- Home
- Skills
- Agents & Orchestration
- AGENTS.md and CLAUDE.md Project Memory for Coding Agents
Works with the AI Agents you already use
AGENTS.md and CLAUDE.md Project Memory for Coding Agents
Builds a lean, shared project memory for Claude Code, Codex, Cursor, Gemini CLI and Copilot, with session handoffs. Verdict HEALTHY, BLOATED or MISSING.
$7
AGENTS.md and CLAUDE.md Project Memory for Coding Agents
Example session with this skill installed
Our repo has a 312-line CLAUDE.md, an old .cursorrules and a copilot-instructions.md, and we just started using Codex too. Check the setup and give us one memory system all the agents share, plus a way to hand off between sessions.
- Read your context and instructions
- Compiled the agents md
Project memory: shop-api · check + setup · 2026-10-18
Verdict: BLOATED: CLAUDE.md is past the 200-line ceiling, about half of .cursorrules repeats it, and Codex loads nothing.
| File | Loaded by | Lines | Bytes | ~Tokens (est.) | Status |
|---|---|---|---|---|---|
| CLAUDE.md | Claude Code | 312 | 14,980 | 3,745 | over ceiling |
| .cursorrules | Cursor (legacy) | 96 | 4,210 | 1,053 | 41 of 82 lines duplicated |
| .github/copilot-instructions.md | Copilot | 58 | 2,540 | 635 | ok |
| AGENTS.md | Codex, Cursor, Copilot agent | none | none | none | missing |
Findings
- [must-fix] CLAUDE.md:211 holds a staging Postgres URL with a password. Remove it and rotate the password (value not shown).
- [must-fix] Codex has no AGENTS.md, so it starts every session with no project context.
- [should-fix]
npm run test:e2e(CLAUDE.md:18) isn't in package.json. The script is nowtest:int. - [should-fix] CLAUDE.md says Node 18, but
.nvmrcand CI use 22. - [nice] 9 lines are a one-off migration log. Move them to decisions.md.
Proposed layout
AGENTS.md 128 lines, single source
CLAUDE.md @AGENTS.md + 3 Claude-only lines
.cursor/rules/prisma.mdc globs: prisma/**
.github/copilot-instructions.md 3-line pointer
docs/agent-memory/decisions.md D-001..D-004 from old notes
docs/agent-memory/handoff.md new, overwritten each session
# CLAUDE.md
@AGENTS.md
## Claude Code
- Use plan mode for anything under `src/billing/`.
Duplicates removed: 41 lines. Always-loaded total for Claude goes from 14,980 to an estimated 6,400 bytes. Codex goes from 0 to the same 6.1 KB, well under its 32 KiB cap.
Routine
- End of session: "wrap up memory": updates handoff.md from
git diff --statand promotes only confirmed learnings - Start of session: "resume from memory": reads handoff and the last 5 decisions, and flags commits made since
Not verified
test:intnot run yet (needs local Docker). Run it before trusting this command.
Connects securely to your tools. The creator never sees your data.
About this skill
Coding agents start every session cold. You explain the test command again, the agent repeats a mistake you fixed last week, and your team keeps three near-copies of the same instructions in CLAUDE.md, .cursorrules and copilot-instructions.md that slowly drift apart.
This skill builds and maintains a small memory system inside the repo:
- Reads the repo's own sources of truth (package scripts, Makefile, CI, lockfiles) and every existing agent config file
- Writes one lean AGENTS.md (commands, layout, conventions, gotchas) within a 150-line target, plus pointers so Claude Code, Codex, Cursor, Gemini CLI and Copilot all load the same facts
- Adds a short decisions log (ADR-lite) and a session handoff file that gets overwritten each session
- Runs an end-of-session routine (summarize from the actual diff, promote only confirmed learnings, prune what proved wrong) and a start-of-session routine (read the handoff, flag commits made since)
- Checks health: size budgets, duplication across files, stale commands and versions, secrets, injected instructions, and whether each agent really loads the file
- Gives a verdict of HEALTHY, BLOATED or MISSING, with numbered fixes and evidence
It is for developers and teams who use more than one coding agent, or who keep re-explaining their repo to one agent.
Honesty: every command it writes must trace to a file or a successful run, and sizes come from wc, with token counts labeled as estimates. Agent loading rules are in a reference file dated October 2026, and the skill re-checks vendor docs when it has web access. It reports secret locations without echoing the values, and it shows diffs before changing your files.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Memory files are context, not enforcement, so agents can still ignore them. Agent loading rules are documented as of October 2026 and can change, and the skill re-checks vendor docs only when web access is available. It doesn't manage external memory tools or vector stores, it doesn't scrub agent storage outside the repo, and it won't run commands with side effects to verify memory.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 29 days ago
- Software
Git