/goal Runs
Use /goal when the human intent is larger than one task. The plan stays in todos.md, and the agent uses the local CLI or MCP tools to claim work, update state, record evidence, and report completion.
Install
bun install -g @hasna/todosExecution model
| Surface | Role |
|---|---|
| Tasks | The concrete work items an agent can claim, start, comment on, verify, and complete. |
| Projects | The workspace boundary for related tasks, plans, runs, comments, and artifacts. |
| Plans | The ordered intent: acceptance criteria, dependencies, checkpoints, and verification steps. |
| Runs | The execution trail: status, logs, artifacts, receipts, and final proof. |
Codex recipe
Create a project
Project
todos projects create launchAdd tasks
Task
todos add "Ship install page" --project launchTask
todos add "Verify CLI quickstart" --project launchAttach a plan
Plan
todos plans create launch --from PLAN.mdRun the goal
In Codex, point the session at the plan and require recorded verification before completion.
Codex
/goal execute launch --source todosReview evidence
Runs
todos runs list --project launchTask detail
todos inspect <task-id>Claude Code recipe
Connect Claude Code through the MCP gateway, then ask it to execute the same plan. Claude should read the project, claim one task at a time, run verification, and write the final task comments before marking work done.
MCP catalog
todos docs catalog --surface mcpClaude Code
/goal execute launch with todos MCPWhat a good run leaves behind
- Every claimed task has a status transition and an owner.
- Every completed task has verification commands or artifact references.
- The plan shows what is done, blocked, or still pending.
- The run record preserves logs, receipts, and final status for review.