands are namespaced by plugin: /delivery:ship, /panels:constitution, /principles:instill, and so on.

Plugins

delivery — the development workflow (play → do → panel → ship → merge)

The cycle runs inside a dedicated git worktree (.claude/worktrees/) so simultaneous parallel cycles never pollute each other — /delivery:play (or /delivery:do in direct mode) creates it; the rest inherit it; /delivery:merge tears it down.

Skills:

  • play - Plan the work: fetch a GitHub/GitLab issue (or take a free-text task), explore, design, approve via plan mode, then either execute inline or emit a DAG of execution prompts to .prompts/ for /delivery:do. Creates the cycle's worktree
  • do - Execute the work: run the latest /delivery:play batch (no args), run specific prompts by number, or execute a free-text task directly
  • ship - Ship the work: preflight checks, issue compliance verification, documentation review, and PR/MR creation (--quick for fast path)
  • merge - Land a green PR/MR with semi-linear history (rebase, then merge commit) and clean up the local clone: remove the cycle's worktree, return to and pull the default branch, delete the merged branch locally and remotely, and prune stale refs
  • deliver-milestone - Drive a whole milestone/epic to done across every open issue (implement → panel review → address findings → ship → optional CodeRabbit → merge). Routes by agency: a fully-autonomous run has Claude author and launch a built-in dynamic Workflow; a checkpointed run uses an interactive orchestrator that calls /delivery:play, /delivery:do, /delivery:panel-review, /delivery:ship with approval pauses
  • milestone-review - Audit a finished GitHub milestone from forge data alone: classify every issue by whether its acceptance criteria were answered with evidence, flag Refs-only deliveries and not-planned closures with no successor, harvest unverified review dismissals and added TODO/FIXME lines from its PRs, give a verdict, and draft issues for the gaps
  • panel-review - Multi-persona code review of a diff. Spawns 6 reviewer subagents (Skeptic, Maintainer, Performance Engineer, Caller, Security Reviewer, Tracer) in parallel against a branch, PR, or commit range. --deep raises investigation budgets (auto-suggested on provenance-heavy diffs); --standard suppresses that prompt
  • preflight-checks - Run project-defined code quality checks (make lint/test/security) before shipping
  • issue-compliance - Verify staged changes satisfy linked issue requirements with coverage scoring
  • security-review - Comprehensive security analysis of changes, context, or entire repository
  • whats-next - Discover and pick up work from handoffs, todos, GitHub/GitLab issues, or ideas (prioritized by readiness, with milestone-scoped issues ranked above general ones)
  • run-prompt - Delegate one or more prompts to fresh sub-task contexts with parallel or sequential execution
  • issue-delivery (internal) - Close the loop between a PR/MR and its tracked issue: closing vs. referencing keywords, partial-delivery bookkeeping, and deviation disclosure. Defers coverage scoring to issue-compliance

Agents:

  • shipper - Expert shipping orchestrator for rigorous code delivery (preflight, issue compliance, docs review)

Reviewer subagents (invoked in parallel by the panel-review skill):

  • reviewer-skeptic - Adversarial bug-hunter focused on edge cases and error-handling gaps
  • reviewer-maintainer - Reviews internal naming, structure, test adequacy, convention drift
  • reviewer-performance - Spots hot-path costs: complexity, allocations, lock contention, leaks
  • reviewer-ergonomics - Caller-perspective review of public APIs, contracts, error messages, breaking changes
  • reviewer-security - Diff-focused security review for injection, auth gaps, secrets, unsafe defaults
  • reviewer-tracer - Cross-file data-flow review: traces changed values, columns, FKs, and config keys to every writer and reader to find producer/consumer disagreement

panels — repo-wide health reviews

Skills:

  • constitution - Author or refresh a project's CONSTITUTION.md (mission, audience, principles, non-goals, success criteria). Auto-detects bootstrap vs refresh mode; in refresh mode produces a drift report before updating. Every success criterion carries a Check: — a command that settles it, or a declared judgement call — and refresh flags criteria that have neither. Required input for panel-product
  • panel-engineering - Multi-persona engineering-health review of the whole repo (quarterly). Spawns 5 senior personas (Architect, Security Posture, Operations/SRE, Developer Experience, Maintainability) in parallel against a captured snapshot, produces per-persona reports plus synthesis and proposed-issue drafts, optionally files the issues
  • panel-product - Multi-persona strategic-alignment review against CONSTITUTION.md (quarterly). First scores its Success Criteria into scorecard.md by running the checks the constitution names (asks before running them), then spawns 4 senior personas by default (Mission Steward, Roadmap Reviewer, Audience Advocate, Trust Auditor; Market Strategist is opt-in since panels 1.3.0 — pass --personas mission,market,roadmap,audience,trust to include it, since the list replaces the default) in parallel, then a closing adversarial Rude Q&A foil pass (rude-qa agent) pressure-tests the synthesis for survival; synthesis compares against the previous run, flagging gaps that persist across runs and scorecard regressions; produces per-persona reports plus synthesis, a foil report, and proposed-issue drafts, optionally files the issues. Requires CONSTITUTION.md — run constitution first if absent (--no-foil skips the foil pass)
  • pressure-test - Pressure-test a strategy, pitch, proposal, or roadmap against the adversarial questioning of the rude-qa agent before you bring it to decision-makers — 5 Whys, gap analysis, speed/cost/risk, pre-mortem, a hostile-question rehearsal, and a sharpened ask plus a Monday action

Both panel skills follow one shared run protocol — probe, snapshot, parallel spawn, truncation retry, synthesis rules, issue drafting, filing prompt — kept in docs/panel-protocol.md so a protocol fix lands in one file.

Engineering-panel subagents (invoked in parallel by the panel-engineering skill):

  • engineering-architect - Whole-repo architecture: module boundaries, coupling, layering, scalability shape
  • engineering-security - Whole-repo security posture: secrets handling, dependency hygiene, threat surface, SECURITY.md adequacy
  • engineering-ops-sre - Operability: observability, deployability, runbooks, failure modes, CI/CD health
  • engineering-dx - Developer experience: onboarding, build/test ergonomics, docs, contributor path
  • engineering-maintainability - Long-term carrying cost: test coverage patterns, convention drift, dead code, refactor debt

Product-panel subagents (invoked in parallel by the panel-product skill):

  • product-mission - What has been built: shipped work vs. stated mission, principle adherence, scope discipline
  • product-market - Market positioning: differentiation, competitive context, category fit, clarity of value proposition (scale-aware; opt-in — runs only when named in --personas)
  • product-roadmap - What is planned and ruled out: open issues/milestones vs. stated direction, priority alignment, sole owner of non-goal discipline
  • product-audience - Who it serves: audience-fit (sole owner), friction at the value moment, unmet needs, audience surprise (distinct from engineering-dx)
  • product-trust - Trust signals: promise vs. reality, transparency, expectation-setting, accountability signals

Strategy foil (standalone via the pressure-test skill, and the closing pass of panel-product):

  • rude-qa - Adversarial strategy sparring partner that runs a "Rude Q&A" over an initiative before it reaches decision-makers, ending with a sharpened ask and a Monday action

authoring — building and auditing Claude Code resources

Creation skills:

  • create-agent-skills - Comprehensive workflow for building agent skills with references, templates, and workflows
  • create-hooks - Build custom hooks with examples and troubleshooting guides
  • create-slash-commands - Build slash commands with argument handling and tool restrictions
  • create-subagents - Design and implement specialized subagents with orchestration patterns
  • create-claudemd - Create, author, or migrate CLAUDE.md and .claude/rules/ files following Anthropic best practices (create / rules / migrate modes). Pairs with audit-claudemd for the audit side
  • create-prompt - Expert prompt engineer that creates optimized, XML-structured prompts with intelligent depth selection
  • create-meta-prompts - Create optimized prompts for Claude-to-Claude pipelines with research, planning, and execution stages
  • create-plans - Create detailed project plans with milestones, phases, and checkpoints

Audit skills:

  • audit-skill - Audit and validate skill structure and quality
  • audit-slash-command - Audit and validate slash command implementation
  • audit-subagent - Audit and validate subagent definitions
  • audit-plugin - Audit plugin structure: directory layout, plugin.json/marketplace.json validity, version sync, resource integrity
  • audit-hooks - Audit hooks.json configuration for correctness, security, event types, matchers, and best practices
  • audit-prompt - Review prompt files for clarity, structure, and effectiveness
  • audit-claudemd - Audit a CLAUDE.md or .claude/rules/ file/dir for conciseness, stale references, scope, and path-scoping (drives the claudemd-auditor agent)
  • audit-docs - Audit project documentation for dead links, orphaned files, drift/staleness, and duplicated facts (DRY) — the runtime enforcement of the documentation-standards link-don't-duplicate discipline (drives the docs-auditor agent)

Migration skills:

  • graft - Graft a resource, a whole plugin, or a pattern from one home into another — within this marketplace or across repos — carrying its closure, rewriting every reference that must change, registering it in the target, and verifying nothing dangles. Handles a clone, a migration, and a re-graft of something grafted before

Standards skills:

  • issue-standards - The canonical tracked-issue standard (types, anatomy, evidence, labels, readiness) for GitHub and GitLab: what an issue body must carry, acceptance criteria as observable conditions, and the evidence each one owes before it closes. Ships copy-paste blocks and drop-in forge issue templates. Pairs with delivery:issue-delivery (PR-side linking) and delivery:issue-compliance (coverage scoring)
  • documentation-standards - The canonical project-documentation standard (types, locations, organization, formatting): a lean OSS-style root over a structured docs/ reference layer, with templates for each doc type. Pairs with audit-docs (drift/dead links), docs-analyzer (code-driven updates), and create-claudemd (CLAUDE.md authoring)
  • docs-analyzer - Semantic analysis of code changes to identify documentation that needs updating
  • heal-skill - Self-improvement workflow for skills

Agents: the auditor counterparts (skill-auditor, slash-command-auditor, subagent-auditor, plugin-auditor, hooks-auditor, prompt-auditor, claudemd-auditor, docs-auditor).

This plugin also owns the design docs behind the collection: the Handyman Principle and design guidelines.

principles — the canon

  • instill - Install or re-sync a canon profile into a repo's .claude/rules/, your user profile's always-on context, or the repo-root CLAUDE.md when a repo has no .claude/. Two profiles, one canon: engineering for someone shipping code (9 durable principles + 6 agent operating rules + privacy, issue-writing, and issue-delivery floors), universal for non-code work with Claude (the ideas that survive generalization, in wording that names no code). You get one per scope, never both — the skill audits the scope for overlap, migrates a pre-profile install in place, and writes a managed block you re-sync as the canon evolves. See docs/engineering-principles.md
  • privacy-redaction (internal) - Determine a destination's visibility, then redact local and internal specifics before they land in it — the procedure behind the privacy floor in the canon

toolbox — productivity utilities

  • add-to-todos - Add items to your todo list with context from conversation
  • check-todos - Review and manage your todo list
  • park - Park current session context or capture cross-project ideas for later pickup
  • history - Read Claude Code conversation history from ~/.claude/history.jsonl and present recent sessions (date, project, topic, session ID) as a scannable table, with a claude --resume tip
  • debug-like-expert - Deep analysis debugging mode for complex issues with methodical investigation protocols

Decision-making frameworks (consider/):

  • consider:10-10-10 - Evaluate decisions across three time horizons (10 minutes, 10 months, 10 years)
  • consider:5-whys - Root cause analysis by asking "why" five times
  • consider:eisenhower-matrix - Prioritize tasks by urgency and importance
  • consider:first-principles - Break down problems to fundamental truths
  • consider:inversion - Think backwards by considering what to avoid
  • consider:occams-razor - Favor simpler explanations and solutions
  • consider:one-thing - Identify the single most important action
  • consider:opportunity-cost - Evaluate what you give up by choosing something
  • consider:pareto - Apply 80/20 principle to find highest-leverage actions
  • consider:second-order - Analyze downstream consequences beyond immediate effects
  • consider:swot - Assess Strengths, Weaknesses, Opportunities, and Threats
  • consider:via-negativa - Improve by removing rather than adding

Domain expertise (expertise/):

  • expertise/iphone-apps - Comprehensive iPhone app development expertise (SwiftUI, App Store, testing, CI/CD)
  • expertise/macos-apps - Complete macOS app development knowledge (AppKit, document apps, system APIs)

Configuration

Personal Claude Code tooling config kept in config/ (not plugin resources):

  • ccstatusline/settings.json - ccstatusline status line layout, including the native weekly-usage widget that reads the rate-limit data Claude Code pipes to the status line. Run make install to symlink ~/.config/ccstatusline here.

Attribution

The majority of resources in this collection are adapted from taches-cc-resources by glittercowboy. This includes:

  • All decision-making frameworks (consider:*)
  • Prompt engineering workflows (create-prompt, run-prompt, create-meta-prompts)
  • Context management (whats-next, add-to-todos, check-todos)
  • Debugging tools (debug-like-expert skill)
  • Extension creation tools (all create-* and audit-* skills)
  • All agent definitions
  • All skills including meta skills and domain expertise

References

About

My personal Claude Code plugin collection.

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages