메뉴
When and how to write a human-action-needed entry to memory/{agent}/needs.md. Agents write these entries when they hit blockers that only a human can resolve.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Complete reference for the factory CLI — all commands, arguments, options, and usage. Use when an agent needs to invoke factory commands, check correct syntax, or understand available options.
When you encounter a failure, correction, or new pattern, write a learning and optionally propose a skill.
How the factory workspace is organized and what each directory means.
What to examine during a reflection pass and how to write observations to memory/{agent}/needs.md.
When and how to write scenario holdouts for factory verification.
How to integrate scenario evaluation into verification reports.
| name | human-action-needed |
| description | When and how to write a human-action-needed entry to memory/{agent}/needs.md. Agents write these entries when they hit blockers that only a human can resolve. |
Write a needs entry when you encounter a blocker that:
agents.yaml, files in universe/)can_write or can_read in agents.yaml)factory cleanup-specs, factory rebuild, factory resolve)Write an observation entry (non-blocking) during any work — not only reflection passes — when you notice friction, limitations, tooling gaps, architectural issues, or improvement opportunities. Observations in your response text are ephemeral; writing them to needs.md makes them durable and actionable by the factory.
Do NOT write a needs entry for:
Write to memory/{your-agent-name}/needs.md. Entries are appended; the file is never deleted:
## {agent}-{kebab-case-description}
- status: open
- created: {ISO 8601 timestamp}
- category: {permission-change | config-edit | manual-intervention | approval | observation}
- blocked: {what you cannot do, or what friction you observe}
- context: {which task, which run, what was attempted}
### Exact Change
{the literal text, YAML block, command, step, or suggested improvement the human needs to review}
Before writing, check your own memory/{agent}/needs.md for an existing open entry with the same {id} heading. If one exists with status: open, do not create a duplicate. If it exists with status: resolved, a new occurrence is a new entry — append it.
| Category | Use when |
|---|---|
permission-change | Agent's can_read or can_write needs expansion |
config-edit | A READ-ONLY config file needs modification |
manual-intervention | One-time human action outside the codebase |
approval | Proposed action requires human sign-off before proceeding |
observation | Non-blocking feedback — kernel automatically promotes to specs/factory-internal/ for structured triage. |
Note the needs entry in your daily log. The human runs factory needs to see all open blocker entries and factory needs --resolve {id} after taking action.
Observations are automatically promoted to specs/factory-internal/ by the kernel post-run pass and triaged by the operator via factory triage. They no longer appear in factory needs output — use factory triage --list to review them.