Framing from Conversation Transcripts
Produce a frame document from one or more conversation transcripts. The frame captures the "why" — what problem to solve and why this one, not the others.
Before You Start
Ask the user:
- Which transcripts? Get specific file paths. Read them in the order the user specifies — conversation order often matters because ideas build across calls.
- What's the topic area? A rough sense of what these conversations were about (e.g., "AI features," "onboarding redesign") so you know what to listen for.
What a Frame Document Contains
A frame has three sections:
1. Source — What was actually said
Capture verbatim quotes from the transcripts. This is ground truth. Everything else in the document is interpretation of this material.
- Attribute quotes to speakers
- Include enough context that the quote makes sense standalone
- Add brief connective tissue between quotes where needed, but keep your voice minimal
2. Pre-work — What options surfaced and why this one
Survey the landscape of options that came up across conversations. This is not background — it's the argument for why we're framing this particular problem.
For each option that surfaced:
- What is it?
- Who benefits?
- How strong is the signal? (How many people raised it independently? Did others pick it up or let it drop?)
Then make the case for the one to pursue now:
- Why is it more urgent or important than the others?
- Why not the others right now? (This is "why not now," not "why not ever" — don't dismiss ideas, just explain why they aren't the priority.)
Do not invent a roadmap or sequence for the other options. Whether and when they happen is a future decision. The only claim you're making is: this one first, for these reasons.
3. Problem / Outcome — The specific frame
Problem: What's broken, what pain exists. Render as bullets for easy scanning.
Outcome: What success looks like. Also bullets. Stay high-level — this is not solution-specific.
4. Less about / More about — Boundary on the solution space (optional)
A guardrail that orients the reader toward what kind of solution fits and away from what kind doesn't. It sits after Problem/Outcome as a separate section. Same format — two symmetric bullet lists.
What it does: Prevents someone from reading the Problem and Outcome correctly but proposing the wrong kind of solution. For example, a problem about getting programs into the system could reasonably lead someone to propose "AI that advises coaches on what to program" — which technically addresses the problem but completely misses the point.
When to include it: When there's a common misunderstanding or an obvious-but-wrong direction that people could easily head toward. The signal is that people in the conversations are actively drawing a line — saying what this is NOT. If nobody felt the need to draw that line, you probably don't need the section.
How it surfaces: This probably won't emerge mechanically from a first pass through the transcripts. It's more likely to surface during review — either the shaper notices a pattern of people saying "not about X," or someone reading the frame proposes something that fits the Problem/Outcome but misses the point. That's the moment where you realize the boundary needs to be made explicit.
What it looks like:
markdown## Less about - [What this project is NOT trying to solve] - [The wrong direction people might naturally head toward] ## More about - [What kind of solution actually fits] - [The real nature of the problem being addressed]
No quotes or attribution needed — it's a synthesis. Keep it to the key points.
The Key Discipline: Evidence, Not Editorializing
After writing each line in Problem and Outcome, ask: who said this, and where?
- If you can point to a specific person and moment → keep it, cite it
- If it's directly implied by what someone said → keep it, mark it as implied with your reasoning
- If you can't trace it back → drop it
Common traps to avoid
Don't connect dots and present inferences as facts. "Coaches bounce during trials because data entry is too hard" sounds plausible but if nobody said it, don't write it as established truth.
Don't embellish for vividness. Adding specificity that nobody stated ("every exercise, set, rep, and percentage") to make the problem sound more concrete is editorializing.
Don't inflate the options list. An idea that one person mentioned and nobody picked up is not an "option with traction." Be honest about signal strength. Ask yourself: did multiple people raise this independently? Did others build on it or let it die?
Don't editorialize in the Problem statement. Phrases like "TB has to earn the switch" or "the single biggest friction point" are rhetoric, not evidence. Make the case using what was said.
Don't present your framing as their words. Keep your interpretive voice separate from what was actually said. The Source section is theirs. Problem/Outcome are your distillation — and should be traceable back to Source.
Reviewing the Frame
After drafting, examine every line in Problem and Outcome:
- Read each bullet
- Ask: what's my evidence for this claim?
- If the evidence is a direct quote — cite it parenthetically
- If the evidence is implied — say so explicitly
- If there's no evidence — drop the line
Do the same for the pre-work options table. For each option listed, verify:
- Did it actually get traction in the conversations? (Multiple people, independently?)
- Or was it one person's idea that others didn't pick up?
Be willing to shrink the options list. Fewer real options are better than a padded list.
Document Format
markdown--- shaping: true --- # [Topic] — Frame ## Source ### [Speaker] ([Date]) > "Verbatim quote..." > "Another quote..." [Brief connective context where needed.] ### [Speaker] ([Date]) > "Verbatim quote..." --- ## Pre-work: [Topic] Options Landscape [N] options surfaced with real traction across conversations: | Option | What it does | Who benefits | Signal strength | |--------|-------------|--------------|-----------------| | **A. [Name]** | ... | ... | ... | | **B. [Name]** | ... | ... | ... | ([Note on ideas that were mentioned but didn't get traction — dropped.]) **Why A now:** [Evidence-based argument for urgency/importance.] --- ## Problem - [Bullet — traceable to source] - [Bullet — traceable to source] ## Outcome - [Bullet — high-level, not solution-specific] - [Bullet — high-level, not solution-specific] --- ## Less about - [What this is not] ## More about - [What this is]
What This Skill Does NOT Do
- Does not shape solutions. The frame is the "why." Shapes (the "how") come after, in a separate shaping doc.
- Does not sequence future work. It picks one thing as the priority now. It does not roadmap the rest.
- Does not summarize conversations. The Source section captures key quotes. The rest of the document is a distillation, not a summary.
