session-audit

name	session_audit
status	codified
version	1.0.0
tier	episodic
description	Scan project memory · age distribution + ungrouped count (L1 build readiness) + per-namespace entity stats · NO LLM.
inputs	[{"project (str, optional)":"project slug · defaults to active"},{"mem_dir (path, optional)":"override mem_dir resolution"},{"lookback_days (int, default 30)":"age bucketing threshold"}]
outputs	["total_sessions","age_distribution (fresh / recent / aged · session counts per bucket)","ungrouped_count (sessions not covered by L1 index)",{"entity_namespace_stats (counts per ns":"wiki/people/companies/concepts/sessions)"}]

session_audit

Pre-L1-build readiness check · before triggering cli_self_evolve --all, this skill tells you which projects have enough ungrouped sessions to make L1 collapse worthwhile (threshold default 3 in self_evolve).

When to use

• Pre-L1-build: which projects should I run self_evolve on?
• Memory health: are sessions piling up uncollapsed?
• Entity graph: how many [[ns/name]] links across the corpus?

Why codified

• 3 distinct call-sites already (CLI / self_evolve / paper3 metrics)
• Single pure-Python scan · O(N) over .md files · deterministic
• Reused by lifecycle_report internally (DRY)