Goal Command

Turn /goal into a persistent execution loop instead of a planning prompt.

Goal Command is an OpenClaw plugin that adds a closed-loop /goal command. It creates a durable run directory, injects execution guidance into the next agent turn, and forces the run toward exactly one terminal state: DONE, BLOCKED, or FAILED.

Why It Exists

Normal agent chats often stop at:

• “Here is the plan”
• “Ready to execute”
• “Let me know if you want me to continue”

That is not enough for objective work. Goal Command makes the run state explicit and persistent so the agent must execute, validate, and close the loop.

What It Does

When you run:

/goal fix the broken checkout test and validate it passes

The plugin creates:

goals/runs/<timestamp-slug>/
├── goal.md
├── status.md
├── feature_spec.md
├── plan.md
├── validation.md
└── decision_log.md

For concrete goals it sets:

state: EXECUTION_READY
nextAction: execute-now

The injected prompt then tells the agent:

• EXECUTION_READY means execute now, not explain
• do not stop after creating a plan
• update status to DONE, BLOCKED, or FAILED
• validate before DONE
• write blockers/evidence into run files

Commands

/goal <objective>
/goal status [runId]
/goal resume [runId]
/goal sync [runId]

/goal cancel is currently reserved for a future version.

Abstract vs Concrete Goals

The plugin classifies vague goals as NEEDS_SPEC and starts a short spec-coach flow first.

Examples:

/goal improve my app

→ asks blocking spec questions first.

/goal update the landing page headline and verify build passes

→ starts execution mode.

Obsidian / Project Notes

By default, synced notes go under:

<agent-workspace>/OpenClaw

Optional config:

{
  "goalCommand": {
    "obsidianRoot": "OpenClaw"
  }
}

obsidianRoot may be workspace-relative or an absolute path.

Safety Notes

• The plugin writes markdown run files inside the agent workspace.
• /goal sync writes markdown project/goal notes under the configured Obsidian root.
• It does not execute shell commands by itself.
• External/destructive work still depends on the agent’s normal approval and safety rules.
• DONE requires validation evidence; effort alone is not completion.

Best Used For

• multi-step coding tasks
• debugging chains
• project cleanup
• publishing/checklist work
• anything where stopping at “plan created” would be failure

by brasco05 · built for OpenClaw closed-loop execution