Skip to content

_index.md

Latest commit

 

History

History
70 lines (57 loc) · 3.16 KB

_index.md

File metadata and controls

70 lines (57 loc) · 3.16 KB

Knowledge Activation — Skill Registry

schemaVersion: 1.0.0

last-updated: 2026-03-19

Skill Index

name version tier autonomous-safe confidence.initial half-life-days decay-model next-skills
bootstrapping-agent-directives 1.0.0 meta false 1.0 null none —
discovering-institutional-knowledge 1.0.0 meta true 1.0 null none creating-skills (on gap detected)
creating-skills 1.0.0 meta true 1.0 null none assessing-skill-quality (unconditional)
assessing-skill-quality 1.0.0 meta true 1.0 null none —
reporting-skill-gaps 1.0.0 meta true 1.0 null none —
uninstalling-knowledge-activation 1.0.0 meta false 1.0 null none —
[infra-standards-stub] stub org false — 45 exponential —
[security-baselines-stub] stub org false — 45 exponential —
[build-pipeline-stub] stub team false — 30 exponential —
[signing-stub] stub team false — 30 exponential —

Decision Dependency Graph

bootstrapping-agent-directives depends-on: [] depended-on-by: [all skills — bootstrapper installs the directives that activate them]

discovering-institutional-knowledge depends-on: [knowledge-sources.json slot configuration] depended-on-by: [CLAUDE.md hedging directive]

creating-skills depends-on: [skill schema, _index.md] depended-on-by: [discovering-institutional-knowledge (on gap), reporting-skill-gaps (on gap)]

assessing-skill-quality depends-on: [validators/skill-quality-checklist.md, validators/skill-schema.md] depended-on-by: [creating-skills, CONTRIBUTING.md gate]

reporting-skill-gaps depends-on: [gap-log.ndjson, hooks/flush-gap-log.sh, gh CLI] depended-on-by: [CLAUDE.md gap directive, Stop hook]

uninstalling-knowledge-activation depends-on: [] depended-on-by: []

Activation Directives

Canonical source — bootstrapping-agent-directives writes these into CLAUDE.md

  1. Before every task: read this file, identify matching skills, inject within context budget before responding
  2. When uncertain or hedging: invoke discovering-institutional-knowledge instead of guessing
  3. When learning undocumented knowledge: invoke creating-skills to capture it
  4. When a skill fails, is missing, or is wrong: append to .claude/knowledge-activation/gap-log.ndjson
  5. Never invoke bootstrapping-agent-directives autonomously

Context Budget Reference

Full config: .claude/knowledge-activation/context-budget.json

Injection order: meta → org → team → session → task

Overflow strategy: truncate-lowest-confidence

Summarization threshold: 180,000 tokens

Event Types

Core canonical types (v1 exhaustive):

skill.invoked, skill.completed, skill.gap-detected, skill.confidence-below-threshold,

gap-report.queued, gap-report.filed, gap-report.deduplicated,

quality-gate.started, quality-gate.passed, quality-gate.failed,

bootstrap.completed, plugin.stale-detected,

context-budget.layer-exhausted, context-budget.summarization-triggered

Subdomain extensions: [subdomain].[verb] — register below:

example-mobile: (none yet)