LightDevCoder/skills provides 34 first-party Agent Skills designed to work together across project planning, coding, and review, or run individually on demand. Each package lives in skills/<name>/ and defines its own behavior in SKILL.md.
Release: v0.2.0 is published from the
v0.2.0tag — the release line carries the full 34 first-party Skills (the 33-package architecture plushumanizer) — and is the current stable release. The exact commit is recorded in the release receipt.
The repository organizes capabilities into focused areas:
- Project Workflow: end-to-end delivery from project bootstrap to release.
- Clarification & Research: structured questions and primary-source investigation before building.
- Execution: bounded implementation tasks with host-aware execution planning.
- Review: read-only specialist checks and final project acceptance.
- Specialized Workflows: dedicated tooling for manuscripts, knowledge bases, language learning, and kanban boards.
- Router:
ask-lightinspects workspace state to suggest next steps.
Skills follow the progressive disclosure patterns of Matt Pocock Skills and the host-evidence inspection approach of Sol Advisor as design references without adding runtime dependencies.
Install Light Skills using the interactive Skills CLI:
npx skills add LightDevCoder/skillsInstall a single Skill:
npx skills add LightDevCoder/skills --skill project-review
npx skills add LightDevCoder/skills --skill researchPin to the v0.2.0 release:
npx skills add LightDevCoder/skills#v0.2.0Target a specific Agent directly:
npx skills add LightDevCoder/skills --agent claude-codeSee Installation for advanced options (explicit agent targets, copy mode, non-interactive CI flags), manual file copying, and verification notes.
$ask-light next # Suggest the next appropriate Skill from current context
$project-init # Bootstrap project structure and tracker contracts
$clarify # Clarify requirements through targeted questions
$project-clarify # Clarify project decisions using repository context
$implement # Execute a ready ticket with verification
$project-review # Run final acceptance checks: PASS / FAIL / BLOCKED
A typical project progresses through these stages, though you can start directly at any stage:
project-init
↓
project-clarify
↓
project-spec
↓
project-tickets
↓
implement
↓
project-review
↓
release-workflow
project-init: sets up tracker contracts and initial configuration.project-clarify → project-spec → project-tickets: clarifies requirements, writes the specification, and splits work into executable tickets.implement: implements one ticket at a time with automated checks.project-review: verifies quality against frozen baselines;review-loophandles iterative fixes.release-workflow: runs release validation, tagging, and publication.
Direct paths for common tasks:
clarify # Standalone brainstorming and clarification
implement # Implement a well-defined ticket directly
diagnosing-bugs → implement # Diagnose an issue, then apply the fix
release-workflow # Publish an approved release
$ask-light # Route unclear tasks to the right Skill
See docs/workflows/ for full workflow guides.
$ask-light next
$ask-light workflow
ask-light is the Light workflow advisor, navigator, and router. It inspects project and workflow evidence, reasons about the best next Skill, explains why, and waits for approval before a validated host-aware transition (model-invoked targets may begin where supported; user-invoked targets fall back to rendering the exact invocation when direct Host transition is unavailable).
See ask-light and docs/workflows/.
| Group | Skills | Details |
|---|---|---|
| Project | project-init, project-clarify, project-spec, project-tickets, implement, project-review, release-workflow |
CATALOG.md |
| Clarification & Research | socratic (engine), clarify, project-clarify, decision-map, research, prototype, to-questionnaire |
clarification-system |
| Execution | implement, agent-config, tdd, diagnosing-bugs, resolving-merge-conflicts |
execution |
| Review | review-loop (engine), generic-review, code-review, project-review (acceptance) |
review-system |
| Specialized | manuscript-ops, kb-init, learn-anything, language-learning, kanban-worker, eli5, recap |
specialized-workflows |
| Productivity | handoff, humanizer, wizard, wait-what, writing-for-agents |
CATALOG.md |
See CATALOG.md for full descriptions, invocation modes, and package paths.
| Origin | Policy | Repository Treatment |
|---|---|---|
| First-party | Collection owner authored | Maintained in skills/<name>/. |
| Approved Port (Matt Pocock) | Upstream behavior preserved with ATTRIBUTION.md |
Self-contained in skills/<name>/ without upstream runtime dependencies. |
| Third-party unmodified | External upstream | Recommended for direct installation; not duplicated here. |
| Modified third-party | Managed in private LightDevCoder/skills-3rdParty |
Retains full patches, licenses, and sync locks. |
| Retired standalone | Consolidated into collection | Documented with migration history in release records. |
Approved Matt Ports (11 packages): research, prototype, tdd, handoff, diagnosing-bugs, wizard, teach, wait-what, to-questionnaire, writing-for-agents, resolving-merge-conflicts. Each package contains ATTRIBUTION.md and runs without external runtime dependencies.
Adapted origin (1 package): humanizer is a substantially transformed first-party capability based on blader/humanizer (2.11.2) with a thin Chinese adaptation layer informed by op7418/Humanizer-zh; both MIT licenses are preserved in its ATTRIBUTION.md.