LightDevCoder/skills provides 36 first-party Agent Skills designed to work together across project planning, coding, and review, or run individually on demand. Each package lives in skills/<category>/<name>/ and defines its own behavior in SKILL.md.
Release: v0.2.5 is the current stable release of all 36 first-party Skills. Its immutable release manifest and post-publication release receipt record the scope and verification.
All categories and collection guides. The 36 source packages are grouped by purpose; names and invocation modes are unchanged.
- project — 8 skills
- engineering — 5 skills
- review — 4 skills
- thinking — 5 skills
- knowledge — 5 skills
- writing — 3 skills
- productivity — 6 skills
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 profile-driven execution configuration.
- 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 the latest collection from the default branch main:
npx skills add LightDevCoder/skillsInstall a single Skill from main (for example, the updated agent-config Skill):
npx skills add LightDevCoder/skills --skill agent-config
npx skills add LightDevCoder/skills --skill project-review
npx skills add LightDevCoder/skills --skill researchNote: An unqualified repository source (
LightDevCoder/skills) follows the default branchmain, delivering the latest admitted features and integrations.
To install the reproducible stable release snapshot, pin the #v0.2.5 tag:
npx skills add LightDevCoder/skills#v0.2.5
npx skills add LightDevCoder/skills#v0.2.5 --skill project-retroPrevious releases (such as #v0.2.4, #v0.2.2, #v0.2.1, and #v0.2.0) remain available for reproducible installs:
npx skills add LightDevCoder/skills#v0.2.4
npx skills add LightDevCoder/skills#v0.2.2
npx skills add LightDevCoder/skills#v0.2.1
npx skills add LightDevCoder/skills#v0.2.0Note: The
#v0.2.0tag remains an immutable, reproducible snapshot of the v0.2.0 release line and is not modified by subsequent work onmain.
npx skills add LightDevCoder/skills --agent claude-codeThe agent-config Skill uses an optional companion MCP server for host inspection and profile persistence, maintained in LightDevCoder/agent-config:
git clone https://github.com/LightDevCoder/agent-config.git
cd agent-config && npm ci && npm run build && npm install -g .
agent-config setup --checkask-light and agent-config support optional semantic acceleration using TypeSafe Jev System One models for fast intent calibration, ambiguity detection, and abstract task profiling:
export TYPESAFE_API_KEY="your-api-key"When configured, Jev enhances candidate choices within deterministic boundaries. When unconfigured or offline, both Skills gracefully fall back to zero-dependency deterministic baselines with no change in fail-closed safety.
See 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 workspace and task tracker settings
$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-retro (agent evaluates friction)
project-init: sets up project layout and task tracker settings.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 acceptance criteria;review-loophandles iterative fixes.release-workflow: runs release validation, tagging, and publication.project-retro: evaluates session friction (navigation, automated checks, instruction clarity, tool usage) and proposes concrete improvements at the conclusion of the workflow.
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 (primary harnesses: 10 native adapters + fallback), 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, light-travelpage |
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/<category>/<name>/. |
| Approved Port (Matt Pocock) | Upstream behavior preserved with ATTRIBUTION.md |
Self-contained in skills/<category>/<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 (2 packages): 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.
light-travelpage substantially transforms do-tongxue/Travel-Plan-Page with protected D1 collaboration, validation, recovery and generation/update tooling; its MIT provenance is in ATTRIBUTION.md.