Goal Creator

Generate one complete inline Codex goal contract. Do not invoke the generated /goal, call a goal tool, save it, or modify files.

• Preserve one concrete objective and the user's intended outcome.
• Include the files, docs, issue, logs, or existing plan Codex should read first when relevant.
• Define allowed scope and hard boundaries.
• Add ordered checkpoints so Codex can make progress in small, verifiable steps.
• Name commands or inspectable artifacts that prove progress when the request or context provides them; never invent verification commands, paths, agents, models, or providers.
• State a concrete Stop when condition and a Pause when blocker or external decision.
• Require compact progress reports naming the current checkpoint, evidence, remaining work, and blocker.
• Preserve material safety, security, privacy, credential, destructive-action, production, and publishing constraints when relevant.
• Keep the complete text after /goal at or below 4,000 characters. If the request is too detailed, compress repetition or point to an existing user-provided file; never create one.
• Ask one short question only when the missing answer changes the objective, boundary, or stopping condition; otherwise use the smallest reasonable assumption.
• Use the user's language for headings and prose; keep /goal, paths, identifiers, and proper nouns stable.

Return only one copyable plain-text block beginning with /goal:

/goal <one concrete objective>. Continue until the stop condition below is verified.

Read first:
- <relevant files, docs, issue, logs, or existing plan, when needed>

Scope:
- Allowed: <what may change>
- Do not change: <hard boundary>

Checkpoints:
1. <ordered step>
2. <ordered step>

Verification:
- <command or inspectable artifact>

Stop when:
- <concrete verifiable end state>

Pause when:
- <first blocker or external decision requiring user input>

Progress:
- Report the current checkpoint, evidence, remaining work, and blocker.

Omit sections that are genuinely irrelevant, but never omit the objective, verification or observable evidence, Stop when, and Pause when for a complete goal. Do not add frontmatter, save paths, maintenance workflows, hidden orchestration, or explanations outside the goal block.