| name | reflect |
| description | CEO structured reflection on recent decisions, routing choices, and team health. Identifies patterns in what worked and what didn't, updates MEMORY.md with strategic lessons, and proposes process improvements. Run monthly or after notable incidents. |
Reflect
Reflection is how the CEO improves over time rather than just executing the same patterns repeatedly. It turns experience into durable adjustments.
When to Run
- Monthly (first weekly-review of the month)
- After a notable incident (missed deadline, article published with errors, production outage)
- When the weekly-review identifies a recurring pattern (same process debt 3 weeks in a row)
- Explicitly requested
Reflection Questions
Work through each question and write a 1–3 sentence answer:
Content pipeline:
- What articles took longer than expected to publish, and why?
- Which articles performed best (traffic, syndication reach)? What made them different?
- Did the bilingual workflow (UA draft → Translator → Editor) introduce any bottlenecks?
Team and routing:
- Were there routing mistakes (work sent to the wrong agent, then redirected)?
- Did any agent repeatedly produce output that needed rework?
- Were cross-cutting issues handled cleanly or did the dual-issue pattern create confusion?
Process:
- What was done manually that should be automated?
- What issue or task took the most back-and-forth to resolve?
- Were there any HITL interventions (human had to manually fix something)?
Quality:
- Did any published article have post-publication corrections?
- Did any deploy fail or require rollback?
- Were there security or dependency issues found after merge?
Output
Write a reflection note to MEMORY.md:
### Reflection — {YYYY-MM-DD}
**What worked:**
- {specific example}
**What didn't:**
- {specific pattern with 1-line root cause}
**Adjustments:**
- {concrete change to process, routing rule, or skill guidance}
**Open questions:**
- {things to watch over the next month}
Then, for each "adjustment" identified:
- If it affects a skill → create an issue to update that SKILL.md
- If it affects agent routing → create an issue to update the relevant AGENTS.md
- If it's a process question for humans → post a comment in the root issue flagging it for human decision
Invariant
Reflection produces action items, not just notes. Every insight that warrants change becomes an issue. Notes without issues are forgotten.
