agentleFS
Sign inStart free
bersimaobersimao/ai-memory-system/skills/memory-write/SKILL.md

memory-write

Skill2 starsChanged 21 days agoMITView on GitHub

Save, correct, or retrieve durable shared memory across Claude Code and Codex. Use for remembered preferences, prior decisions, project history, explicit save requests, and implicit references to earlier work.

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. opensite-ai/opensite-skills★ 15same name
    2. yktsnet/dotfiles-public★ 1same name
    3. ztnkv/personal-operating-system★ 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 repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use ~/.claude/scripts/mem.

    line 8
  2. Very specific:

    Read the existing target and check duplicates, with mem search too: it covers MEMORY-COLD.md, which the index no longer shows.

    line 23
  3. Distinctive:

    Put additions in a temporary file, then:

    line 26
  4. Distinctive:

    Every accepted transaction records originals and replacements before writing.

    line 37
  5. Very specific:

    When an index fills: with salience on, run mem rebalance RELATIVE_PATH first on a project index (weak lines to MEMORY-COLD.md).

    line 43

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Shared memory

Use ~/.claude/scripts/mem. Markdown is authoritative; both harnesses use the same journaled writer. Do not create a parallel harness-specific memory database.

For recall, run mem search "query" --scope all, inspect project labels, and expand relevant REQUEST_ID:RESULT_ID results before relying on them. Try domain identifiers, English/Portuguese alternatives, and --semantic if needed. A lexical miss does not establish absence. Verify that a result from another project applies to this task.

For saves, preserve existing KB/domain-skill routing when a skill owns the fact. Preferences belong in context/USER.md; shared workflow facts in context/MEMORY.md; project facts in that project's context/topics/ with an index pointer. In an index with ## <topic> sections, a new page's pointer goes at the end of that section's hub (the topics/ page linked on the section's first line), not into MEMORY.md. An always/never rule belongs in the repo's instruction file, not in memory. Resolve the project using node ~/.claude/hooks/project-store.js --json --resolve "$PWD". Read the existing target and check duplicates, with mem search too: it covers MEMORY-COLD.md, which the index no longer shows. Do not store credentials or secrets.

Put additions in a temporary file, then:

~/.claude/scripts/mem prepare RELATIVE_PATH --text-file /tmp/addition.md --operation append --reason "why" > /tmp/memory-proposal.json
~/.claude/scripts/mem apply /tmp/memory-proposal.json

Inspect proposals before applying. A correction uses replace with the complete new text and requires apply --reviewed; archive also requires review. Honor existing user authorization and confirm removals when not already authorized. Never force a stale revision (regenerate it from the current file) or rewrite memory with direct file tools. Every accepted transaction records originals and replacements before writing. The lock coordinates these writers; arbitrary external file edits do not participate in the lock.

Caps, split, salience

Caps are characters: USER 1,375; global MEMORY 4,000; project index 2,500 or its .cap. When an index fills: with salience on, run mem rebalance RELATIVE_PATH first on a project index (weak lines to MEMORY-COLD.md); still over, or any other index → mem split RELATIVE_PATH: it moves the largest ## sections whole to their hub and leaves one pointer line each; with nothing movable it archives the index verbatim in index-*.md, linked from the new index. Finished topic: mem close-section RELATIVE_PATH "## X" (section to its hub, pointer under the closed heading, config closed_heading, default ## Closed). Never compress facts to fit. Scheduled curation runs the same split.

Salience (opt-in, config salience: true; optional salience_recent_days limits it to stores with a recent session): curation keeps a project index's strongest lines (fresh, recalled, opened, recalled from another project counts double, or pinned with mem pin RELATIVE_PATH "fragment") within cap - 300 and moves weak ones to MEMORY-COLD.md beside it: searched and recalled, not loaded at startup; a cold line used again comes back. mem rebalance RELATIVE_PATH --dry-run lists the moves.

Lossless full-index archives stay beside the index, preserving existing relative link destinations. Ordinary topic pages stay in topics/.

Checkpoints and diagnostics

Hooks capture incremental assistant output, including short sessions. Capture is a fallback record, not a substitute for explicitly saving decisions with their rationale. Without hooks: mem checkpoint --session ID --cursor UNIQUE_EVENT --text "decision and why". Distilled candidates and raw checkpoints are unverified until checked against their evidence. Startup snapshots are bounded (a Claude Code session started by /clear omits the latest checkpoint); retrieve omitted sources as needed and inspect the full manifest with mem snapshot --json.

python3 ~/.claude/scripts/memory-doctor.py reports installation and corpus health. mem report measures retrieval engagement; opening a result does not prove correctness.

Confirm the saved location. Accepted writes are immediately visible to lexical search; semantic search sees them after the next index run.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free