Skip to content

api-reference.md

Latest commit

 

History

History
121 lines (88 loc) · 3.75 KB

api-reference.md

File metadata and controls

121 lines (88 loc) · 3.75 KB
Raw

Grok Build Goal API Reference

Canonical reference for the /goal slash command and the update_goal agent tool.

/goal Slash Command

Set, manage, or check an autonomous goal. Grok works toward the objective across turns and reports progress.

Syntax

/goal <objective>
/goal status
/goal pause
/goal resume
/goal clear

Arguments

Argument Description
<objective> Natural-language goal. Include verifiable done conditions when possible.
status Show current goal state, recent progress, and whether goal mode is active.
pause Pause goal mode without clearing the objective.
resume Resume a paused goal.
clear Exit goal mode and clear the active objective.

Examples

/goal Migrate the auth module to the new API. Done when all auth tests pass and no imports from legacy/auth remain.
/goal status
/goal pause
/goal resume
/goal clear

Availability

/goal appears in the command list only when:

  1. The goal feature is enabled for your session, and
  2. The update_goal tool is in the session toolset.

update_goal Tool

The agent calls this tool to report progress without flooding the conversation. Humans typically use /goal status; the agent uses update_goal.

Parameters

Parameter Type Description
message string (optional) Short progress log. Visible in tool output, not surfaced to the pager dashboard.
completed boolean (optional) Set true only when the goal is fully achieved. Ends goal mode. Use with message for a completion summary.
blocked_reason string (optional) Set when truly stuck after 3+ consecutive failed attempts at the same problem. Pauses the goal as blocked. Never use for success.

Examples (agent-side)

Progress update:

{ "message": "Migrated token refresh; 12/18 auth tests passing" }

Completion:

{ "completed": true, "message": "All auth tests pass; legacy imports removed" }

Blocked:

{ "blocked_reason": "OAuth provider sandbox credentials expired; need human to rotate secrets" }

Rules of Thumb

  1. Call update_goal at meaningful milestones — not every tool call.
  2. Use completed: true only when the verifier agrees (tests, checklist, separate sub-agent).
  3. Use blocked_reason sparingly — after genuine exhaustion of alternatives, not on first error.
  4. Never put success text in blocked_reason.

Goal Mode Lifecycle

stateDiagram-v2
    [*] --> Idle
    Idle --> Active: /goal objective
    Active --> Active: agent turns + update_goal(message)
    Active --> Paused: /goal pause
    Paused --> Active: /goal resume
    Active --> Completed: update_goal(completed: true)
    Active --> Blocked: update_goal(blocked_reason)
    Active --> Idle: /goal clear
    Paused --> Idle: /goal clear
    Completed --> Idle: automatic
    Blocked --> Active: human unblocks + /goal resume
Loading

Related Commands

Command Relationship
/loop Scheduled recurring work — often feeds goals
scheduler_create Durable scheduling; pair with loops, not goals
Task (subagents) Launch verifier in worktree isolation
Skills Encode scoping, verification, and completion checks

External State Convention

Recommended project files when using goals:

File Purpose
GOAL.md Active objective, done condition, progress log, blockers
AGENTS.md Project rules including "always verify before completed: true"
.grok/skills/goal-verifier/ Skill that checks done conditions independently

See starters/minimal-goal/ and templates/GOAL.md.template.