Objetivos de aprendizaje
- Delegate exploration without losing control.
- Write a reusable skill.
- Consolidate conflicting findings.
Split by questions
A good division separates research, implementation and verification. Do not send several agents to edit the same files: it creates conflicts and makes changes hard to attribute.
- Agent A: repository map, read-only.
- Agent B: tests and risks, read-only.
- Agent C: minimal patch proposal, no writing.
- Human owner: decide, apply and validate.
A useful skill
A skill is a working recipe with an objective, boundaries, steps and evidence. Start with one repeated task before generalizing.
Goal: review changes before merge. Input: diff and requirements. Limits: do not modify files; do not invent tests. Output: prioritized findings with evidence and missing tests.
Comprueba que funciona. Run two read-only tasks in parallel and compare their references before making a decision.
Guardar y reabrir el proyecto.
More agents do not create more truth; without clear missions they create more coordination work.
Siguiente misión recomendada
Si has guardado la evidencia de esta lección, continúa con «Local DeepSeek models and quantization». Si no, repite la comprobación antes de avanzar.