Use when deciding whether something learned this session is worth persisting to agent memory, when writing or updating a memory file, and when a saved memory turns out to be wrong. Defines the four memory genres, the selection bar, the evidence-anchor rule, and correction discipline. The repo carries specs/ADRs/beads; memory carries what none of those can.
87%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Memory is not a diary. A 30-file memory corpus from a four-day port kept ~25 commits/day sustainable across amnesiac sessions because every file passed a bar: its absence already burned a red→green cycle, or would burn one next session. Files that don't pass that bar are noise that dilutes recall.
Don't save what the repo already records: code structure (read the code), past work (git log), open work (the tracker), requirements (specs/feature files), settled decisions (ADRs). Memory holds the residue: empirical facts, traps, seams, and discipline — the things a fresh session cannot derive and will otherwise rediscover the expensive way.
Every claim carries its anchor, or it's vibes:
That triple — artifact anchor, tracker anchor, evidence anchor — is what makes a memory corpus auditable instead of folklore.
Memories are falsifiable. When a saved claim is disproven:
**CORRECTION (was wrong here):** marker,
preserving the wrong belief as a warning — a silent rewrite erases the
trap for the next session that re-derives the same wrong belief.description is load-bearing — recall decisions
are made on it, so write it as the compressed rule itself.[[link]] and index line
resolves to a real file. Mechanical: scripts/memory-lint.bb <memory-dir> checks dangling wiki-links, dangling/missing index
entries, unindexed files, frontmatter, and oversize — run against the
field corpus it found two dangling links nobody had spotted.igouss/gitweb-rs
9ef2429
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.