Teaches agents how to discover, select, and invoke skills from the skill library. Use this skill whenever you're uncertain which skill applies to a task, when composing multiple skills for complex work, or when you need to understand what skills are available. This is your go-to when facing an ambiguous task and need to figure out the right approach before diving into implementation.
Agents use this skill to navigate the skill library at runtime. It teaches discovery, selection, invocation, and composition so agents can choose the right skill without human guidance.
skills/GUIDE.md — the master index with quick-selection table and task-type mappingskills/ directory — each subfolder contains a SKILL.mdskills/_shared/conventions.md — cross-cutting rules all skills followWhen facing a task:
Most tasks need exactly one skill. Pick it and go.
User asks: "Fix the login timeout bug"
→ implementing-features (clear bug fix with acceptance criteria)
Compose only when one skill's output feeds another skill's input.
User asks: "Design and implement a caching layer"
→ designing-systems (produces ADR with approach)
→ implementing-features (uses ADR as spec)
→ writing-tests (validates implementation)
When invoking a skill:
I'm using the [skill-name] skill to handle this task.
Pre-flight:
- [x] Condition 1 met
- [x] Condition 2 met
Proceeding with execution steps...
Skills execute one after another, each consuming the previous output.
planning-tasks → implementing-features → writing-tests
Rule: Only chain when output A is input B. Independent tasks run separately.
When planning reveals unknowns, fork into research before continuing.
planning-tasks
├── researching-options (unknown 1)
├── researching-options (unknown 2)
└── continue with implementing-features
When implementation requires an architecture decision, gate on design.
designing-systems (must complete first)
└── implementing-features (uses ADR as spec)
Setup and context management. Typically run once or periodically.
initialize-repo — set up .context/ for a new projectcontext-loader — how to read .context/ efficientlycontext-maintenance — keep .context/ accuratecontext-review — full codebase scan to sync docsDefine what to build and how. Run before implementation.
planning-tasks — break goals into task liststask-plan — canonical plan format for handoffsdesigning-systems — architecture decisions (ADRs)design-first — triage: design first or implement directly?researching-options — evaluate libraries/approachesBuild and validate. The core work loop.
implementing-features — write/modify codewriting-tests — write/run testssystematic-debugging — structured debugging processCross-cutting quality practices. Referenced during other skills.
common-constraints — universal agent rulesverification-checklist — pre-completion checklisttesting-discipline — TDD and test quality practicescommit-discipline — git commit conventionsReflection and improvement. Run after tasks complete.
task-retrospective — structured reflectionknowledge-graduation — promote patterns to permanent docs