| name | audit-ai-memex |
| description | Audit AI Memex health — detect stale entries, spoke-hub drift, broken references, and dependency propagation risks across the knowledge base |
Audit AI Memex
Assess the health and freshness of the AI Memex knowledge base. Run from the
lqdev.me repo root where _src/resources/ai-memex/ and graph.json are available.
When to Run
- After importing entries (verify integration with existing knowledge graph)
- At session end ("Anything stale or broken in the Memex?")
- When updating a hub entry (check what depends on it)
- Periodically to catch drift between spokes and hub
- When user asks: "How fresh is my Memex?", "What needs updating?"
Quick Audit
Run the backing script from the lqdev.me repo root:
& "$HOME/.agents/skills/audit-ai-memex/scripts/audit-memex.ps1"
The script outputs a structured report covering:
- Entry freshness — age since last update, flagging entries older than 90 days
- Spoke-hub drift — entries where spoke has a newer
last_updated_date
- Broken references —
related_entries pointing to non-existent files
- Orphan entries — entries with 0 inbound references (isolated knowledge)
- Critical hubs — entries with 3+ inbound references (high propagation risk)
Interpreting Results
Staleness
Entries unchanged for 90+ days aren't necessarily stale — patterns and references
may be stable. Focus on:
project-report entries (likely outdated as projects evolve)
- Entries whose
tags reference fast-moving technologies
Spoke-Hub Drift
If a spoke entry is newer, use the import script to update:
& "$HOME/.agents/skills/import-ai-memex/scripts/import.ps1"
Broken References
Fix by either:
- Removing the dead reference from
related_entries
- Creating the missing entry if the knowledge gap is real
Update Propagation
When updating a critical hub entry (3+ inbound refs), review its dependents.
Read references/REPORT-FORMAT.md for the dependency impact template.
Gotchas
graph.json must exist at _public/resources/ai-memex/graph.json for dependency
analysis. Run dotnet run first if it's missing.
- Entries written directly in the hub (no spoke equivalent) are never flagged as
drifted — they're hub-only by design.
- The
related_entries field is seeded by the knowledge graph at build time.
An entry showing 0 inbound refs may gain connections after the next build.