debrief-and-learn

Files (1)

• SKILL.md

SKILL.md (first lines)

# Debrief And Learn

This is an internal meta skill. Humans should not need to reference it directly.

## Purpose

After build work completes, inspect the outcome and extract reusable learning for Darwin together with the operator.

The goal is not just to finish the task. The goal is to make Darwin better at the next task.

## Run After

- approved build work has been executed
- relevant verification has been run
- the current task outcome is understood well enough to describe honestly
- the operator has had a chance to share how the result felt from their side

## Debrief Questions

Answer these after each meaningful build task:

1. What was the user trying to achieve?
2. What plan was chosen?
3. What succeeded quickly?
4. What created confusion, rework, or delay?
5. How did the operator feel about the result and the workflow?
6. What patterns should Darwin reuse next time?
7. What skills, prompts, or docs should Darwin improve?

## Output

Produce a concise learning record with:

- task summary
- verification evidence
- friction points
- successful patterns
- operator feedback
- proposed Darwin improvement, if one meaningfully emerged
- whether the improvement belongs in a skill, prompt, or documentation, if applicable

If needed, store the record in `.opencode/darwin-debrief.md` only as a temporary working artifact while the Darwin improvement is being applied.

The preferred outcome is to canonize the learning immediately by updating the relevant Darwin skill, command, prompt, or documentation in the same session.

It is also valid to conclude that the existing Darwin workflow held up well and no meaningful update is warranted for this task.

## Guardrails

- Do not invent lessons without evidence from the task.
- Prefer small reusable improvements over broad vague ideas.
- If no meaningful lesson emerged, say so.
- Do not force a Darwin update just to satisfy the ritual of debriefing.
- If the task revealed a repeated pain point, route it into Darwin's self-improvement loop.
- If operator feedback is missing, ask for it before treating the debrief as complete.
- Do not leave a clear and small Darwin improvement sitting only in a debrief artifact.

From tykowale/darwin, MIT.

More from tykowale/darwin