My personal Claude Code plugin collection — five focused plugins in one marketplace, so you install only what you need.
# 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 utilitiesSlash commands are namespaced by plugin: /delivery:ship, /panels:constitution, /principles:instill, and so on.
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 worktreedo- Execute the work: run the latest/delivery:playbatch (no args), run specific prompts by number, or execute a free-text task directlyship- Ship the work: preflight checks, issue compliance verification, documentation review, and PR/MR creation (--quickfor 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 refsdeliver-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:shipwith approval pausesmilestone-review- Audit a finished GitHub milestone from forge data alone: classify every issue by whether its acceptance criteria were answered with evidence, flagRefs-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 gapspanel-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.--deepraises investigation budgets (auto-suggested on provenance-heavy diffs);--standardsuppresses that promptpreflight-checks- Run project-defined code quality checks (make lint/test/security) before shippingissue-compliance- Verify staged changes satisfy linked issue requirements with coverage scoringsecurity-review- Comprehensive security analysis of changes, context, or entire repositorywhats-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 executionissue-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 toissue-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 gapsreviewer-maintainer- Reviews internal naming, structure, test adequacy, convention driftreviewer-performance- Spots hot-path costs: complexity, allocations, lock contention, leaksreviewer-ergonomics- Caller-perspective review of public APIs, contracts, error messages, breaking changesreviewer-security- Diff-focused security review for injection, auth gaps, secrets, unsafe defaultsreviewer-tracer- Cross-file data-flow review: traces changed values, columns, FKs, and config keys to every writer and reader to find producer/consumer disagreement
Skills:
constitution- Author or refresh a project'sCONSTITUTION.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 aCheck:— a command that settles it, or a declared judgement call — and refresh flags criteria that have neither. Required input forpanel-productpanel-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 issuespanel-product- Multi-persona strategic-alignment review againstCONSTITUTION.md(quarterly). First scores its Success Criteria intoscorecard.mdby 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,trustto include it, since the list replaces the default) in parallel, then a closing adversarial Rude Q&A foil pass (rude-qaagent) 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. RequiresCONSTITUTION.md— runconstitutionfirst if absent (--no-foilskips the foil pass)pressure-test- Pressure-test a strategy, pitch, proposal, or roadmap against the adversarial questioning of therude-qaagent 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 shapeengineering-security- Whole-repo security posture: secrets handling, dependency hygiene, threat surface, SECURITY.md adequacyengineering-ops-sre- Operability: observability, deployability, runbooks, failure modes, CI/CD healthengineering-dx- Developer experience: onboarding, build/test ergonomics, docs, contributor pathengineering-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 disciplineproduct-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 disciplineproduct-audience- Who it serves: audience-fit (sole owner), friction at the value moment, unmet needs, audience surprise (distinct fromengineering-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
Creation skills:
create-agent-skills- Comprehensive workflow for building agent skills with references, templates, and workflowscreate-hooks- Build custom hooks with examples and troubleshooting guidescreate-slash-commands- Build slash commands with argument handling and tool restrictionscreate-subagents- Design and implement specialized subagents with orchestration patternscreate-claudemd- Create, author, or migrateCLAUDE.mdand.claude/rules/files following Anthropic best practices (create / rules / migrate modes). Pairs withaudit-claudemdfor the audit sidecreate-prompt- Expert prompt engineer that creates optimized, XML-structured prompts with intelligent depth selectioncreate-meta-prompts- Create optimized prompts for Claude-to-Claude pipelines with research, planning, and execution stagescreate-plans- Create detailed project plans with milestones, phases, and checkpoints
Audit skills:
audit-skill- Audit and validate skill structure and qualityaudit-slash-command- Audit and validate slash command implementationaudit-subagent- Audit and validate subagent definitionsaudit-plugin- Audit plugin structure: directory layout, plugin.json/marketplace.json validity, version sync, resource integrityaudit-hooks- Audit hooks.json configuration for correctness, security, event types, matchers, and best practicesaudit-prompt- Review prompt files for clarity, structure, and effectivenessaudit-claudemd- Audit aCLAUDE.mdor.claude/rules/file/dir for conciseness, stale references, scope, and path-scoping (drives theclaudemd-auditoragent)audit-docs- Audit project documentation for dead links, orphaned files, drift/staleness, and duplicated facts (DRY) — the runtime enforcement of thedocumentation-standardslink-don't-duplicate discipline (drives thedocs-auditoragent)
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 withdelivery:issue-delivery(PR-side linking) anddelivery:issue-compliance(coverage scoring)documentation-standards- The canonical project-documentation standard (types, locations, organization, formatting): a lean OSS-style root over a structureddocs/reference layer, with templates for each doc type. Pairs withaudit-docs(drift/dead links),docs-analyzer(code-driven updates), andcreate-claudemd(CLAUDE.md authoring)docs-analyzer- Semantic analysis of code changes to identify documentation that needs updatingheal-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.
instill- Install or re-sync a canon profile into a repo's.claude/rules/, your user profile's always-on context, or the repo-rootCLAUDE.mdwhen a repo has no.claude/. Two profiles, one canon:engineeringfor someone shipping code (9 durable principles + 6 agent operating rules + privacy, issue-writing, and issue-delivery floors),universalfor 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.mdprivacy-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
add-to-todos- Add items to your todo list with context from conversationcheck-todos- Review and manage your todo listpark- Park current session context or capture cross-project ideas for later pickuphistory- Read Claude Code conversation history from~/.claude/history.jsonland present recent sessions (date, project, topic, session ID) as a scannable table, with aclaude --resumetipdebug-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 timesconsider:eisenhower-matrix- Prioritize tasks by urgency and importanceconsider:first-principles- Break down problems to fundamental truthsconsider:inversion- Think backwards by considering what to avoidconsider:occams-razor- Favor simpler explanations and solutionsconsider:one-thing- Identify the single most important actionconsider:opportunity-cost- Evaluate what you give up by choosing somethingconsider:pareto- Apply 80/20 principle to find highest-leverage actionsconsider:second-order- Analyze downstream consequences beyond immediate effectsconsider:swot- Assess Strengths, Weaknesses, Opportunities, and Threatsconsider: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)
Personal Claude Code tooling config kept in config/ (not plugin resources):
ccstatusline/settings.json- ccstatusline status line layout, including the nativeweekly-usagewidget that reads the rate-limit data Claude Code pipes to the status line. Runmake installto symlink~/.config/ccstatuslinehere.
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-expertskill) - Extension creation tools (all
create-*andaudit-*skills) - All agent definitions
- All skills including meta skills and domain expertise
- Plugin Marketplaces Documentation
- Claude Code Documentation
- taches-cc-resources - Source of many resources in this collection