project-memory
Skill1 starsChanged 23 days agoMITView on GitHub
Creates and keeps the .builder/ project files (BRIEF, CURRENT, DECISIONS, LOG, LAUNCH, SPEND) that let any new session continue a build without the user re-explaining anything, and writes hand-off prompts for a fresh session. Use it when starting any build, at every gate or milestone, before ending a long session, when context is getting full, when the user says "new session", "hand off", "save where we are", "continue this later", or when anything is deferred "until launch".
- Reads credentials
27 other projects use a version of this.
Adopted by
Most recently taken up by
- roronoazoroshao369/vibe-coding-ostheir own version, first seen Nov 15, 2026
- Sdamirsa/dot-claude-ifftheir own version, first seen Nov 15, 2026
- CheehooLabs/SpureeSkillstheir own version, first seen Nov 8, 2026
- cobibean/project-memory-skilltheir own version, first seen Nov 8, 2026
- RRFRRF/project-memory-skilltheir own version, first seen Nov 8, 2026
- xuyewild/multi-ai-skillstheir own version, first seen Oct 25, 2026
Other versions27 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 27
- SpillwaveSolutions/project-memory★ 88same name
- CheehooLabs/SpureeSkills★ 18same name
- RRFRRF/project-memory-skill★ 8same name
- scaleto/Antigravity-Super-Skill-Architecture★ 8same name
- cobibean/project-memory-skill★ 7same name
- me-cedric/agent-compass★ 3same name
- tasuku-9/project-memory-skill★ 3same name
- jukrap/ai-agent-playbook★ 2same name
- sakuruamayday-ops/enterprise-project-application-assistant★ 2same name
- Sdamirsa/dot-claude-iff★ 2same name
17 more not listed
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 repository7 other files it gives its agents.
- appSkill
- backend-loginSkill
- builder-setupSkill
- buildSkill
- researchSkill
- voice-agentSkill
- websiteSkill
The file
We split this skill into the rules it gives an agent
12 rules. 0 are common, 10 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 22 edited versions we read keep these.
10
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
- could be automatic: a setting could enforce it
- Very specific: line 15
Create them from templates/ the first time (copy the folder, then fill in BRIEF.md from stage 1).
- Distinctive: line 32
Who decided what is sacred: Only *client* and *user* decisions can be locked.
- Distinctive: line 36
Record observations, not guesses: If something broke and the cause isn't proven, write "cause unknown".
- Distinctive: line 38
Check paths before writing a hand-off: Every file a hand-off mentions must exist.
- Very specific: line 39
Commit before risky work: if the project uses git (git add -A && git commit -m "checkpoint: <gate>").
- Distinctive: line 43
When context is getting full, the work changes phase, or the user asks for a new session:
- Distinctive: line 44
Update CURRENT.md (especially Next step) and LOG.md.
- Distinctive: line 52
When the hand-off is for another AI tool (Codex, Cursor), add: "Work only in <folder or branch>.
- Distinctive: line 53
Before finishing, list every file you changed."
- Distinctive: line 56
If two sessions (or another AI tool, like Codex or Cursor) edit the same folder, they overwrite each other's work.
- Distinctive: line 58
Check git status for changes this session didn't make.
- Distinctive: line 61
Never deploy files this session didn't change without saying so.
“Common” is measured: kept by at least 6 in 10 of the 22 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
Project memory
In the builds this kit was made from, users re-explained things Claude should already have known 4–17 times per project. Once a project kept these files and Claude wrote its own hand-off prompts, that dropped to zero in every build that used them. This skill is that system.
The files (.builder/ in the project root)
Create them from templates/ the first time (copy the folder, then fill in BRIEF.md from stage 1). At the same time, keep
them private: add .builder/ and .env* to .gitignore, and to .vercelignore if the project deploys to Vercel. BRIEF.md
can hold a client's phone number and budget, so it must never be published. The publish check reminds you if it's missing.
| File | What goes in it | Update when |
BRIEF.md | The approved Task Card: goal sentence, audience, priority order, first real use, domain/accounts, caps | Stage 3 approval; any scope change the user approves |
CURRENT.md | What's true right now: live URLs, what works, what's in progress, Next step, who owns which folder | Every gate, every deploy, end of session |
DECISIONS.md | One row per decision: what · who decided (client / user / Claude-proposed) · date · locked? | Whenever something is decided |
LOG.md | Dated, newest first: what was done, what was learned, what broke and the real cause (or "cause unknown") | Every gate |
LAUNCH.md | Every "later", "before launch", "when we go live" item, as checkboxes, each marked blocker or not | The moment anything is deferred |
SPEND.md | Paid services: balance · cap · spent so far | Before and after any paid call (the spend guard checks this file) |
RESEARCH.md | Research findings (written by the research skill) | Stage 2 |
The session-start hook reads CURRENT.md and the open LAUNCH.md items into every new session automatically.
Rules that matter
- Who decided what is sacred. Only client and user decisions can be locked. A Claude-proposed decision is a default that can be revisited. In the data, Claude's own design opinions got written down as "settled", and a passing remark became "the client asked for no menu bar". Write who actually said it, with a date or quote.
- Inferred goals are labelled
inferred, not confirmeduntil the user confirms them. - Record observations, not guesses. If something broke and the cause isn't proven, write "cause unknown". A wrong cause in memory misleads every later session.
- Check paths before writing a hand-off. Every file a hand-off mentions must exist. One hand-off named a file that didn't.
- Commit before risky work if the project uses git (
git add -A && git commit -m "checkpoint: <gate>"). This gives the user an undo point. Never commit.envfiles or secrets. Check.gitignorefirst.
Hand-off to a new session
When context is getting full, the work changes phase, or the user asks for a new session:
- Update CURRENT.md (especially Next step) and LOG.md.
- Give the user this block to paste into the new session:
Continue the <project> build. Read .builder/CURRENT.md, then .builder/BRIEF.md and the open items in
.builder/LAUNCH.md. Next step: <one line>. Don't re-ask anything those files answer.Tell the user exactly how to start it: "Open a new Claude Code session in this folder and paste the block above."
When the hand-off is for another AI tool (Codex, Cursor), add: "Work only in <folder or branch>. Don't touch <list>.
Before finishing, list every file you changed."
Several sessions or tools in one project
If two sessions (or another AI tool, like Codex or Cursor) edit the same folder, they overwrite each other's work. This happened in 7 of 14 builds. Before starting:
- Check
git statusfor changes this session didn't make. If there are any, ask whose they are before touching them. - Give each session its own git worktree (
git worktree add ../<project>-<task> -b <task>), or record in CURRENT.md which folder each session owns. - Never deploy files this session didn't change without saying so. The publish guard warns about this.
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.

