six-thinking-hats

Edward de Bono's method, adapted for an agent working with a human: one perspective at a time, so critique, enthusiasm and invention stop colliding in the same paragraph.

When

A choice with real trade-offs (architecture option, scope cut, build vs buy), a stalled discussion, or a plan about to be committed. Not for a task whose answer is already clear.

Steps

1. The hats

Hat	Perspective	Rule
White	Facts and data	No interpretation; state what is known and what is missing
Red	Feelings and intuition	Gut reaction, no justification required
Black	Caution	Risks, obstacles, drawbacks
Yellow	Optimism	Benefits, opportunities, value
Green	Creativity	Unconventional options, no filtering yet
Blue	Process	Frames the session, summarises, concludes

2. Sequence

1. Default order: Blue (problem, goal, time) → White → Red → Black → Yellow → Green → Blue (sort the ideas, decide, list next steps). Not every hat has to be used; a single hat is fine (Green alone to generate options).
2. Everyone wears the same hat at the same time; the facilitator keeps Blue throughout.
3. Keep each hat short (a minute or two per participant in a live group); Green may run longest.

3. Applying it with an agent

1. Write one section per hat in the chosen order, headed by the hat name, and do not mix: no risks in the Yellow section, no verdict before the final Blue.
2. The final Blue states the decision or the open question, and the next action.

Output / checkpoint

One short section per hat used, then a Blue conclusion with decision and next steps, shown to the human before acting on it.

Guardrails

• Do not let Red justify itself or Black veto before Yellow and Green have spoken.
• A structure for thinking, not a source of facts: White must say what is unverified.
• Skip it for simple decisions; the ritual costs more than it returns.

Origin

Method by Edward de Bono; summary read 2026-10-14 at manager-go.com (page "6 chapeaux de la réflexion"). Rewritten in our own words; the hat roles and the typical sequence are the method's, the agent adaptation is own guidance. 🟡 not dogfooded.