- Home
- Skills
- Agents & Orchestration
- Agent Memory OS
Works with the AI Agents you already use
Agent Memory OS
Gives Cursor, Claude Code, Copilot and Gemini one shared memory per repo. Installs a small .ai/ layer every agent reads before working and updates before stopping — no re-explaining your codebase.
$5
Agent Memory OS
Example session with this skill installed
Use the agent-memory-os skill to set up shared memory for this repo.
- Read your context and instructions
- Compiled the agent memory os
I have completed the context audit and initialized the shared memory layer. The audit found significant duplication between your .cursorrules and CLAUDE.md files.
Agent Context Audit & Memory Scaffolding
| # | Savings | Action | Location |
|---|---|---|---|
| 1 | ~450 tk | Delete duplicate linting rules | .cursorrules:12 vs CLAUDE.md:45 |
| 2 | ~320 tk | Move deployment steps to a separate skill | CLAUDE.md:80-120 |
| 3 | ~150 tk | Remove stale reference to deleted /docs/api-v1 | .cursorrules:5 |
Shared Memory Layer Created:
I have scaffolded the .ai/ directory with the following state:
.ai/context.md: Project scope and stack detected frompackage.json..ai/architecture.md: Module map based on directory structure..ai/memory.md: Initial conventions and gotchas..ai/session.md: Current state: "Memory layer initialized; audit pending application."
Next steps
- Run
python scripts/audit.py . --apply-lessto remove the identified duplicates. - Review
.ai/architecture.mdto ensure the detected module boundaries are correct. - Instruct your next agent session to "Read .ai/session.md to begin."
Connects securely to your tools. The creator never sees your data.
About this skill
Your agent forgets everything between sessions and completely when you switch tools. You re-explain the stack, re-list the conventions, re-read the same files. Then you open a different agent and do it again.
Agent Memory OS installs a shared memory layer in the repository itself: five plain markdown files in .ai/ that every agent reads before it works, and updates before it stops.
WHAT IT INSTALLS
.ai/context.md what the project is, the stack, what is out of scope .ai/architecture.md layers, boundaries, module map, external services .ai/decisions.md append-only log of choices that are expensive to undo .ai/memory.md conventions, gotchas, module knowledge .ai/session.md current state and the handoff record between agents
Plus the boot block for your rules file (AGENTS.md, CLAUDE.md, .cursorrules instructions for all three) and five starter templates.
WHY THIS ISN'T JUST "WRITE A CLAUDE.MD"
Three mechanisms do the work:
-
An init-proof line. The agent's first output must be
Init Complete: Read Session [Yes]…. If it skipped the bootstrap, you see it in the transcript instead of finding out three hours later. -
A verify-by command in every handoff. Work moves between agents only with the exact command that proves the state is real, so an agent never builds on an unverified handoff.
-
One home per fact. A rule that lives in two files eventually contradicts itself and when it does, your agent follows whichever copy it read last, so behaviour changes at random.
WHAT YOU GET
- SKILL.md — install procedure, boot block, file triggers, search-order ladder, verification checklist, and troubleshooting for the five failures that actually happen
- references/handoff.md handoff schema, resume procedure, multi-agent routing, conflict cases
- references/memory-rules.md what belongs in which file, freshness rules, pruning, migration for existing repos and monorepos
- assets/ five starter files, ready to fill in
PERMISSIONS
Reads and writes markdown. No shell, no network access, no secrets, no build step. The only files it creates live under .ai/ plus the boot block in your existing rules file.
LICENCE
The OmniAgent-OS repository is MIT. This packaged skill is licensed to you for use in unlimited projects, including client work. It may not be resold or redistributed.
BEST FOR
Anyone running more than one agent on one repo, long-running projects, and anyone who has watched an agent confidently break something a previous agent had already fixed.
NOT FOR
Throwaway scripts, or repos small enough that an agent can read the whole thing in one pass.
Read-only by default. UI/UX search requires local Python 3.x. Memory layer requires agents to follow a specific read/write protocol. Audit token savings are estimates based on standard encoding.
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 any payment.
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 21 days ago
Requires
Permissions
File Scopes
Creates only markdown under .ai/ plus the boot block in the existing rules file. No shell, no network, no secrets, no build step. No account to connect.