🧠 agent-memory-keeper
Persistent project memory (MEMORY.md) so your agent stops repeating the same mistakes. A free, open-source Agent Skill for Claude Code, Cline, Cursor, Codex and 75+ AI agents.
Install agent-memory-keeper
# Claude Code (all 51 skills)
/plugin install all-skills --marketplace milad-ahmd/agent-skills
# Cursor, Codex, Cline, Windsurf, Copilot + 75 more agents
npx skills add milad-ahmd/agent-skills --skill agent-memory-keeper
# Shell installer
curl -fsSL https://raw.githubusercontent.com/milad-ahmd/agent-skills/main/install.sh | bash -s -- agent-memory-keeperIncluded in packs: startup-cto-pack, developer-pack, skill-builder-pack
What to say to your agent
Remember that we use pnpm, not npmWhat do you remember about this project?
When it triggers
Gives your coding agent a persistent project memory. It maintains a concise MEMORY.md of verified commands, conventions, decisions, gotchas, and user preferences, reads it at the start of work, and updates it whenever the user corrects the agent or a lesson is learned, so the same mistakes are not repeated across sessions. Use at the start of any session in a project with MEMORY.md, or when the user says "remember", "don't forget", "you did this wrong again", "what do you remember", or "forget".
Full skill instructions (SKILL.md)
Agent Memory Keeper 🧠
Stop re-learning the same project every session.
When to read
At the start of every task in a repo, read MEMORY.md at the repo root if it exists. Obey it unless the user says otherwise.
When to write (immediately, without waiting to be asked)
- The user corrects you ("no, we use pnpm", "don't touch generated files").
- You discover a verified command (how to test, build, lint or run locally) or a gotcha (a flaky test, a required env var, an order of operations).
- A decision is made, with the reason for it.
- The user states a preference (code style, commit style, review strictness, language).
- The user says "remember …".
Format (keep it under ~150 lines; update entries instead of appending duplicates)
# Project Memory
_Last updated: 2026-10-25_
## Commands (verified)
- Test: `pnpm test -- --run` (vitest; plain `pnpm test` starts watch mode)
## Conventions
- API handlers live in `src/routes/*`, one file per resource
## Decisions
- 2026-10-12: Chose Zod over Yup: already used in the frontend
## Gotchas
- `DATABASE_URL` must point at the docker DB, or migrations hang
## Preferences
- Conventional Commits; never push without asking
## Open threads
- Migrating auth to v2: halfway, see `docs/auth-v2.md`Commands the user can say
- "What do you remember?" → summarise MEMORY.md.
- "Forget X" → remove the entry and confirm.
- "Memory cleanup" → merge duplicates, delete stale entries, re-verify the commands.
Integration
- Claude Code: add
@MEMORY.mdtoCLAUDE.mdso it loads automatically. - Cline / Cursor / Codex: reference it from
.clinerules,.cursor/rulesorAGENTS.md. - Team memory: commit the file. Personal memory: add it to
.gitignore. Ask the user which they prefer the first time.
Rules
- Never store secrets, tokens, passwords or personal data.
- Only record facts that have been verified, or that the user stated. Mark anything uncertain with
(unverified). - Keep entries short, specific and dated. Memory that is too long gets ignored.
View source on GitHub ⭐ Star the collection