About goal-loop
A goal-oriented agent skill that guides work through small actions, focused verification, and repeated critique. It defines completion criteria from the goal and repository context, limits rounds, and pauses before consequential operations.
Key features
- 01Infer completion criteria
- 02Choose the smallest useful action
- 03Verify each state-changing action
- 04Review remaining work each round
- 05Limit rounds with a default of five
- 06Pause before consequential operations
Use cases
- 01Completing repository tasks in defined rounds
- 02Checking changes with targeted tests or diffs
- 03Tracking blockers that prevent safe progress
- 04Stopping before commits, deployment, or irreversible actions
Install
Copy the command and run it in a terminal.
$
npx skills add https:/ / github.com/ darknesskiller/ skills --skill goal-loopAfter installing, type/goal-loopin Claude Code to confirm it loaded.
Bundled filesagents/openai.yaml
Details
From the project READMEGoal Loop
Steps
- Parse goal. Read
max_rounds(default5). Reject non-positive integer. - Infer completion criteria from goal and repo context. Record uncertainty as blocker only when it prevents safe next action.
- Choose smallest useful next action. Route through existing skill when one fits.
- Verify with narrowest useful check: test, diff, command output, artifact.
- Critique: what proves or disproves completion? What remains? Smallest next action?
- Criteria evidenced: stop
completed. max_rounds reached: stopmax_rounds. Else repeat from 3.
Rules
A round = one state-changing action + verification. Tool calls are not rounds. Do not spend rounds restating goal, polling without change, or speculative cleanup. Pause before commits, pushes, PR, deploy, production, credentials, irreversible ops.
Status
End with exactly one: completed, blocked (name unblocker), max_rounds (name remaining criteria).
Comments
Markdown supported