القائمة
What to examine during a reflection pass and how to write observations to memory/{agent}/needs.md.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
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 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.
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.
When and how to write scenario holdouts for factory verification.
How to integrate scenario evaluation into verification reports.
| name | agent-reflection |
| description | What to examine during a reflection pass and how to write observations to memory/{agent}/needs.md. |
When invoked with a reflection prompt (not a heartbeat), follow these instructions rather than the usual heartbeat behavior.
Read each area before writing. Only write observations where you find genuine friction, misalignment, or improvement opportunities.
Your role description — Does it accurately describe what you actually do? Is anything missing or misleading?
Your skills list — Which always-loaded skills do you actively use? Which available skills have you never activated? Are skills missing that would help your work?
Your access scopes — Have you ever needed to read or write outside your current can_read/can_write? Are there directories in your scopes you never touch?
Your recent daily logs — Read the last 5 entries. What patterns emerge? Repeated NO_REPLY? The same workaround appearing multiple times? Recurring blockers?
Your private memory — Is it accurate and current? Is anything missing you keep rediscovering?
Your pipeline inputs — Are they consistently populated? Is anything stale, malformed, or missing that upstream agents should be producing?
Your pipeline outputs — Is your output format working for downstream agents? Have you seen signs it isn't being consumed correctly?
The learnings directory — Are there learnings in learnings/ relevant to your work that haven't been incorporated into your skills or memory?
Use the same format as shared/human-action-needed with category: observation:
## {agent}-{kebab-case-description}
- status: open
- created: {ISO 8601 timestamp}
- category: observation
- blocked: {description of friction, misalignment, or improvement opportunity}
- context: {what prompted this — reflection pass, pattern across runs, specific incident}
### Exact Change
{suggested improvement — may be concrete or a description of what should be different}
The blocked field for observations describes the friction, not an inability to work. The ### Exact Change section describes the suggested improvement, even if you're uncertain it's the right fix.
Apply the standard deduplication check: if an open entry with the same ID already exists, do not create a duplicate.
For genuine blockers discovered during reflection, use the appropriate blocker category (permission-change, config-edit, manual-intervention, approval) rather than observation.