Skip to content

topoteretes/cognee / .agents/skills/multiagent-memory-setup

multiagent-memory-setup

Unverified

Import the user's own Claude Code and/or Codex session history (the prompts they typed and the agents' answers) into one cognee dataset, then answer questions over it, across every agent and per agent. Use when someone wants to load, import or backfill their previous coding-agent sessions into cognee, give several coding agents one shared memory of past work, or ask what their earlier sessions were about. Runs locally; reads ~/.claude and ~/.codex read-only, writes only to cognee's memory.

Import coding-agent sessions

Runs the cookbook in examples/cookbooks/import_agent_sessions/. It first finds the coding agents installed on this machine and the sessions each has kept. One LLM key (LLM_API_KEY, OPENAI_API_KEY or ANTHROPIC_API_KEY) runs cognee and imports every agent's history. The chosen agents go into the dataset imported_agent_sessions, and each question is answered over all of them. Setup, the scripts and what each one does are in the cookbook's README.md.

Run every command from the repo root.

Demo

When the user has no Claude Code or Codex history, or wants to see it work first, run the demo instead of the steps below. It needs one key and imports sample sessions of both agents (a made-up web shop), never the user's own. It is what the cookbook runs without --agents: uv run python examples/cookbooks/import_agent_sessions/import_agent_sessions.py (--check first to check the key; exit 2 with a MISSING: line means no key is set). Give the user the [ask] answers and tell them they come from sample data. Pass --ask "..." for their own question.

Steps

  1. Run uv run python examples/cookbooks/import_agent_sessions/import_agent_sessions.py --check.
    • Exit 2: each [setup] MISSING: line names one fix. Tell the user, using the README's "What it needs" table, and stop. Never ask the user to paste a key into the chat.
    • Exit 0: go on.
  2. Tell the user what was found, from the [find_agents] lines: each agent, how it is installed, and how many sessions it has and where they were started. Name each Not imported: app (Claude desktop chats, Cursor's own agent, ChatGPT) and why its chats can't be read.
  3. Ask the user which agent sessions to ingest into cognee, with your question tool when you have one: one choice per agent the OK: line names, several allowed. Skip asking only when the user already named the agents. Don't choose for them, and never leave --agents out of the import: without it the run is the demo on sample data.
  4. Ask the user which projects to import, from the [find_agents] Project: lines, with "All projects" as one choice. Every imported session is sent to the LLM provider, and a machine often holds work the user can't share, so never assume all. When there are more projects than your question tool shows, list them and ask the user to name theirs. Skip asking only when the user already named the projects.
  5. Run the import with --agents set to the user's choice, like --agents claude,codex, and one --project <path> per chosen project, copying the path from its Project: line (leave --project out for all projects). A project includes the folders inside it. Start with the default --limit 20 unless the user asked for more, and pass --since YYYY-MM-DD for a period. Pass their question with --ask "..."; without one the run asks "What did I work on in these sessions?": uv run python examples/cookbooks/import_agent_sessions/import_agent_sessions.py --agents <choice> --project <path> --ask "...". It takes a few minutes for 20 sessions.
  6. The answers are on the [ask] lines: all sessions> over everything, and with two agents one <agent> only> line each. Give them to the user.
  7. Later questions don't need a new import: uv run python examples/cookbooks/import_agent_sessions/scripts/ask.py "..." --agents claude,codex (list only the agents the user chose).
  8. If a step fails, the run exits 1 and the traceback's last line says why. Run that script from examples/cookbooks/import_agent_sessions/scripts/ on its own to look closer.

Rules

  • Session history is the user's private data. Don't quote it beyond what answers the question, and don't print .env.
  • Every imported session is sent to the provider of the key cognee runs on and costs LLM calls, so run the import only when the user asked for it, and raise --limit only when they want more history.
  • Pass --clear only when the user asked to start over: it forgets everything the dataset holds. It never touches ~/.claude or ~/.codex.
  • Redaction catches common key formats and secret-named values (DB_PASSWORD=...) only. If the user is about to import a lot, suggest they look at a few files in examples/cookbooks/import_agent_sessions/sessions/ first.

Clean up

uv run cognee-cli forget --dataset imported_agent_sessions
uv run cognee-cli forget --dataset imported_agent_sessions_demo
rm -r examples/cookbooks/import_agent_sessions/sessions examples/cookbooks/import_agent_sessions/sample
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 463kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 318kUpdated Claude CodeCodexUnverifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 318kUpdated Claude CodeCodexUnverifiedInstall