Fifty-nine skills,
all in main context.
Reference for the skills. A skill runs in main context with the whole conversation visible, which is what lets it decide anything. The baseline ships one subagent, and everything else that might have been one is a skill instead.
Categories
If these counts and the real total ever drift apart, audit-baseline fails the build. There are 15 categories, assigned by hand rather than derived from disk. That is why they live beside the deriver, and why this section can give you counts without a per-skill grouping.
4
10
5
5
3
1
1
1
4
10
1
2
4
5
3
The roster
59 baseline-owned skills, in alphabetical order. Each summary is excerpted from the first sentence of the skill's own description. It is never reworded, because a skill prompt is a contract and editing it from a template would quietly change how the skill behaves in a session (the roster derives from the files themselves).
move the slug's workflow artifacts (intake, scout, research, spec, approvals, swarm state, security reports, rendered diagrams) to…
Drift check between the baseline implementation on disk and the claims in `docs/init/seed.md` + cross-references in CLAUDE.md, README.md, and the…
PM-mode brainstorm helper. Captures the requirement via Socratic dialogue before any entry phase (`/intake`, `/spec`, `/tdd`) drafts its artifact.
Draft a Business Requirements Document (BRD) for cross-functional or stakeholder-heavy work that needs more structure than an intake.
documentation edits, governance count bumps, vendored-skill content updates, configuration tweaks, formatting, typo fixes, dependency bumps where no…
Analyze a codebase and recommend Claude Code automations (hooks, subagents, skills, plugins, MCP servers).
The primary tool for code-navigation questions in ANY language (frontend or backend) — walk the structural graph from an entry point (page, route,…
MANDATORY skill for ALL code generation.
Commit Preparation and Execution. Stages and commits the work.
Split a dirty working tree into single-concern Conventional Commits.
EXPERIMENTAL. Join or leave an org-team coordination channel as a session peer.
When the user wants to write, rewrite, or improve marketing copy for any page — including homepage, landing pages, pricing pages, feature pages,…
Orchestrates `impeccable` for every design task inside a workflow phase.
orchestrator for documentation work. Surveys the diff, routes any page on a documentation surface through `technical-writer`, standalone technical…
Write and maintain technical documentation.
Generate or repair a project's .gitignore.
End-to-end workflow orchestrator. Walks the 11-phase pipeline, invoking each phase skill in order inside an internal loop, yielding at consent gates…
Remove signs of AI-generated writing from text.
Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract,…
Write the production code that makes a pre-decided set of failing tests pass, within an explicit write_set, following an explicit behavior contract.
Draft a Workflow Phase 1 intake document capturing a new request — the problem, the desired outcome, constraints, and testable acceptance criteria.
Integration and Verification. Runs the full test suite, stamps the binding verdict at .claude/state/last_test_result, optionally runs a cross-engine…
Review the auto-extracted candidates in `.claude/memory/_pending.md` and commit keepers to the canonical memory files (`landmarks.md`,…
a flat pod of up to four peer Claude Code sessions claims lane-tagged slices of an approved spec over the baseline MCP channel and implements them…
hosts the two behaviours that distinguish the `power` workflow track from spec-entry.
documentation body, intake problem statements, spec context, RCA summaries, marketing copy, README sections, PR descriptions.
Draft a Root Cause Analysis for an incident, outage, or repeated test failure.
Bring documentation prose down to a high-school reading level without losing precision.
Research and Solution Exploration. Surfaces 2–4 candidate solution approaches with concrete tradeoffs, grounded in current library docs (fetched…
Cycle-end retrospective converting recurring judgment failures into durable enforcement.
Read-only view of the execution roadmap.
Derive an execution roadmap from a project's vision/spec docs by first principles, then diff that fresh derivation against the existing roadmap to…
sync the project's execution roadmap (project.json → roadmap.path, default docs/roadmap-execution-plan.md) to the just-landed work.
Write executable failing tests from a recipe handed to you by the main context.
Codebase Scouting and Constraint Discovery.
OWASP-aligned security review of pending code changes.
Mechanical cleanup pass over the branch diff, followed by a `code-structure` review pass and a `verify` re-stamp.
Draft a Workflow Phase 4 technical spec from an intake (and optionally a BRD + scout + research memo).
Cross-consistency review of a drafted spec's diagrams.
Preflight a spec draft without saving. Runs the same checks as the write-boundary hooks — PlantUML syntax, required diagram presence, AC-to-sequence…
Extract every PlantUML block from docs/specs/<slug>.md and render each to SVG under docs/specs/_rendered/<slug>/, with an index.md listing them in…
Oracle-bound spec-review check that every structured Rollout prerequisite binds to an enforcement-type acceptance criterion.
Bootstrap the central system spec at `docs/system/` for a repository that has never had one.
every spec AC must trace to a resolvable upstream AC in the intake (and BRD if present), and no upstream AC is silently dropped.
a mechanical, exit-code-driven check that every feature in a sprint manifest is provably complete across three dimensions (a done-record reference,…
Decompose an MVP vision into a sprint manifest — a prioritized feature list where every feature carries explicit done-criteria (a done-record…
Propose the next dependency-ready sprint from the ALREADY-DECOMPOSED roadmap — standup's active sibling.
Read-only release + backlog recap. Reports the last release, commits-since-tag classified by conventional-commit type with the semver bump they…
Execute a swarm plan wave by wave with filesystem isolation via git worktrees.
Decompose an approved spec into a dependency-ordered swarm plan — one task per component, each with an explicit write_set.
Report the health of the central system spec at `docs/system/` across seven checks — coverage gaps, stale elements, dangling anchors, duplicate…
TDD coordinator. Decides the scenario recipe and the implementation contract in main context, writes them to a state file, seeds per-worker tasks…
When the user wants to create step-by-step technical tutorials, quickstarts, or code walkthroughs.
Standard operating procedure for producing a documentation page end to end — gather source material, classify, draft against the measured corpus…
Write technical documentation that matches the measured profile of professionally-written docs.
pick the workflow entry phase (intake / spec / tdd / chore) and record the workflow statefile that the Track Guard reads.
Reconcile baseline-versioned files that the `create-baseline upgrade` CLI staged for LLM-assisted semantic merge.
Contract document for the binding test verdict file at .claude/state/last_test_result.
On-demand "what's new" generator. Main context (which knows the impending change) writes keepachangelog-style entries to a fragment file at…
Ownership
A skill is baseline-owned when its SKILL.md frontmatter says owner: baseline. Every other skill on disk belongs to the project. Leaving the line out is the default on purpose, so a repository that already carries skills of its own can install the overlay without anyone annotating a single file (an explicit owner: user is permitted, never required).
To keep a maintainer's working skills out of the package, the build ships a skill only when it claims baseline ownership. Ownership decides one further thing. audit-baseline hash-checks the baseline-owned paths and nothing else, which leaves your own skills outside the audit.
Sub-skill contracts
Five skills call another skill as part of their job, and four of those calls are mandatory. Where a skill declares none, its work is mechanical enough not to need one.
Calls code-structure.
Calls code-structure. Also checks current docs for any third-party API.
Calls impeccable to do the design. Writes no product code itself.
Always calls humanizer. Also calls one register skill, picked by whoever ran it.
Mechanical. Declares no sub-skill.