think-reframe

What this skill does

# Think-Reframe - Problem Redefinition Before Problem Solving

Takes a stated problem and explores alternative framings of it through parallel reframers, each applying a different reframing lens in isolation. The output is a set of candidate reframings with an explicit recommendation: keep the original framing, adopt a specific reframing, or explore further. Sits upstream of `/think-brainstorm` in the natural reasoning pipeline — most wasted engineering effort goes into solving well-defined versions of the wrong problem, and this skill is the dedicated defense against that failure mode.

**This skill produces no tangible artifacts.** It is a consultant, not an implementer. No code, no tickets, no commits. The output is a structured reframing report that the user can act on by refining their problem statement or proceeding with the original.

## Roles

**Judge (you, running this skill):**
- Capture the stated problem in a written brief
- Extract stated and unstated premises (lightweight reconnaissance, not confrontation)
- Choose appropriate reframing lenses
- Spawn reframers and synthesize the alternatives into a report with a recommendation

**Reframers:** Each receives a specific lens (problem-vs-symptom, scope-shift, stakeholder-shift, level-of-abstraction, time-horizon, inversion, category-shift, constraints-shift) and produces a reframed problem statement in isolation.

## Workflow

### 1. Receive the Problem

The problem may arrive as:
- **Conversation context** — summarize it back, confirm
- **A document** — read the file (design doc, ticket, problem statement)
- **Fresh user input** — capture it verbatim

**Produce a written brief** of the stated problem as you understand it. Reframers operate on this brief. Ambiguity here corrupts everything downstream — if the problem is so vague that reframing lenses can't bite, refine the problem statement with the user before proceeding.

### 2. Extract Premises

**Lighter than `/think-brainstorm`'s assumption-validation phase.** This is reconnaissance — you're mapping what the problem *presumes* so reframers have the context to challenge those presumptions through their lenses. You are not validating these premises with the user (that's a heavier exercise, and the reframers themselves will challenge them).

**Identify:**
- **Stated premises** — what the problem statement explicitly asserts
- **Unstated premises** — what the problem statement assumes without saying
- **Implied scope** — what the problem is about, what it's not about
- **Implied category** — is this framed as technical / people / process / etc.?
- **Implied stakeholders** — whose problem is this?
- **Implied time horizon** — immediate, medium, long-term?

This snapshot is passed to reframers as context. The reframers' job is to challenge and shift these.

### 3. Choose Reframing Lenses

Select 3-6 lenses from the palette based on the problem's shape. The orchestrator decides autonomously.

**Available lenses:**
- **problem-vs-symptom** — is the stated problem a symptom of something upstream?
- **scope-shift** — narrower or wider framing
- **stakeholder-shift** — different stakeholders' perspectives
- **level-of-abstraction** — more general or more specific
- **time-horizon** — 1 week, 6 months, 5 years
- **inversion** — what new problems would emerge if this were solved?
- **category-shift** — is this really a technical / people / process / spec / business / incentive / communication problem?
- **constraints-shift** — which "fixed" constraints are actually negotiable?

**Selection heuristics:**
- Problem described as "we need X" → include stakeholder-shift (is X what different parties need?), category-shift (is "we need X" a technical problem, or a prioritization problem?)
- Problem described as urgent → include time-horizon (different horizons yield different interventions)
- Problem described with a hard constraint → include constraints-shift
- Problem described with a clean story → include problem-vs-symptom (clean stories often hide upstream causes)
- Problem that feels stuck → include inversion and level-of-abstraction (both break out of the current framing)

**Drop lenses that don't fit.** Not every lens applies to every problem. A clearly-factored technical problem may not need category-shift; a personal productivity problem may not need stakeholder-shift. Better 3 fitted lenses than 7 forced ones.

### 4. Spawn Reframers (Parallel, Isolated)

Spawn one `THK - Reframer` agent per chosen lens, in parallel. Each receives:
- The written problem brief (from step 1)
- Its assigned lens
- The premise snapshot (from step 2)
- Instruction to produce a reframed problem statement with an explicit "what changed" diff and "when this framing applies" note

**No cross-talk between reframers.** NGT principle — independent reframing first, synthesis second. Isolated reframers produce more distinct alternatives; coordinated ones anchor on each other.

Collect all reframings.

### 5. Synthesize

Combine the isolated reframings into a coherent report:

**5a. Assess meaning-shift per reframing.** For each reframing, judge how much it materially shifts the problem's meaning. A reframing that just restates the problem in different words is not productive. A reframing that would lead to a different intervention is productive. Classify each as:
- **Materially shifts meaning** — adopting this reframe would lead to a meaningfully different solution
- **Refines the framing** — sharpens the original without replacing it
- **No meaningful shift** — the original framing holds up through this lens (this is still valuable calibration)

**5b. Look for composite framings.** Sometimes two lenses compound — e.g., "stakeholder-shift to operator + time-horizon 5 years" = "what does this look like to the operator who has to maintain this for 5 years?" These composite framings can be sharper than either parent. Only include if the composition is genuinely illuminating — don't manufacture them.

**5c. Identify standout reframings.** The 1-3 reframings that most change what solving the problem would mean. These are the recommendations to seriously consider.

**5d. Form an orchestrator recommendation.** One of:
- **Keep original framing** — reframings refined the framing but didn't materially shift it; the original is sound
- **Adopt reframing X** — one reframing materially shifts the problem in a way the user should consider adopting
- **Further explore** — multiple reframings materially shift the problem; the user should pick among them rather than receive a single recommendation from this skill

### 6. Report

**Final report format:**

```
## Reframing Report

**Stated problem:** [one-line summary of original framing]
**Lenses applied:** [list]

### Original Framing

[Restate the problem as originally stated, with the premise snapshot
making implicit assumptions visible. This is the baseline each reframing
is measured against.]

### Standout Reframings

[1-3 reframings that materially shift the problem's meaning. For each:]

#### [lens name]

**Reframed:** [concrete alternative problem statement]

**What changed:** [explicit diff from the original framing]

**When this framing applies:** [conditions under which this reframing
is most useful]

### Composite Framings

[Optional. Reframings the orchestrator constructed by combining lenses.
Only include if the composition is genuinely illuminating.]

### Refinements (Not Materially Shifting)

[Reframings that sharpened the framing without replacing it. Listed
briefly for completeness.]

### Where the Original Framing Held Up

[Honest reporting: which lenses produced "no meaningful shift." This is
calibration signal — it tells the user which dimensions of their framing
are robust.]

### Recommendation

One of:
- **Keep original framing** — no reframing materially shifted the problem; proceed as framed
- **Adopt reframing X** — reframing X materially shifts the problem in a way worth adopting
- **Further explore** — multipl