agentleFS
Sign inStart free
BrendenKennedyBrendenKennedy/claude-for-ai-platforms/.claude/skills/memory/SKILL.md

memory

Skill1 starsChanged 3 months agoMITView on GitHub

Cross-session working memory + the branch-per-session git workflow — the read/write process for `.claude/memory/` (`sessions/`, `roadmap.md`, `reference/`). RECALL: when earlier work is referenced, grep/read `sessions/` newest-last and answer from it. RECORD: after substantive work, add `sessions/YYYY-MM-DD-<slug>.md` from the template + update the roadmap. REFERENCE: promote reusable patterns to `reference/`. BRANCH: don't commit straight to main — branch per session/wave, commit only when asked, land, note branch+hash in the session record. Load when starting or wrapping a unit of work, or when past sessions come up. Triggers: record the session, wrap up, close out, session note, roadmap, what did we do, what did we decide, last session, recall, remind me, start a branch, feature branch, land this, merge to main, commit this.

We split this skill into the rules it gives an agent

5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 75 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the relevant entry (or the latest 1–3) and answer from it.

    line 35
  2. Very specific:

    Add .claude/memory/sessions/YYYY-MM-DD-<slug>.md from the template below — a *refined* summary, not a transcript (~one screen: outcomes + …

    line 38
  3. Very specific:

    Update roadmap.md: check off finished items, move them to Done (recent), add follow-ups.

    line 40
  4. Very specific:

    If you established a reusable "how we do X", add/adjust a reference/ note (or promote it to a skill).

    line 41
  5. Very specific:

    Branch per unit of work, off main: Don't commit straight to the trunk — cut a branch named for the work (e.g.

    line 76

“Common” is measured: kept by at least 6 in 10 of the 75 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.

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

Memory — journaling past work + branch hygiene

On-demand: load when you start a unit of work (branch it), wrap one up (record it), or when the user references earlier sessions (recall it). .claude/memory/ holds the stored notes (the DATA); this skill is the PROCESS that reads and writes them — pulled in on demand, never auto-loaded. (This is the in-repo, project-scoped memory — not the personal auto-memory at ~/.claude/projects/**/.)

The data store (where notes live)

PathHolds
.claude/memory/sessions/dated refined summaries of each substantive session (YYYY-MM-DD-<slug>.md, newest-last)
.claude/memory/roadmap.mdliving backlog: next · in-progress · done-recent
.claude/memory/reference/stable "how we do X" notes that recur but don't warrant a full skill
.claude/memory/scaffold-journal.mdobserved quality of the scaffold itself (skills/agents/commands/hooks) — dogfooding wins/friction/coordination-gaps/missing-features; /scaffold-retro harvests it

(.claude/memory/policy/ also lives under memory — authored governance canon — but it's owned by the governance skill, not this one.)

Recall — when the user references earlier work

Triggers: "what did we do / decide on X", "in the last few sessions", "remind me", "previously".

  1. ls .claude/memory/sessions/ (dated names sort newest-last) and/or grep it for the topic.
  2. Read the relevant entry (or the latest 1–3) and answer from it. Don't re-derive what a note records.

Record — after substantive work

  1. Add .claude/memory/sessions/YYYY-MM-DD-<slug>.md from the template below — a refined summary, not a transcript (~one screen: outcomes + current state, not every step). Cross-link related sessions.
  2. Update roadmap.md: check off finished items, move them to Done (recent), add follow-ups.
  3. If you established a reusable "how we do X", add/adjust a reference/ note (or promote it to a skill).
  4. Scaffold check (self-observed — you were in the session): did any part of the .claude/ config help or hurt? A skill that didn't surface, an agent handoff that dropped context, a hook that fought you, a command you wished existed? Log a row to scaffold-journal.md per its schema. "None this session" is a valid, common answer — don't manufacture entries. (/scaffold-retro later clusters these into what to build; this step just captures them while specifics are fresh.)

Naming: ISO date + kebab slug (2026-01-15-project-bootstrap.md).

Session template (copy into the new file — keep these sections)

# Session: <title>

**Date:** YYYY-MM-DD · **Focus:** <one-line focus>

## Summary
<2–4 sentences: the goal of the session and what was actually accomplished.>

## Changes & artifacts
- <path> — <what changed / why>

## Key decisions
- <decision> — <one-line rationale>

## State
- <what works now> · <what's in progress> · <known gaps>

## Follow-ups
- <open thread> → ../roadmap.md

## Related
- <link to prior/related session, or "first session">

Branch management — the git workflow around a session

  • Branch per unit of work, off main. Don't commit straight to the trunk — cut a branch named for the work (e.g. <topic>-<slug>). Matches the harness rule: on the default branch, branch first. Trivial docs-only touch-ups are the only reasonable exception.
  • Commit only when the user asks. End every commit message with the project's required trailer(s), if any (<PLACEHOLDER — e.g. a Co-Authored-By line>).
  • Land it. "Land" = merge the branch into main (locally, or push + open a PR — <PLACEHOLDER: this repo's landing convention>). Record the branch name in the session note + roadmap up front (it's known before you commit); add the commit hash in a small follow-up commit or read it from git later.

Close-out (the wrap-up sequence)

Record the session note → update the roadmap → (commit) → land the branch. One coherent unit of work = one session note. Keep the roadmap status and the session prose from drifting. The /wrapup command drives this close-out end-to-end (record → optional commit → land) as a checklist — prefer it over reconstructing the steps from memory.

Gotcha

The .claude/memory/ notes are the DATA; this skill is the only source of the RECALL / RECORD / BRANCH rules — consolidated here (not in CLAUDE.md) so they surface exactly when you're starting, branching, or wrapping up work, and don't tax always-on context every session.

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