mk-cc-resources

Design the sprint from requirements. Decides module boundaries, delegates per-module architecture to sub-architects in parallel, packs closed task specs into dependency-ordered sprints. Produces ARCH.md, decisions index, per-task specs, sprint manifest. Run after /research, before /build. After architect finishes, every remaining question is implementation, not design.

Cascade a version bump across plugin.json + marketplace.json plugin entry + mk-cc-all bundle (if affected) + marketplace.json metadata + RELEASE-NOTES.md entry. Accepts patch/minor/major bump type. Validates semver consistency across all touch points. Use when shipping changes to any plugin. Composable — @ship references this; /plugin-scaffold creates v1.0.0 directly (doesn't call this).

Generate session handoff document — what was done (with file/commit refs), what remains (priority-ordered), critical context (decisions, gotchas, rejected approaches), blockers. Saves a NEW timestamped handoff to .claude/handoffs/ each time (history is never overwritten), updates the .claude/handoffs/INDEX.md ledger, and refreshes .claude/handoff.md as the latest-alias /resume reads. Triggers /claude-md-sync if more than 10 files changed or impact-map sections touched. Use when ending a session, pausing mid-task, or switching projects.

Execute the sprint. Reads task specs from /architect, dispatches task agents in dependency-ordered waves, writes code + tests, verifies each agent's output against disk before recording completion. Drift surfaces loudly. Produces sprint completion records + SPRINT-REPORT.md. Run after /architect, before /review.

Turn project pitch into build-ready SPEC.md through interactive questions. Loops on open design questions until every section closes. Output is testable and unambiguous — no "TBD" or "agent decides X". First step of the pipeline. Run before /research.

Build a functionality glossary + DRY audit for a codebase. A deterministic Python engine indexes every function (Python/TS/JS/C# via AST, tree-sitter), fingerprints 5 signals, and clusters duplicate implementations; Claude sub-agents label functionalities against a controlled verb vocabulary, review each cluster (Pass B), and substrate-verify instances (Pass C). Produces GLOSSARY.yaml (frozen schema, consumed by future /dry-refactor) + GLOSSARY.md. Use when the codebase feels WET, before a refactor pass, or before /architect when a new module may overlap existing code.

Resume from previous session handoff. Reads .claude/handoff.md (the latest-alias), validates branch matches, checks for commits since handoff, compares pipeline phase, reports any state drift. Presents remaining work priority-ordered and recommends the single best first action. Marks the handoff consumed without destroying history — the permanent copies + ledger in .claude/handoffs/ are never truncated (migrates an older single-file handoff into that history). Use at session start.

Spec compliance audit before shipping. Walks SPEC.md top to bottom; for every design decision, reads code at the cited locator and verifies implementation matches. Distinct from /review (bug-hunt on sprint output) — verify is the final gate before complete. Run after /review passes, before marking project complete.

Code review on current sprint. Hunts bugs (does the code work?) and drift (does the code match the spec?). Every finding has file:line + verbatim quote, re-validated against disk. Confirmed critical findings block sprint advance. Run after /build, before /verify.

Pipeline self-heal. Two jobs — (1) recover from any state (fresh, mid-flight, corrupt, code-without-spec) by reading artifacts and proposing walk-forward via legal transitions; (2) sweep stale claims from outstanding work register and disposition each per user confirm. Applies only on user confirm — never silent mutations.

Route findings to the right phase. Reads items from /research, /review, or /verify (gaps, findings, drift) and categorizes each — one disposition per item. Routes to /elicit, /research, /architect, or /build depending on what the item needs. Categorizes, never resolves. Surfaces ambiguity to the user.

Propose targeted CLAUDE.md updates after code changes. Reads git diff (last 20 commits), cross-references against CLAUDE.md sections — impact map, shared modules table, file locations, architecture reference, known issues. Shows current-vs-proposed for each stale section. User approves each edit individually. Callable standalone or triggered by /handoff when staleness detected.

Diagnose this session — find multi-step workflow chains you did manually, friction with skills that fired badly or should have fired, installed plugins that fit but went unused, and clear coverage gaps. Reads conversation + git activity (session scope) or also handoffs + memory + recent commits (wide scope). Output is diagnostic — issues with evidence + root causes + where to look for fixes. Does NOT propose diffs or apply changes. Use when something felt off, when reviewing what worked, or periodically to find workflow gaps.

Metrics-driven retrospective — commits, files changed, lines +/-, tasks completed vs failed, drift items, blockers carried over. Reads git log, .pipeline/ sprint records, QA reports, handoff archives. Gaps and failures listed FIRST, then patterns, then what worked. Concrete recommendations for next session. Use after sprint completion or significant work session.

Show pipeline state. Three modes — init (create state file), status (show current phase + sprint), next (recommend next command). Reads .pipeline/state.yaml, validates against legal transitions, flags degraded states. Backs /init, /status, /next commands.

Code-mode DRY audit after /build. Runs the code-glossary engine on the sprint's code — indexes every function, clusters duplicate implementations across files, surfaces extraction candidates. Propose-only — writes .pipeline/glossary/GLOSSARY.{yaml,md}, never modifies source. Also renders MAP.md (functionality map) consumed by /architect and /build before designing. Optional phase; run after /build, before or alongside /review.

Spec-level DRY pass between /architect and /build. Clusters the current sprint's task specs across sub-architects, surfaces overlapping functionality ("task-042 and task-067 both describe fetch-user-from-db"), and proposes consolidations. Propose-with-confirm — every merge needs user OK; originals archived before any edit. Optional phase; run after /architect packs the sprint, before /build.

Research what could inform the best implementation. Multi-perspective parallel agents (best-practices, ecosystem, examples, risks, costs) synthesize into REQ.md with rationale, references, and testable acceptance criteria. Focused on the specific decisions the spec leaves open — never vague. Run after /elicit, before /architect.

Execute extractions a code-glossary run proposed — MVP preflight + dry-run only. Reads GLOSSARY.yaml (frozen schema v1), runs the 7 Appendix-A pre-flight gates (baseline tests, git-clean, target module, verification status, confidence floor, substrate-verify, gitignore), then prints the planned helper + per-site edit plan WITHOUT writing any source file. Use after /code-glossary marks clusters extractable and you want to see exactly what an extraction would change before committing to it. Live execution (writes, rollback, test-after-each) is a later version behind its own gate.

Cross-document drift audit. Compares CLAUDE.md + README.md + marketplace.json + bundle plugin.json against disk state (plugins/, versions). Finds stale references (benched plugins as active), version mismatches between docs and reality, plugins on disk missing from docs, outdated architecture descriptions. Proposes targeted edits per file for user approval. Broader scope than /claude-md-sync (which covers CLAUDE.md only). Use when shipping, after plugin/version changes, or when docs feel stale.

Bootstrap a new mk-cc-resources plugin with proper structure. Generates directory tree + plugin.json + SKILL.md skeletons for named skills + marketplace.json entry + mk-cc-all bundle update + README active-plugins table entry + dedicated README section + CLAUDE.md architecture tree entry + RELEASE-NOTES.md v1.0.0. Use when starting a new plugin from scratch. Mechanical 9-step chain — does it all in one invocation. Asks for plugin name + skill names + descriptions.

Audit a plugin's skill set against current best practices. For each skill, scores frontmatter quality, body length vs 500-line guide, conciseness, description discoverability, single-responsibility, and architecture coherence. Dispatches parallel review agents (one per skill). Produces per-skill scorecard with letter grades + systemic vs per-skill issues + ranked fixes. Diagnostic only — never applies changes. Use when reviewing a plugin's skill quality or before a major release.

Single-step skill fixture for T-979 AC-3 boundary test (D-Rd12-10).

Configure alert sounds — volume, mute/unmute, custom sounds per event

Tracks questions and bugs across a project. Auto-detects handlers, researches from project context in the background, logs to an Excel tracker, generates meeting agendas, and tracks reported bugs with investigation. Auto-gitignores itself. Use /note to ask a question, /note bug to log a bug, /note init to set up.

Explore the schema and values of any data file (XLSX, CSV, JSON) using the scout CLI. Use when the user asks to examine, index, or explore a data file's structure.