Create Goal

Create Goal

Create a durable Codex /goal contract from spec-kit artifacts.

This skill is the Merlin-owned bridge from Spec Kit artifacts to Codex /goal. It keeps AB Method's useful idea of a persistent GOAL.md, but replaces AB Method project structure with spec-kit feature directories.

When To Use

Use when:

• the user wants to run Codex /goal;
• a feature has spec.md, plan.md, and tasks.md;
• autonomous implementation needs a stop condition, verification loop, and stable source-of-truth list.

Do not use when the idea has no spec yet. Route to speckit-specify or brainstorming first.

Locate The Feature Directory

Find the active feature directory in this order:

1. .specify/feature.json field feature_directory;
2. environment variable SPECIFY_FEATURE_DIRECTORY;
3. the only or newest specs// directory;
4. a branch-matching specs/-/ directory.

If multiple plausible directories exist, choose the one most clearly referenced by the user or current branch. Ask only if choosing would materially risk writing the wrong goal.

Required Inputs

The feature directory must contain:

• spec.md
• plan.md
• tasks.md

Read these if present:

• research.md
• data-model.md
• contracts/
• quickstart.md
• checklists/
• memory/constitution.md
• .specify/memory/constitution.md

Refusal Conditions

Do not write GOAL.md if:

• no feature directory can be found;
• any required input is missing;
• spec.md contains unresolved [NEEDS CLARIFICATION];
• plan.md lacks a concrete technical approach or verification plan;
• tasks.md is not executable as a checklist;
• constitution or checklist gates are unresolved.

Report the blocking files and exact fixes needed.

Output Path

Write or update:

specs//GOAL.md

Do not write a separate docs/goals/ tree for spec-kit projects.

GOAL.md Contract

The goal must include:

1. Title and short objective.
2. Source-of-truth artifact list with relative paths.
3. Success criteria tied to spec.md.
4. Execution rules tied to tasks.md.
5. Implementation plan summary tied to plan.md.
6. Verification loop: tests, quickstart, Playwright/browser proof, QA, review.
7. Progress protocol: mark tasks in tasks.md, keep evidence current.
8. Safety boundaries: do not change unrelated scope, preserve user-story independence, follow constitution.
9. Stop condition: all tasks complete, checks pass, acceptance scenarios satisfied, no unresolved gates.
10. Resume note: if interrupted, re-read this GOAL.md, tasks.md, and latest proof before continuing.

Use templates/GOAL.md as the shape. Keep it concise enough to paste into Codex /goal.

Process

1. Inspect the feature directory and required artifacts.
2. Validate refusal conditions.
3. Extract objective, acceptance scenarios, constraints, tasks, and verification commands.
4. Write GOAL.md.
5. Stop and report the output path plus any unresolved risks.

Do not start implementation from this skill. The next step is Codex /goal.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: merllinsbeard
• Source: merllinsbeard/merlin-skills
• License: MIT

Install and usage instructions live in the source repository linked above.