Skip to content

CATALOG.md

Latest commit

 

History

History
375 lines (297 loc) · 23.1 KB

CATALOG.md

File metadata and controls

375 lines (297 loc) · 23.1 KB
Raw

First-Party Skill Catalog

中文目录

This catalog is synchronized from the 34 admitted package directories under skills/. It is an inventory — not a static workflow router and not a record of what is installed on a particular Agent host. The package's SKILL.md remains the behavior authority.

Collection status

Field Value
Collection Light Skills — Composable Agent Workflows
Package count 34 admitted first-party Skills
Current state Released v0.2.0 with 34 first-party packages
Stable release v0.2.0 (34 packages; previous stable was v0.1.6 with 9 packages)
Installation authority docs/INSTALLATION.md
Discovery check tests/test_collection_discovery.py · tests/test_composition.py
Evidence v0.2.0 release evidence

v0.1.1 shipped five packages; v0.1.2 added recap and language-learning (seven); v0.1.3 migrated the test toolchain; v0.1.4 added kanban-worker; v0.1.5 tightened kanban scheduling and identity; v0.1.6 added kb-init (nine). v0.2.0 released the full 33-package architecture across project workflow, clarification, execution, review, and specialized tools, and the v0.2.0 line was extended with the humanizer admission (34 packages; see CHANGELOG.md).

No package in this table is an unmodified upstream copy. Approved Matt PORTs carry ATTRIBUTION.md and no upstream runtime dependency.

Admitted Skills

agent-config

  • Purpose: Map inspectable Agent Host evidence (models, agents, parallelism, worktree) to one safe execution plan with role-clear fallbacks.
  • When to use: Complex work where reviewer independence, isolation, or parallelism affects structure.
  • Invocation: Model-invoked.
  • Package: skills/agent-config/
  • Status: Admitted first-party; NEW architecture (Sol Advisor design reference, host-agnostic).
  • Evidence: references/host-evidence-schema.md, plan-schema.md.
  • Installation path: skills/agent-config/ in a host-recognized Skills root.

ask-light

  • Purpose: Act as the Light workflow advisor, navigator, and router: inspect project/workflow state, recommend the next Skill with reasoning, and transition safely after user approval.
  • When to use: The next step is unclear; you need project-aware routing, collection navigation, or standalone routing.
  • Invocation: User-invoked only; read-only before approval. After approval, model-invoked targets may begin where supported; user-invoked targets follow Host transition policy and render the exact invocation when direct transition is unavailable.
  • Package: skills/ask-light/
  • Status: Admitted first-party; REFACTOR (router built last after the full map).
  • Evidence: Semantic-map, provenance, availability, and top-result tests under skills/ask-light/tests/; user guide.
  • Installation path: skills/ask-light/ in a host-recognized Skills root.

clarify

  • Purpose: One-invocation continuous clarification for a vague idea, requirement, or process — no formal SPEC or project workflow. Asks the current frontier as a round of independent questions and accepts batch replies.
  • When to use: Idea/brainstorm is foggy and no project context is required.
  • Invocation: User-invoked only.
  • Package: skills/clarify/
  • Status: Admitted first-party; ADAPT (Matt grill-me → Light, via socratic).
  • Evidence: SKILL.md, references/WORKFLOW.md, ATTRIBUTION.md.
  • Installation path: skills/clarify/ in a host-recognized Skills root.

code-review

  • Purpose: Read-only specialist reviewer for a bounded git diff along Standards and Spec axes.
  • When to use: Reviewing a branch/PR diff, or when review-loop / project-review requests a software check.
  • Invocation: Model-invoked (read-only; never repairs or verdicts).
  • Package: skills/code-review/
  • Status: Admitted first-party; ADAPT (Matt code-review, two-axis method preserved).
  • Evidence: references/WORKFLOW.md, SMELL-BASELINE.md, ATTRIBUTION.md.
  • Installation path: skills/code-review/ in a host-recognized Skills root.

decision-map

  • Purpose: Plan a large, foggy, multi-session effort as a persistent decision map of tickets.
  • When to use: Many dependent decisions; work spans sessions; fog must clear before project-spec.
  • Invocation: User-invoked only.
  • Package: skills/decision-map/
  • Status: Admitted first-party; ADAPT (Matt wayfinder).
  • Evidence: references/MAP-CONTRACT.md, ATTRIBUTION.md.
  • Installation path: skills/decision-map/ in a host-recognized Skills root.

diagnosing-bugs

  • Purpose: Diagnosis loop for hard bugs and performance regressions with a tight feedback signal.
  • When to use: Something broken/throwing/failing/slow and the cause is not obvious.
  • Invocation: Model-invoked.
  • Package: skills/diagnosing-bugs/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt baseline preserved).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/diagnosing-bugs/ in a host-recognized Skills root.

eli5

  • Purpose: Explain any topic, code, or error at a chosen audience level.
  • When to use: User asks "explain like I'm 5", "to my boss", or needs a non-technical framing.
  • Invocation: Model-invoked (explain on explicit request).
  • Package: skills/eli5/
  • Status: Admitted first-party; MIGRATE — NO REWRITE (from upstream DreambigOu/ELI5 @ a766623, via temporary migration fork LightDevCoder/ELI5).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/eli5/ in a host-recognized Skills root.

generic-review

  • Purpose: Read-only default reviewer for ordinary artifacts — finds omissions, wrong output, contradictions, usability gaps.
  • When to use: No specialist reviewer is more appropriate.
  • Invocation: Model-invoked (read-only; never verdicts).
  • Package: skills/generic-review/
  • Status: Admitted first-party; NEW.
  • Evidence: SKILL.md.
  • Installation path: skills/generic-review/ in a host-recognized Skills root.

handoff

  • Purpose: Compact the current conversation into a handoff document for the next agent.
  • When to use: Closeout or resumption across sessions/agents.
  • Invocation: User-invoked only.
  • Package: skills/handoff/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt handoff).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/handoff/ in a host-recognized Skills root.

humanizer

  • Purpose: Rewrite AI-sounding English or Chinese text so it reads naturally without changing what it says.
  • When to use: Editing or reviewing prose with inflated claims, sales language, vague sources, stock AI words, filler, or chatbot artifacts.
  • Invocation: Model-invoked.
  • Package: skills/humanizer/
  • Status: Admitted first-party via full-path PASS; released on the v0.2.0 line; ADAPT of blader/humanizer (2.11.2) plus a thin zh adaptation, MIT attribution preserved.
  • Evidence: SKILL.md, ATTRIBUTION.md, admission record.
  • Installation path: skills/humanizer/ in a host-recognized Skills root.

implement

  • Purpose: Execute one bounded, already-decided work item (code, doc, config, Skill, generic task) with verification and review handoff.
  • When to use: A ticket or SPEC slice is ready and unambiguous.
  • Invocation: User-invoked only.
  • Package: skills/implement/
  • Status: Admitted first-party; ADAPT (Matt implement → general-purpose executor).
  • Evidence: references/WORKFLOW.md, ATTRIBUTION.md.
  • Installation path: skills/implement/ in a host-recognized Skills root.

kanban-worker

  • Purpose: Pick up and execute one Light-Kanban task per scheduled run; resumes owned work and reviewFeedback before new claims.
  • When to use: Scheduled Light-Kanban board work.
  • Invocation: Model-invoked; manual entry point is supported.
  • Package: skills/kanban-worker/
  • Status: Admitted first-party via full path (review-loop agent-skill PASS); renamed from light-kanban-worker in v0.1.6.
  • Evidence: Contract and behavior tests under skills/kanban-worker/tests/, user guide.
  • Installation path: skills/kanban-worker/ in a host-recognized Skills root.

kb-init

  • Purpose: Design and initialize a maintainable knowledge base via interview, then implement only after approval.
  • When to use: Creating or rebuilding a wiki, reference library, or research archive.
  • Invocation: User-invoked only.
  • Package: skills/kb-init/
  • Status: Admitted first-party via full path (review-loop agent-skill PASS); released in v0.1.6.
  • Evidence: Contract tests under skills/kb-init/tests/, user guide.
  • Installation path: skills/kb-init/ in a host-recognized Skills root.

language-learning

  • Purpose: Tutor any language through six modes — lessons, flashcards, conversation, grammar, quizzes, translation.
  • When to use: Learning or practicing a foreign language.
  • Invocation: User-invoked only.
  • Package: skills/language-learning/
  • Status: Admitted first-party via prompt-only fast-track PASS; released in v0.1.2.
  • Evidence: Contract tests under skills/language-learning/tests/, user guide.
  • Installation path: skills/language-learning/ in a host-recognized Skills root.

learn-anything

  • Purpose: Turn sufficiently evidenced conversations, notes, or workflows into reusable Agent Skill methods.
  • When to use: Source material may contain a repeatable, evidence-backed method.
  • Invocation: User-invoked only.
  • Package: skills/learn-anything/
  • Status: Admitted first-party; PRESERVE — NO REWRITE.
  • Evidence: package contract, user guide.
  • Installation path: skills/learn-anything/ in a host-recognized Skills root.

manuscript-ops

  • Purpose: Route and govern manuscript engineering from notes to multilingual, multi-format deliverables.
  • When to use: Manuscript scope, risk, batches, reviews, or formats need governing.
  • Invocation: Model-invoked; manual entry point is supported.
  • Package: skills/manuscript-ops/
  • Status: Admitted first-party; PRESERVE — NO REWRITE.
  • Evidence: package contract, user guide.
  • Installation path: skills/manuscript-ops/ in a host-recognized Skills root.

project-clarify

  • Purpose: Clarify a real project's unresolved decisions from inspected project facts; returns a bounded handoff for project-spec. Uses the same frontier-round interaction as clarify with project-aware evidence.
  • When to use: Existing project has unclear requirements; facts already in repo should not be re-asked.
  • Invocation: User-invoked only.
  • Package: skills/project-clarify/
  • Status: Admitted first-party; ADAPT (Matt grill-with-docs).
  • Evidence: references/project-clarification-contract.md, ATTRIBUTION.md.
  • Installation path: skills/project-clarify/ in a host-recognized Skills root.

project-init

  • Purpose: Idempotently bootstrap the stable Light project and tracker contracts consumed by downstream Project Skills.
  • When to use: New project needs a minimal, confirmed starting point.
  • Invocation: User-invoked only.
  • Package: skills/project-init/
  • Status: Admitted first-party; REFACTOR (repository bootstrap; full clarification remains in project-clarify).
  • Evidence: Contract and behavior tests under skills/project-init/tests/; user guide.
  • Installation path: skills/project-init/ in a host-recognized Skills root.

project-review

  • Purpose: Project-level final acceptance — freeze a baseline, compose reviewers, issue PASS/FAIL/BLOCKED.
  • When to use: Completed project needs acceptance before release-workflow.
  • Invocation: Model-invoked; manual entry supported.
  • Package: skills/project-review/
  • Status: Admitted first-party; NEW (migrated final-acceptance logic from old review-loop).
  • Evidence: SKILL.md, references/profiles/.
  • Installation path: skills/project-review/ in a host-recognized Skills root.

project-spec

  • Purpose: Turn already-clarified outputs into a formal project SPEC without reopening an interview.
  • When to use: Decisions are clarified and a SPEC is needed for project-tickets.
  • Invocation: User-invoked only.
  • Package: skills/project-spec/
  • Status: Admitted first-party; ADAPT (Matt to-spec).
  • Evidence: references/, ATTRIBUTION.md.
  • Installation path: skills/project-spec/ in a host-recognized Skills root.

project-tickets

  • Purpose: Turn an approved SPEC into a dependency-ordered, tracer-bullet ticket graph.
  • When to use: SPEC is approved and executable tasks are needed.
  • Invocation: User-invoked only.
  • Package: skills/project-tickets/
  • Status: Admitted first-party; ADAPT (Matt to-tickets).
  • Evidence: references/, ATTRIBUTION.md.
  • Installation path: skills/project-tickets/ in a host-recognized Skills root.

prototype

  • Purpose: Build a throwaway prototype to answer a design question.
  • When to use: State model or UI logic needs a quick feel-check before commitment.
  • Invocation: Model-invoked.
  • Package: skills/prototype/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt prototype).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/prototype/ in a host-recognized Skills root.

recap

  • Purpose: Show one concise line about the current session without replacing or compacting conversation history.
  • When to use: User explicitly invokes $recap.
  • Invocation: User-invoked only; $recap is the sole entry.
  • Package: skills/recap/
  • Status: Admitted first-party; v0.1.2 released the prior form; the manual-only amendment is unreleased pending current-candidate acceptance.
  • Evidence: Current amendment tests in tests/test_functional_closure.py; frozen historical tests remain under skills/recap/tests/; user guide.
  • Installation path: skills/recap/ in a host-recognized Skills root.

release-workflow

  • Purpose: Publish a completed project — synchronize docs, run quality gates, tag, and release.
  • When to use: Project has passed project-review and is ready to publish.
  • Invocation: Model-invoked (or manual entry where supported).
  • Package: skills/release-workflow/
  • Status: Admitted first-party; MIGRATE — NO REWRITE (from LightDevCoder/release-workflow).
  • Evidence: SKILL.md.
  • Installation path: skills/release-workflow/ in a host-recognized Skills root.

research

  • Purpose: Investigate an external question against high-trust primary sources and capture findings.
  • When to use: Local preset or facts are insufficient; need external evidence.
  • Invocation: Model-invoked.
  • Package: skills/research/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt research).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/research/ in a host-recognized Skills root.

resolving-merge-conflicts

  • Purpose: Resolve an in-progress git merge or rebase conflict.
  • When to use: Merge/rebase halted with conflicts.
  • Invocation: Model-invoked.
  • Package: skills/resolving-merge-conflicts/
  • Status: Admitted first-party; PORT — NO REDESIGN.
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/resolving-merge-conflicts/ in a host-recognized Skills root.

review-loop

  • Purpose: Lightweight review/repair engine — resolve reviewer, invoke, receive findings, return repair, re-run.
  • When to use: Any artifact with a reviewer and a bounded repair window.
  • Invocation: Model-invoked; manual entry supported.
  • Package: skills/review-loop/
  • Status: Admitted first-party; REFACTOR + SPLIT (final acceptance moved to project-review).
  • Evidence: SKILL.md, references/.
  • Installation path: skills/review-loop/ in a host-recognized Skills root.

socratic

  • Purpose: Core clarification engine — internal decision frontier with rounds of independent questions, options, recommendations, batch replies, and shared-understanding confirmation.
  • When to use: Underlies clarify, project-clarify, decision-map; not a standalone project workflow.
  • Invocation: Model-invoked (engine for other Skills).
  • Package: skills/socratic/
  • Status: Admitted first-party; ADAPT (Matt grilling).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/socratic/ in a host-recognized Skills root.

tdd

  • Purpose: Test-driven development — red → green → refactor loop with real tests.
  • When to use: Implementing coding features test-first or fixing bugs with regression cover.
  • Invocation: Model-invoked.
  • Package: skills/tdd/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt tdd).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/tdd/ in a host-recognized Skills root.

teach

  • Purpose: Teach a new skill or concept within the workspace.
  • When to use: User wants a guided lesson on a topic.
  • Invocation: User-invoked only.
  • Package: skills/teach/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt teach).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/teach/ in a host-recognized Skills root.

to-questionnaire

  • Purpose: Turn an undecided question into a questionnaire for the person who holds the information.
  • When to use: Information is held by another person, not the current user.
  • Invocation: User-invoked only.
  • Package: skills/to-questionnaire/
  • Status: Admitted first-party; PORT — NO REDESIGN.
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/to-questionnaire/ in a host-recognized Skills root.

wait-what

  • Purpose: Re-pitch the last message that did not land.
  • When to use: User says "wait, what?" or similar confusion.
  • Invocation: User-invoked only.
  • Package: skills/wait-what/
  • Status: Admitted first-party; PORT — NO REDESIGN.
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/wait-what/ in a host-recognized Skills root.

wizard

  • Purpose: Interactive bash wizard for human-only steps (provisioning, secrets, dashboards, cutovers).
  • When to use: Task needs a guided human walk-through, not an agent-auto step.
  • Invocation: Model-invoked.
  • Package: skills/wizard/
  • Status: Admitted first-party; PORT — NO REDESIGN (Matt wizard).
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/wizard/ in a host-recognized Skills root.

writing-for-agents

  • Purpose: Author or edit agent-facing documents (Skills, AGENTS.md, CLAUDE.md) for model consumption.
  • When to use: Creating or improving agent instructions or Skill packages.
  • Invocation: Model-invoked.
  • Package: skills/writing-for-agents/
  • Status: Admitted first-party; PORT — NO REDESIGN.
  • Evidence: SKILL.md, ATTRIBUTION.md.
  • Installation path: skills/writing-for-agents/ in a host-recognized Skills root.

Source-state boundaries

State Where it belongs Catalog treatment
First-party This repository Listed above when admitted.
Approved Port (Matt) This repository with ATTRIBUTION.md Listed above; self-contained, no Matt runtime dependency.
Direct upstream Original upstream repository Mentioned as dependency; never copied here unmodified.
Modified third-party skills-3rdParty Listed in that private repository's source catalog after fork admission.
Deprecated or archived Released migration record Listed with replacement and migration guidance.

See maintenance for synchronization and admission for the ownership gate.