audit-what-the-agent-remembers
Use when you want a full inventory of stored memory before trusting it. Enumerates entries, dates, and use counts so dead weight and gaps are visible.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 81
- Forks
- 17
01 前言 Front Matter
- name
- audit-what-the-agent-remembers
- description
- Use when you want a full inventory of stored memory before trusting it. Enumerates entries, dates, and use counts so dead weight and gaps are visible.
02 资源文件 3
- 许可证LICENSE1.0 KB
- 参考文档SOURCE.md356 B
- 参考文档TRUST.auto.yaml2.2 KB
03 SKILL.md 正文
Audit What the Agent Remembers
You cannot reason about memory you have not listed. An audit enumerates every stored fact with its age and whether it was ever used, turning a black box into a table.
Procedure
- Enumerate the store:
find ~/.hermes/memories -type f -name '*.md' | sort. - Dump every entry with its date:
grep -rn "^20[0-9][0-9]-" ~/.hermes/memories/. - Count by age bucket and print the histogram; a healthy store is mostly recent, with a small aged tail.
- Flag entries with no date — they cannot be aged or pruned, so they are the first gap.
- Flag entries with no source — unverifiable by construction.
- Cross-check for the same key appearing twice:
grep -oE '^[a-z-]+ *=' ... | sort | uniq -d. - List entries never referenced since creation, if use is tracked; the never-used set is the pruning queue.
- Compare the store against the questions you actually ask — a store full of facts you never query is misrouted (see
route-a-fact-to-memory-session-or-skill). - Write the audit to
notes/memory-audit.md: counts, undated, unsourced, duplicated, unused. - Report the four numbers before making any change; audit first, prune second.
Pitfalls
-
Auditing by memory of what you stored instead of dumping the store to a file.
-
Counting entries by file and missing that one file holds forty facts.
-
Ignoring undated entries in the histogram, so the age picture is wrong.
-
Treating a large store as healthy when most entries are never read.
-
Deleting during the audit rather than after it, so the inventory is incomplete.
-
Reporting totals without the file they came from, so the audit cannot be re-run.
-
Auditing only the memory store and ignoring skills and session notes that also hold facts.
-
Skipping the duplicate check, so a fact counted once is actually stored three times.
Verification
wc -l notes/memory-audit.md; grep -c "undated\|unsourced\|duplicate\|unused" notes/memory-audit.md
# passes when the four counts are present and sum to the total entry count
Report to the user: total entries, and the counts of undated, unsourced, duplicated, and unused.
04 同仓库更多 Skills
recipe-performance
Diagnose a specific MetaMask user flow's real-app performance and Sentry critical-user-flow telemetry with deterministic mm-harness recipes. Use for CUF boundaries, cold/warm/background-resume behavior, latency regressions, lifecycle tags, duplicate or blocking network requests, long tasks, layout/paint cost, screen redraws, React rerenders, or browser trace analysis where tests alone cannot prove the real app.
mexc-spot-websocket
MEXC Spot WebSocket workflows for public streams, protobuf handling, depth maintenance, listen keys, and private streams without exposing API keys.
mexc-spot-rest
MEXC Spot REST workflows for public data, signed account requests, orders, wallet/admin actions, and secure signing without exposing API keys.
mexc-futures-websocket
MEXC Futures WebSocket workflows for public streams, private account/order/position streams, ping/pong, subscriptions, and depth maintenance.
0xklen-version-and-refresh-sanctions-lists
暂无简介
verify-tests-cover-changed-lines
Use when a PR adds code that the existing suite may not exercise. Measures patch coverage — the share of newly changed lines hit by tests — before approving.