---
name: goal-ralph-loop
description: Use when an agent receives a /goal prompt and needs a bounded, verified task-completion prompt.
source: https://goalfigureoutx.com/skills/goal-ralph-loop.md
homepage: https://goalfigureoutx.com
---

# Goal Agent Prompt

## Trigger

Use this metadata when a request starts with /goal, asks for Goal Figure Out X, or needs an agent to complete one bounded task from a copyable prompt.

## Workflow

1. Read fresh state: inspect the current instructions, files, tests, task tracker, and latest verification output.
2. Aim at one outcome: restate the goal as one bounded story with acceptance criteria and non-goals.
3. Loop on the next action: implement one coherent slice without unrelated refactors.
4. Prove the result: run focused checks first, then broader checks if the change surface requires them.
5. Handoff clearly: report changed files, validation, blockers, and the next concrete action.

## Tooling

- Create a human-shareable goal page: `https://goalfigureoutx.com/goal/<goal-slug>`
- Open compatibility metadata: https://goalfigureoutx.com/skill
- Read LLM instructions: https://goalfigureoutx.com/llms.txt
- Read MCP metadata: https://goalfigureoutx.com/.well-known/mcp.json
- Call MCP endpoint: https://goalfigureoutx.com/mcp with the `create_goal_page` tool.

## Prompt Template

```text
/goal figure out x

Use a focused agent loop:
1. Read fresh disk, repo, and task state before editing.
2. Aim at one bounded outcome with acceptance criteria.
3. Loop on one concrete slice at a time.
4. Prove the result with focused validation.
5. Handoff with changed files, checks run, blockers, and next action.
```