FeaturesSkills Catalog

Skills Catalog

This is the authoritative summary of templates/skills/INDEX.md: every skill, agent, and command ported into ai-core-kit, across both layers. For each item the INDEX records its layer, the source repo + license it was vendored or re-authored from, and the trigger that wires it into a fork (a project.manifest.yaml value, an archetype, or “always”).

For the complete capability list of every skill (what each does + its trigger, grouped by meta-builders / engineering / language packs / design-system / telemetry), see the Skills Reference. Companion reference pages: Agents, Commands, Hooks, MCP, and Observability.

Commands (e.g. /rpi/research) drive agents (code-explorer, architect) which invoke skills (coding-standards); the spine lives in two layers — META (.claude/skills/, builds the kit) and CHILD (templates/skills + agents + commands, rendered into a fork). The RPI loop research → plan → implement runs requirement-parserarchitectcode-reviewer / security-reviewer, gated research → plan by constitutional-validator.

The two layers — never conflate them

  • META = .claude/skills/ — skills that help build the kit. Not rendered into forks; never gated by a child contract.
  • CHILD = templates/skills/, templates/agents/, templates/commands/ — the payload /ack-init (P4) renders into a fork, gated by the child manifest.

License discipline. Apache-2.0 items (anthropics/skills) are copied/adapted with the NOTICE preserved; MIT items (ecc, claude-skills, claude-code-best-practice) are re-authored in kit style with attribution. Proprietary Anthropic doc skills (docx/pdf/pptx/xlsx) were never read, copied, or derived from. Full ledger: License ledger & references.

META — build the kit (not rendered into forks)

SkillSourceLicenseProvenanceTrigger
skill-creatoranthropics/skillsApache-2.0Vendored + adaptedMETA only — author/benchmark kit skills.
mcp-builderanthropics/skillsApache-2.0Vendored + adaptedMETA only — build/extend an MCP server (incl. a child’s features.mcp server).
skill-validatorauthor-new (wraps lint-frontmatter.py)Apache-2.0 (kit)OriginalMETA only — validate kit skills/agents/commands against frontmatter rules.
cost-telemetry (META)author-newkit (MIT)OriginalMETA only — interpret the offline aggregator over a kit build transcript.

skill-creator/agents/{analyzer,comparator,grader}.md are internal sub-agent prompts of the skill, not standalone kit agents — they carry no kit agent frontmatter and are not linted as agents.

CHILD — engineering & product skills

Top-level templates/skills/, archetype/manifest-gated.

SkillSourceLicenseTrigger
coding-standardsauthor-newMITAlways — cross-language quality floor.
error-handlingauthor-newMITAlways for code archetypes (backend-api, fullstack, cli, library).
code-tourauthor-newMITAlways — onboarding/PR/RCA walkthroughs; emits .tours/.
architecture-decision-recordsauthor-new (Nygard ADR)MITAlways — records to docs/adr/.
production-auditauthor-newMITbackend-api, fullstack, cli. Skip docs-only/library.
cost-auditauthor-newMITfeatures.cost_telemetry == true OR archetype runs paid jobs/agents.
cost-telemetry (CHILD)author-newMITfeatures.cost_telemetry == true. Runs telemetry/aggregate.py.
agent-evalauthor-newMITfeatures.agent_eval == true.
frontend-a11yauthor-newMITfullstack with React/Next UI (framework in [next, remix]).
ui-design-systemalirezarezvani/claude-skillsMIT (re-authored)fullstack / UI archetypes, or features.design_system == true.
saas-scaffolderalirezarezvani/claude-skillsMIT (re-authored)saas / fullstack with auth+billing.
spec-to-repoalirezarezvani/claude-skillsMIT (re-authored)Greenfield scaffolding / new-project flows.

The fullstack design-system ships three additional skills (shadcn-ui, brand-guidelines, frontend-design-guidelines) only when design_system.install: true — see Design system.

CHILD — reusable agents

templates/agents/. The RPI + review fleet. ECC supplies the engineering reviewers; claude-code-best-practice supplies the RPI-flow agents.

AgentSourceLicenseModelTrigger
architectaffaan-m/eccMITopusAlways — design/ADR/trade-offs; RPI plan phase.
code-exploreraffaan-m/eccMITsonnetAlways — discovery step of RPI.
code-revieweraffaan-m/eccMITsonnetAlways for code archetypes — pre-PR review.
security-revieweraffaan-m/eccMITsonnetCode archetypes touching auth/input/payments; pre-release.
silent-failure-hunteraffaan-m/eccMITsonnetCode archetypes doing I/O/DB/network/transactions.
refactor-cleaneraffaan-m/eccMITsonnetAlways for code archetypes — dedicated cleanup pass.
requirement-parserclaude-code-best-practice (rpi/)MITsonnetRPI research step 1; render when features.rpi == true.
constitutional-validatorclaude-code-best-practice (rpi/)MITopusRPI research→plan gate; render when the child has a constitution/archetype.

CHILD — RPI & product commands

templates/commands/.

CommandSourceLicenseTrigger
/rpi/researchclaude-code-best-practiceMITfeatures.rpi == true. RPI step 1 (GO/NO-GO).
/rpi/planclaude-code-best-practiceMITfeatures.rpi == true. RPI step 2 (planning docs).
/rpi/implementclaude-code-best-practiceMITfeatures.rpi == true. RPI step 3 (phased exec + gate).
/prdauthor-newMITProduct/SaaS archetypes or features.product == true.
/riceauthor-newMITProduct/SaaS archetypes or features.product == true.

CHILD — language / framework / DB packs

templates/skills/lang/, a deterministic manifest-gated render set. The authoritative trigger table is templates/skills/lang/INDEX.md.

PackSourceLicenseTrigger (manifest condition)
python-patternsaffaan-m/eccMITproject.language == python
python-testingaffaan-m/eccMITproject.language == python
typescript-patternsauthor-newMITproject.language == typescript
go-patternsaffaan-m/eccMITproject.language == go
rust-patternsaffaan-m/eccMITproject.language == rust
node-api-patternsauthor-new (informed by ECC nestjs)MITproject.framework in [express, nestjs]
react-patternsaffaan-m/eccMITproject.framework in [next, remix] (React fullstack only)
postgres-patternsaffaan-m/eccMITpersistence.db == postgres
prisma-patternsaffaan-m/eccMITpersistence.orm == prisma
docker-patternsaffaan-m/eccMITcontainerization — no manifest key yet (see gaps).

Deferred / not ported (no silent caps)

These were in the port plan or surfaced as gaps but are not present in the current ported set. Listed so nothing disappears silently — see the Roadmap for the full deferred / not-ported list.

Language / framework / DB enum values not yet ported:

Manifest enum valueDisposition
project.language == javajava-coding-standards (+ testing) routed outside the G4 minimum; future pack.
project.framework == fastapifastapi-patterns (copy from ecc) planned.
project.framework == gingin-patterns author-new, planned.
project.framework == axumaxum-patterns author-new, planned.
project.framework in [sveltekit, nuxt]sveltekit-patterns / nuxt4-patterns author-new. react-patterns must NOT render for these.
persistence.db in [mysql, sqlite, mongodb]mysql-patterns (ecc), sqlite/mongodb author-new.
persistence.orm in [sqlalchemy, drizzle, gorm]all three author-new in the port plan.

Schema/interview gap blocking clean docker-patterns wiring. templates/interview/questions.yaml never asks whether the child uses Docker, so docker-patterns has no enum to gate on. Recommendation: add a project.containerized boolean, OR always-render docker-patterns for backend-api/fullstack.

Anthropics example skills available but not vendored here: algorithmic-art, brand-guidelines, canvas-design, claude-api, frontend-design, internal-comms, slack-gif-creator, theme-factory, web-artifacts-builder, webapp-testing. (doc-coauthoring ships no LICENSE.txt → not vendorable.) Adopt later behind feature flags if needed.

Proprietary — permanently excluded. Anthropic document skills docx, pdf, pptx, xlsx (“All rights reserved”). Never read, copied, or derived from.