| name | skill-integration |
| description | Manages skill-creator integration with GSD workflows including skill loading, session observation, bounded learning guardrails, and pattern detection. Use this skill whenever: executing any GSD phase (to ensure relevant skills are loaded first), starting or resuming a work session (to check for pending suggestions), the user asks about skills, patterns, or skill-creator status, performing code changes that might represent repeating patterns, the user corrects Claude's output (highest-signal observation), or when skill refinement, creation, or suggestion review is discussed. Critical for maintaining the adaptive learning layer.
|
| user-invocable | true |
| version | 1.0.0 |
| format | "2025-10-02T00:00:00.000Z" |
| triggers | ["Manages skill-creator integration with GSD workflows including skill loading, session observation, bounded learning guardrails, and pattern detection"] |
| updated | "2026-04-25T00:00:00.000Z" |
| status | ACTIVE |
Skill-Creator Integration
Manages skill loading, observation, and bounded learning for the GSD ecosystem. Ensures learned knowledge persists across sessions and subagent contexts.
Skill Loading Before GSD Phases
Before executing any GSD phase, load relevant generated skills:
- Check
.claude/commands/ for project-level skills
- Check
~/.claude/commands/ for user-level skills
- Project-level skills take precedence over user-level on conflict
- Load only skills relevant to the current phase and task
- Respect the token budget: 2-5% of context window maximum
Critical: When forking subagent contexts for GSD phases (execute-phase, verify-work), include relevant skills in the subagent's context. Clean context means free of stale conversation history -- not free of learned knowledge. A subagent that can't access learned skills is throwing away everything skill-creator has captured.
For full 6-stage pipeline details, see references/loading-protocol.md.
Token Budget Management
The 6-stage skill loading pipeline (Score, Resolve, ModelFilter, CacheOrder, Budget, Load) manages what gets loaded. When multiple skills compete for budget:
- Phase-relevant skills get priority (e.g., testing skills during
verify-work)
- Recently activated skills rank higher than dormant ones
- Project-level skills override user-level duplicates
- If budget is exceeded, queue overflow skills and note what was deferred
Session Observation Protocol
During all work sessions, maintain awareness of patterns worth capturing:
- Tool sequences that repeat across sessions (e.g.,
git log -> read file -> edit -> test -> commit)
- File patterns consistently touched during specific GSD phases
- Commands run after test failures or verification issues
- Corrections the user makes to agent output -- these are the highest-signal observations
- Phase outcomes: success, failure, partial, and what was different
If .planning/patterns/ exists, record observations to sessions.jsonl in that directory. This data feeds skill-creator's pattern detection pipeline.
For observation taxonomy, signal strength ranking, and JSONL schema, see references/observation-patterns.md.
Bounded Learning Guardrails
These constraints are NON-NEGOTIABLE:
