From infinite-brain
Periodic vault maintenance: audit orphans, contradictions, confidence gaps, stale nodes, cross-link opportunities, taxonomy health, and summary quality. Run weekly or bi-weekly. Invoked with /organize-vault.
How this skill is triggered — by the user, by Claude, or both
Slash command
/infinite-brain:organize-vaultThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are auditing the Infinite Brain knowledge graph for health issues.
You are auditing the Infinite Brain knowledge graph for health issues.
_system/INDEX.md for current state baseline.1. Orphan Census
Find nodes with zero edges and zero related links. For each: suggest 2-3 connection targets with edge types. If truly standalone, flag with #standalone.
2. Contradiction Scan
Compare nodes in the same namespace for conflicting claims; pay special attention to contradictory edges between hypothesis/fact/decision nodes. Report as: [node-a] contradicts [node-b] about [topic].
3. Confidence Gaps
Find nodes where confidence is missing, 0.0, or inconsistent with content. Suggest updated values based on available evidence. Flag nodes with high confidence but a triggered staleness signal.
4. Stale Node Detection
Check staleness_signal conditions. Flag nodes where verified_at is over 90 days old. Priority: pillars > decisions > facts > patterns > hypotheses.
5. Cross-Link Opportunities Find nodes sharing 2+ tags with no edge. Suggest specific edge types and weights with rationale. Look for implicit hierarchies (patterns implementing pillars, decisions deriving from hypotheses, etc.).
6. Taxonomy Health Compile the master tag list (create one if missing). Flag inconsistent spellings (e.g. "AI" vs "ai" vs "artificial-intelligence"). Suggest merges for near-identical tags.
7. Visibility Health
Find nodes missing visibility. Flag conflicts between content sensitivity or namespace scope and the current setting. Always suggest one of: public, namespace, private, system.
8. Summary Quality Flag summaries over 200 chars or 3+ sentences, empty or placeholder text ("TBD", "description", "1-2 sentences..."), or summaries that don't match node content.
Write the report's prose — issue descriptions, suggested fixes, action text — in the vault's document language (see _system/AGENTS.md → Document Language). Keep file paths, id slugs, type/edge/visibility names, and tags in canonical English.
## Vault Health — [Date]
### Summary
- Total nodes: XX | Orphans: XX | Contradictions: XX | Stale: XX | Cross-link opportunities: XX
### Priority Actions (Top 5)
1. [ ] [high] ...
2. [ ] [med] ...
...
### Details
[File path | Issue | Suggested fix]
After delivering the report and applying any user-selected fixes:
Write a log node to logs/log-organize-vault-YYYYMMDD-HHmmss.md using the log schema from _system/FRONTMATTER-SCHEMA.md:
operation: organize-vaultaffected_nodes: node IDs modified (empty array if user selected "none")summary: one sentence — node count, issue count, fixes appliednpx claudepluginhub jhs512/ib --plugin infinite-brainCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.