| name | goal |
| description | Run or manage a Codex-style persistent goal loop in Cursor Agent chat using goal or cursor-goal checkpoint helpers. Use for /goal, /goal pause, /goal resume, /goal clear, /goal edit, durable objectives, verification loops, and evidence-based completion. No Agent SDK or API key required. |
| disable-model-invocation | true |
Goal Skill (Cursor)
Use this skill when the user invokes /goal or asks for a Codex-style persistent goal loop.
Run the loop in this Cursor chat session. You are the agent. Do not spawn @cursor/sdk or external agent APIs.
Use goal or cursor-goal for durable state, verification, and checkpoint accounting.
Command mapping
/goal → goal status or read .goal/current.json.
/goal <objective> → set state, then work the goal in this chat.
/goal pause | /goal resume | /goal clear | /goal edit <objective> → matching CLI subcommands.
goal "<objective>" --verify "npm test"
Checkpoint loop (each turn)
- Load active goal (
goal status). If missing on /goal <objective>, set with goal.
- If status is not
active, report and stop unless the user asked to resume.
- Optionally run
goal prompt for the continuation contract.
- Make one bounded checkpoint of concrete progress.
- End your response with exactly:
GOAL_STATUS: COMPLETE | CONTINUE | BLOCKED
GOAL_REASON: <one short evidence-based sentence>
- Record:
goal checkpoint --tool-calls <n> (pipe assistant text on stdin if needed).
- Continue while
active until complete, blocked, budget_limited, or user pauses.
Safety
- No destructive verification unless explicitly requested.
- Do not expose secrets.