| name | perspectives |
| description | Multi-perspective analysis of a problem or proposal through evidence, sentiment, risks, benefits, alternatives, and process. Use for a quick "what are we missing?" pass. Do not use to choose a consequential option, run cross-model deliberation, or review code. |
| user-invocable | true |
| argument-hint | [topic or proposal] [optional: perspective] |
Perspectives
Analyze a problem, proposal, or design through six isolated perspectives, then
synthesize what matters. This is a lightweight exploration tool based on Edward
de Bono's Six Thinking Hats framework. It broadens the conversation; it does
not replace decide, consensus, or code-review.
Routing
- Use
perspectives for quick breadth, brainstorming, or "what are we
missing?"
- Use
decide when the user must choose between consequential options and
needs a durable recommendation.
- Use
consensus when a complex assessment or implementation needs independent
cross-model review and user approval gates.
- Use
code-review when code or a diff exists and the goal is to find defects.
Input
Accept a topic or proposal and an optional perspective. If the topic is too
vague to frame as one focused question, ask one clarifying question. Otherwise,
start immediately.
Perspective aliases:
| Perspective | Aliases |
|---|
| Evidence | white, facts, data, unknowns |
| Sentiment | red, feelings, gut, reactions |
| Risks | black, caution, failure |
| Benefits | yellow, optimism, upside |
| Alternatives | green, creative, options |
| Process | blue, meta, next steps |
Full Analysis
When no perspective is specified:
- Restate the topic as a focused one-sentence question.
- Analyze all six perspectives in the order below. Keep each section within
its assigned mode; integration happens only in synthesis.
- Synthesize the decision-relevant insights and smallest useful next step.
Evidence
- Separate verified facts, user-supplied claims, assumptions, and unknowns.
- Cite available evidence. Label unsupported claims; do not invent facts.
- Identify the unknowns most likely to change the conclusion.
Sentiment
- Record the user's stated intuition and stakeholder sentiment.
- When no sentiment was supplied, describe plausible reactions explicitly as
hypotheses, not facts or the agent's own feelings.
- Keep this section short. Feelings matter as signals, not proof.
Risks
- Identify plausible failure modes, downsides, constraints, and second-order
effects.
- Be genuinely critical, but distinguish demonstrated risks from speculation.
Benefits
- Identify plausible benefits, leverage, opportunities, and best-case outcomes.
- State what must be true for each important benefit to materialize.
Alternatives
- Generate options that avoid or change the central trade-off.
- Include doing nothing, staged approaches, smaller experiments, and combined
options when relevant.
- Allow unconventional ideas, but label impractical ones.
Process
- Assess whether the question is framed well and whether more analysis is
useful.
- Identify the owner or decision-maker when known and any evidence needed
before action.
Synthesis
Conclude with:
- The one or two insights most likely to affect the outcome
- Any sentiment that should be acknowledged
- The strongest unresolved unknown
- A concrete next step or recommendation
Do not force a decision when the task is exploratory. If the analysis reveals
a consequential choice, recommend continuing with decide. If it reveals a
need for independent model review, recommend consensus.
Focused Analysis
When the user specifies one perspective, run only that perspective in depth and
skip synthesis. Use 5-10 concise points and a brief framing sentence.
Use the practical perspective names as output headings rather than colors.
