Skills Reference
This is the complete, no-omissions list of every skill in ai-core-kit. A
skill is a model-invoked capability: Claude loads it when its description
trigger matches the work at hand (a human can also invoke it by name). Skills
split along the META vs CHILD boundary:
- META skills live in
.claude/skills/and help build the kit itself. They are never rendered into a fork. - CHILD skills live under
templates/skills/and are rendered into a fork by/ack-init, gated by the project manifest.
For the source-repo + license provenance of each item see the Skills Catalog and the License ledger. The tables below are the capability reference: layer, what each does, and what fires it.
META skills (.claude/skills/) build the kit; CHILD skills (templates/skills/) are rendered into a fork and gated by project.manifest.yaml. cost-telemetry exists in both layers.
META — meta-builder skills (.claude/skills/)
Author, validate, and instrument the kit’s own primitives. Not rendered into forks; never gated by a child contract.
| Skill | Layer | What it does | Trigger / when |
|---|---|---|---|
cost-telemetry | META | Runs the offline ai-core-kit cost aggregator and interprets its output — computes USD spend for a Claude Code run from transcript token-usage lines times a versioned pricing map, attributed by model, feature, agent, and session. | How much a build/session/feature/agent cost, token or cost breakdown or report, mentions telemetry/pricing.json/aggregate.py. |
mcp-builder | META | Builds high-quality MCP (Model Context Protocol) servers that let LLMs reach external services through well-designed tools — covers research, implementation in Python (FastMCP) or Node/TypeScript (MCP SDK), and evaluation. | Build an MCP server, expose this API as MCP tools, FastMCP, wire up the project’s features.mcp server. |
skill-creator | META | Authors, edits, and benchmarks ai-core-kit skills — drafts a new SKILL.md, runs with-skill vs baseline evals, grades them, and optimizes the description for reliable triggering. | Create/author/write a skill, make a SKILL.md, improve/optimize this skill, run evals on a skill, benchmark a skill. |
skill-validator | META | Validates ai-core-kit skills, agents, and slash commands against the kit’s canonical frontmatter and structure rules by running scripts/lint-frontmatter.py and interpreting every finding. | Validate/lint/check this skill, does this SKILL.md follow the kit conventions, why is lint-frontmatter failing, audit the skills. |
cost-telemetryappears in both layers — the META copy interprets a kit build transcript; the CHILD copy (below) is rendered into a fork. Same engine, two homes.
CHILD — engineering & product skills (templates/skills/)
The payload skills agents load during delivery. Each row’s trigger is the
project.manifest.yaml condition (or archetype) that wires it into a fork.
| Skill | Layer | What it does | Trigger / when |
|---|---|---|---|
agent-eval | CHILD | Head-to-head, reproducible comparison of coding agents (Claude Code, Aider, Codex, and others) on this project’s own tasks — measuring pass rate, cost, wall-clock time, and consistency. | Which coding agent is best for us, benchmark these agents, did the new model regress. |
architecture-decision-records | CHILD | Captures architectural decisions as structured ADR documents that live in docs/adr/ alongside the code, recording context, alternatives considered, and rationale. | Record this decision, ADR this, choosing between significant alternatives, why did we choose X. |
code-tour | CHILD | Creates CodeTour .tour files — persona-targeted, step-by-step codebase walkthroughs anchored to real files and line ranges, written to .tours/. | Give me a code tour, onboarding walkthrough, tour this PR, walk me through how X works. |
coding-standards | CHILD | Baseline, cross-language coding conventions for naming, readability, immutability, type safety, and code-smell review — the shared quality floor for any module. | Clean this up, is this idiomatic, review for quality, what are our conventions. |
cost-audit | CHILD | Evidence-first investigation of runaway or anomalous spend in an app or service — tracing the request path to a ranked, file-cited root cause and a burn-ordered fix list. | Cost spike, burn rate, we’re over budget, duplicate jobs, free users hitting the paid model. |
cost-telemetry | CHILD | Runs this project’s offline cost aggregator and interprets its output — computes USD spend for a Claude Code run from transcript token-usage lines times a versioned pricing map. | How much a session/feature/agent/the project cost, token or cost breakdown, mentions telemetry/pricing.json/aggregate.py. Rendered when features.cost_telemetry == true. |
error-handling | CHILD | Robust error-handling contract for this project — typed error hierarchies, the Result pattern, API error envelopes, error boundaries, retries with backoff. | How should this fail, add retries, error envelope, this catch block swallows the error. |
frontend-a11y | CHILD | Accessibility patterns for React/Next.js UIs — semantic HTML, correct ARIA, form labeling and error association, keyboard navigation, focus management, reduced-motion, screen-reader support. | Make this accessible, a11y review, add aria labels, screen-reader support. |
production-audit | CHILD | Local-evidence production-readiness audit for a shipped app — pre-launch reviews, post-merge risk passes, and what-breaks-in-prod questions. | Is this production-ready, what would break in prod, ready to ship. |
saas-scaffolder | CHILD | Generates a production-shaped subscription-SaaS starter — authentication, a database schema, billing/checkout, protected routes, and a working dashboard. | Scaffold a new SaaS, scaffold a Next.js app with auth and payments, wire up Stripe billing. |
spec-to-repo | CHILD | Turns a natural-language project specification into a complete, runnable starter repository — parsing the spec, designing the file tree and schema, generating real code. | Build me an app, create a project from this spec, scaffold a new repo, turn this idea/PRD into code. |
ui-design-system | CHILD | Generates and exports a design-token system from a brand color — color scales, a modular type scale, an 8pt spacing grid, radii, shadows, breakpoints, z-index layers. | Generate design tokens, create a color palette / type scale / spacing system, export CSS variables or SCSS tokens. |
CHILD — language / framework / DB packs (templates/skills/lang/)
A deterministic, manifest-gated render set. Each pack is rendered only when its
manifest condition holds; the authoritative trigger table is
templates/skills/lang/INDEX.md.
| Skill | Layer | What it does | Trigger / when (manifest condition) |
|---|---|---|---|
docker-patterns | CHILD | Docker and Docker Compose patterns for this project — multi-stage Dockerfiles, dev/prod compose layering, healthchecks and service dependencies, networking, volume strategy. | Editing Dockerfile/compose.yaml or wiring local services. |
go-patterns | CHILD | Idiomatic Go conventions for this project — error wrapping, the zero-value rule, accept-interfaces-return-structs, context propagation, concurrency, and package layout. | Editing .go files when project.language == go. |
node-api-patterns | CHILD | Node.js backend API patterns for this project covering both Express and NestJS — layered structure, validated DTOs, centralized error handling, async-safe handlers. | Editing controllers/routes/services for Express or NestJS backends. |
postgres-patterns | CHILD | PostgreSQL patterns for this project — index selection, schema/data-type choices, query optimization, pagination, locking, and security defaults. | Editing migrations or .sql when persistence.db == postgres. |
prisma-patterns | CHILD | Prisma ORM patterns and non-obvious traps for this project — schema/ID strategy, select vs include, transactions, cursor pagination, PrismaClient singleton. | Editing schema.prisma or Prisma queries/migrations when persistence.orm == prisma. |
python-patterns | CHILD | Idiomatic Python conventions for this project — PEP 8 layout, type hints, dataclasses, comprehensions, context managers, exceptions, and packaging. | Editing .py modules when project.language == python. |
python-testing | CHILD | pytest testing strategy for this project — TDD cycle, fixtures and scopes, parametrization, mocking, async tests, and coverage gates. | Creating test_*.py when project.language == python. |
react-patterns | CHILD | React 18/19 patterns for this project — hooks discipline, server/client component boundaries, Suspense + error boundaries, form actions, data-fetching choices. | Editing .tsx/.jsx components for React 18/19. |
rust-patterns | CHILD | Idiomatic Rust conventions for this project — ownership and borrowing, Result/? error propagation, enums to make illegal states unrepresentable, traits, and safe concurrency. | Editing .rs files when project.language == rust. |
typescript-patterns | CHILD | Idiomatic TypeScript conventions for this project — strict compiler settings, type narrowing, discriminated unions, generics, runtime validation at boundaries. | Editing .ts/.tsx when project.language == typescript. |
Language enum values not yet ported (
java,fastapi,gin,axum,sveltekit/nuxt,mysql/sqlite/mongodb,sqlalchemy/drizzle/gorm) are tracked, not hidden — see the deferred list in the Skills Catalog and the Roadmap.
CHILD — design-system skills (fullstack)
The ui-design-system skill (above) is always available to UI archetypes. When
design_system.install: true the fullstack archetype additionally installs a
dedicated design-system payload — shadcn-ui, brand-guidelines, and
frontend-design-guidelines — plus the shadcn MCP server (see
MCP Reference and Design system).
| Skill | Layer | What it does | Trigger / when |
|---|---|---|---|
ui-design-system | CHILD | Design-token system generator (see engineering table). | fullstack / UI archetypes, or features.design_system == true. |
shadcn-ui + brand-guidelines + frontend-design-guidelines | CHILD | The fullstack design-system payload (component install via the shadcn MCP, brand tokens, frontend design guidelines). | Only when design_system.install: true. |
Telemetry skills
The two cost skills are CHILD payload (both MIT) and are detailed in the Observability Reference:
| Skill | Layer | What it does | Trigger / when |
|---|---|---|---|
cost-telemetry | CHILD | Runs telemetry/aggregate.py and interprets its output — the single source of truth for the numbers. | features.cost_telemetry == true. |
cost-audit | CHILD | Evidence-first investigation of why spend spiked; delegates the numbers to cost-telemetry. | Cost spike / burn-rate investigation. |
See also: Skills Catalog (provenance + license), Agents Reference, Commands Reference.