Codex, with the decisions behind the code.

Retrieve requirements, previous work and open blockers before starting the next change.

Install in Codex

OpenAI Codex(v0.132.0)

model:gpt-5.6-sol low/modelto changedirectory:~/lobu-workspace

Lobu MCP ready · 6 tools · workspace Product

Continue the onboarding work from the last agent and report what ships.

I'll retrieve the decision and its sources, inspect the current code, then preview any consequential workspace action.

lobu.search_memory onboarding decision→ 1 decision · 3 sources

lobu.query_sdk read acceptance criteria→ exact source links retained

Read src/onboarding/ConnectAgent.tsx

Edited src/onboarding/ConnectAgent.tsx(+9 −2)

Ran bun test→ passed▸

bun test v1.2.21
✓ onboarding > connects an agent first
✓ onboarding > preserves linked sources
 18 pass
 0 fail

lobu.run_sdk save shipped result→ preview requires approval

Working(0s • esc to interrupt)

gpt-5.6-sol low · ~/lobu-workspace

The context is already there.

Lobu syncs your connected sources in the background. Search the live company model with its records, relationships and source events together.

How memory works

Your team's agents, too.

Codex can call shared organization agents with their own skills, identity and permissions. Use paired devices or sandboxed tools where access is granted.

Shared agents and access

Paste this into Codex.

One prompt connects the session. Prefer to wire it yourself? Run the skill installer instead and skip the prompt.

Connect this Codex session to Lobu, the shared company context layer for agents, over MCP: https://lobu.ai/mcp.

Onboard yourself:
- If you have a CLI: run `codex mcp add lobu --url https://lobu.ai/mcp`, then `codex mcp login lobu` and complete OAuth, then start a fresh Codex task so it loads the new tools.
- If you can't touch a terminal: ask the user to connect Lobu from the Codex Plugins settings, authenticate, and start a fresh task — then continue once connected.

Authenticate when prompted, then:
Read the Lobu skill at https://app.lobu.ai/api/skill/lobu, then run search_sdk for the smallest SDK surface you need. Do not dump the whole workspace.
Search existing memory before answering or saving a duplicate. Use search_memory/save_memory for durable context, query_sdk for reads, run_sdk for writes. Preview consequential writes first and keep every returned source URL, identifier, and approval status exact — never invent one, never fabricate a returned URL or identifier.

Open Codex

npx skills add lobu-ai/lobu --skill lobu