memory-review
Skill2 starsChanged 39 days agoMITView on GitHub
Sort Claude Code's auto memory by the memory rule - preferences into rules, project knowledge into repo docs, progress deleted. Read-only until approved. Run with /memory-review.
- Reads credentials
7 other projects use a version of this.
Adopted by
Most recently taken up by
- cveralyon/axel-setuptheir own version, first seen Nov 8, 2026
- denormans/claude-code-workflowtheir own version, first seen Nov 8, 2026
- letsloose501/sqs-agent-memorytheir own version, first seen Nov 3, 2026
- lorena-bordonaba-pau/youtube-agenttheir own version, first seen Oct 13, 2026
- Hassaan146/claude-skillstheir own version, first seen Jul 25, 2026
- dekimuhq/claude-memorytheir own version, first seen Jun 3, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- Hassaan146/claude-skills★ 18same name
- cveralyon/axel-setup★ 4same name
- finger88/my-claude-skills★ 3same name
- letsloose501/sqs-agent-memory★ 2same name
- lorena-bordonaba-pau/youtube-agent★ 2same name
- denormans/claude-code-workflow★ 1same name
- dekimuhq/claude-memory★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository16 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- codebase-researchSkill
- context-transferSkill
- feature-pr-descriptionSkill
- how-it-worksSkill
- human-voice-writingSkill
- i18n-translateSkill
- impact-checkSkill
- orchestrateSkill
- promotion-pr-descriptionSkill
- pr-review-triageSkill
- qa-testerSkill
- release-notesSkill
- thorough-diff-reviewSkill
- walkthroughSkill
- writeups-cleanupSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 46
every memory, one verdict:
- Distinctive: line 69
rule: add the lines to the named rule file and commit them in the repo that holds it, then delete the memory.
- Distinctive: line 71
docs: write the text into the named doc and commit it in that repo on the branch the user names (else the checked-out branch, as its own …
- Very specific: line 73
delete: remove the file and its index line, and remove [[name]] links to it from the others.
- Very specific: line 84
If local.md exists next to this file, read it.
“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Memory review
Memories are loaded into sessions and steer the work. A stale one is worse than none: it states
something that stopped being true with the confidence of a fact. A misplaced one hides knowledge
from the team: the memory rule (rules/memory.md in countersign-skills, imported by the global
instructions) says memory holds only how the user wants the agent to work. Cross-project
preferences belong in the rules, project knowledge in that repo's docs, and progress nowhere. This
skill sorts every memory by that rule and changes nothing without the user's approval.
Where memories live
Each project's memory is ~/.claude/projects/<key>/memory/ (or under $CLAUDE_CONFIG_DIR), where
<key> is the project's absolute path with every character that is not a letter or digit turned
into -. It holds MEMORY.md, an index of one line per memory, and one file per memory with
name, description and metadata.type frontmatter. When the folder is a link into your memory
repo, review and commit there.
Scope: every project folder that has memory files, or only the projects the user names.
Step 1: mechanical checks, no agents
For each project, read only the frontmatter and MEMORY.md, and list:
- an index line whose file is missing, and a memory file with no index line
- a file with missing or malformed frontmatter
- a
[[name]]link to a memory that does not exist in that project - a
projectmemory whose content states a date, release or deadline that has passed - a memory about progress: a plan, in-flight work, "waiting for X", or a feature that shipped
- the same preference in two projects' memories (a candidate for a rule)
- a memory that repeats the global instructions (
~/.claude/CLAUDE.mdand what it imports) or a skill almost word for word - a project memory folder that is a plain folder while others are links into a memory repo (not adopted yet)
Step 2: one read-only agent per project
Spawn one Explore agent per project with a memory folder, model: "sonnet", at most four at a
time, in one message. Brief each with the project's repo path and its memory folder, and ask for:
- every memory, one verdict:
- rule: a preference true in every project. Name the rule file (a rule module, or the
profile's
CLAUDE.md) and give the exact lines to add, without the story. - docs: project knowledge (a decision, a domain term, a gotcha, an infra quirk). Name the
repo doc for its area, or
docs/decisions.mdwhen none fits, and give the exact text. When the repo is public and the fact is private, the verdict is keep instead. - delete: progress, a shipped feature, something stale, or something already written in the repo, a rule or a skill (name where).
- keep: personal feedback about this project that can't go in its repo, or a pointer to a system outside the repo.
- update: a keep whose text is wrong, with the exact replacement.
- rule: a preference true in every project. Name the rule file (a rule module, or the
profile's
- the evidence for each verdict: the file, line, command output or git state that shows it, checked now, never recalled
- concrete claims checked against the repo: files, functions, flags, scripts and settings still
exist; rules still match
CLAUDE.md,CONTRIBUTINGor the rules folder; branches, releases and pull requests are in the state the memory says - at most 30 lines back, no file contents, and never anything from
.envfiles or other secrets
Step 3: one report, then only what the user approves
Show one table per project: memory, verdict, evidence, and for rule, docs or update the proposed text and where it goes. Ask once, for the whole table, which lines to apply. Apply exactly those:
- rule: add the lines to the named rule file and commit them in the repo that holds it, then delete the memory.
- docs: write the text into the named doc and commit it in that repo on the branch the user names (else the checked-out branch, as its own commit), then delete the memory.
- delete: remove the file and its index line, and remove
[[name]]links to it from the others. - update: rewrite the file, keep its frontmatter, and refresh its index line if the description changed.
Deleting a memory means removing its file and index line and the [[name]] links to it. In a
memory repo, commit once per review, for example chore: review memories for <projects>, with
the reviewed memory folders as the pathspec: git commit -m "<message>" -- <folder> ....
Sessions in other projects share the repo and may have staged their own changes, and a bare
git commit takes those too. Never push unless asked.
If local.md exists next to this file, read it; where it disagrees with this file, it wins.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

