Goal Engineering

Verifiable run-until-done objectives for Grok Build /goal. Patterns, a verifier, and the goal-audit CLI.

View the Project on GitHub cobusgreyling/goal-engineering

Primitives Matrix — Goal Mode Across Tools

Tool-agnostic goal design: the capability matters, not the product name.

Primitive Job in a Goal Grok Build Claude Code Codex Cursor
Set objective Define run-until-done work /goal <objective> /goal /goal Agent + explicit stop condition in prompt
Progress telemetry Log without chat noise update_goal tool Goal progress APIs / state files Goal mode + state Hooks + state files
Pause / resume Human control /goal pause / resume /goal pause / resume Pause/resume Manual session control
Verifier Independent done check goal-verifier skill + Task subagent Reviewer agent Verifier subagent Review mode / second agent
State Survive compaction GOAL.md GOAL.md / progress files Markdown state GOAL.md
Worktrees Safe parallel attempts Task with worktree isolation git worktree Built-in worktree Cloud agent worktree
Skills Scoping + verification .grok/skills/ .claude/skills/ .codex/skills/ .cursor/skills/

Scheduling vs Goals

Use case Grok Claude Code Codex
Run until tests pass /goal all tests pass /goal /goal
Daily triage /loop 1d /loop Automations
Hand off triage → finish Loop writes STATE.md → /goal top item Same Same
/goal <objective with done condition>
  + GOAL.md (state)
  + goal-verifier skill (checker)
  + AGENTS.md (project rules)
  + npm test / pytest (objective gate)

Audit after setup: npx @cobusgreyling/goal-audit . --suggest