memory-bank
Skill0 starsChanged 24 days agoMITView on GitHub
Keep a git-tracked markdown memory bank (memory/ folder) so the agent resumes work across sessions. Use at session start, when a task finishes or a blocker appears, when a tool or documentation problem needs logging, when a significant technical decision is made, before ending a session with unfinished work, and when asked to set up or reset the memory bank for a project or branch.
8 other projects use a version of this.
Adopted by
Most recently taken up by
- thistleknot/skillstheir own version, first seen Nov 8, 2026
- rstudio/shinytheir own version, first seen Nov 1, 2026
- posit-dev/py-shinytheir own version, first seen Oct 17, 2026
- nek1987/auto-agent-harnesstheir own version, first seen Oct 11, 2026
- code-with-mehedi/nextjs-claude-startertheir own version, first seen Aug 25, 2026
- fockus/skill-memory-banktheir own version, first seen Jul 28, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- rstudio/shiny★ 5.7ksame name
- posit-dev/py-shiny★ 1.8ksame name
- fockus/skill-memory-bank★ 32same name
- nek1987/auto-agent-harness★ 4same name
- josiahcrackle9631/memory-bank★ 1same name
- code-with-mehedi/nextjs-claude-starter★ 0same name
- thistleknot/skills★ 0same name
- tsbhatia/coding-skills-for-agents★ 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.
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 5 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 8 edited versions we read keep these.
5
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
- Very specific: line 23
Create missing files in memory/ from the templates/ folder next to this SKILL.md, including friction/README.md.
- Distinctive: line 42
Do not rewrite these files on every turn.
- Very specific: line 58
Commit memory updates at milestones, with the feature commit or in a memory: commit.
- Distinctive: line 59
Never track per-developer local chores (worktree cleanup, local setup) here: they only create merge noise.
- Distinctive: line 64
When the user starts a new branch or worktree and asks to initialise memory:
- Distinctive: line 70
Commit the initialisation in the branch's first commit.
- Distinctive: line 74
If the project also has a semantic or long-term memory tool (for example jev-mem), keep the two from overlapping:
“Common” is measured: kept by at least 6 in 10 of the 8 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 bank
Project state lives in plain markdown files in memory/, tracked in git. Because the folder is tracked, every
branch and worktree carries its own version: context follows the branch, shows up in PR diffs, and needs no service.
Files
| File | Role | Read at session start? |
memory/progress.md | Macro checklist: what is done, what is open, known bugs. | yes |
memory/activeContext.md | Scratchpad: current focus, blockers, files touched, the exact next step. | yes |
memory/decisions.md | Append-only log of significant technical decisions and why (newest first). | no, search it when needed |
memory/archive.md | Finished history moved out of the two files above. | no |
memory/friction/ | Maintenance log of tool/doc/workflow problems to fix later; separate from project memory. | no |
Setup and missing files
Create missing files in memory/ from the templates/ folder next to this SKILL.md, including
friction/README.md. Check individual files even when memory/ already exists (for example, an installer may
have created only friction/). Preserve every existing file; never replace notes with templates. Fill the "Branch"
and "Current focus" lines of a newly created activeContext.md from what you know, and tell the user what you
created. If you cannot locate templates/, create missing files with the headings described in this document.
Session start
- Set up any missing memory-bank files as above, then read
memory/activeContext.md. - Read
memory/progress.md. - Continue from "Next steps". If they look stale against the code or
git log, say so and ask before acting on them.
During work: update only on these triggers
progress.md: a task is completed ([ ]to[x]), or you discover a new task or bug.activeContext.md: focus changes, you hit a blocker, or you conclude a step (touched files, state, exact next step).decisions.md: a significant technical, structural or dependency choice. Append at the top: context, decision, rationale.friction/: a tool errors, a doc misleads, or you need a workaround. Add one small dated file usingmemory/friction/README.mdand keep working. This is a maintenance backlog, not session context; also record current blockers inactiveContext.md.
Do not rewrite these files on every turn. Small, trigger-based edits keep the diffs reviewable.
Keeping it lean
activeContext.md is a scratchpad, not a log: roughly the current thread, its blockers and next steps, plus a short
"just changed" list. When a thread concludes, move it to archive.md under a dated heading. progress.md lists open
items plus the in-flight feature's Done list; move a finished feature's Done list to archive.md once nothing in it is
likely to be referenced again. Never delete from decisions.md or archive.md.
Hand-off (before finishing a turn that leaves work unfinished)
- Checkbox state in
progress.mdis accurate. activeContext.mdnames the exact next step, so a fresh session can start without asking.
Version control
- Commit memory updates at milestones, with the feature commit or in a
memory:commit. Do not leave the tree permanently dirty. - Never track per-developer local chores (worktree cleanup, local setup) here: they only create merge noise.
- On merge, keep the feature branch's
progress.mdandactiveContext.md; mergedecisions.mdby keeping both sides' entries, newest first.
New feature branch
When the user starts a new branch or worktree and asks to initialise memory:
progress.md: new section for the feature, empty Done list, carry forward only still-relevant Open items.activeContext.md: rewrite with the branch name, the new focus, no blockers, only relevant next steps.decisions.md: leave untouched.
Commit the initialisation in the branch's first commit.
Alongside a semantic memory (optional)
If the project also has a semantic or long-term memory tool (for example jev-mem), keep the two from overlapping:
- Current state (focus, in-progress work, blockers, next steps) belongs here, never in the other tool, where it would go stale silently.
- Dated facts that stay true (decisions with reasons, bug causes, gotchas, user preferences) belong in the semantic memory. A significant decision gets both a
decisions.mdentry and a one-fact note there. - Rules the agent must always follow belong in
AGENTS.md/CLAUDE.md.
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.

