| name | develop-idea |
| description | Research, refine, and scope a single video idea Dan already has — validate demand with evidence, find the angle competitors are missing, and define what the video covers, ending in a go / reframe / park verdict. Use whenever Dan brings his own idea rather than asking for new ones — "I want to make a video about X", "is X worth a video?", "help me scope this video", "research this topic for a video", "what should I cover in a video on X", "how should I angle this". Finding ideas from scratch is `video-ideation`; turning a settled, scoped idea into a title/thumbnail/intro is `video-packaging`. This skill is the bridge between them. |
Develop Idea
Dan arrives with an idea; the job is to pressure-test and shape it, not to admire it. A good session answers three questions with evidence: is there demand for this, what angle wins against what already exists, and what exactly should the video cover. The honest outcomes are go, reframe (the idea is close but a different angle or framing is stronger), or park (the evidence says wait or skip — saying so saves Dan a week of production on a video nobody asked for).
Present the resulting brief directly in the conversation — no artifact, no file saved to a repo or working folder.
Workflow
Step 1 — Capture the seed
Restate the idea as you understand it in one or two sentences and note any constraints Dan gave (deadline, effort level, a specific release it ties to). Then check what he actually asked for:
- Full development ("is this worth making?", "help me with this idea") → run all steps.
- A partial ask ("just help me scope it", "what's a good angle?") → do that part well; only pull in other steps if their findings change the answer. Don't run a demand sweep Dan didn't ask for when he's already committed to filming.
Don't interrogate him upfront — the research surfaces better questions than guessing does. Ask at most one clarifying question, and only if the idea is genuinely ambiguous.
Step 2 — Validate demand
Research pointed at this one topic, not a broad sweep. Tactics, tool names, and the channel list are in references/research-guide.md — read it before starting. Run sources in parallel (via subagents if available):
- Dan's own data — has he (or his blog/newsletter) covered this or adjacent topics, and how did it perform? Prior over-performance on a neighboring topic is strong evidence; a prior flop is worth knowing too.
- YouTube signal — do existing videos on this topic pull outlier views relative to their channel size? No videos at all is ambiguous: either an open lane or no demand — search volume decides which.
- Community signal — recent HN/Reddit threads, release notes, conference talks touching the topic. Recurring confusion is tutorial demand; heated debate is "explained/opinion" demand.
- Search demand — what learners type, whether the head terms are rising or flat, and whether what currently ranks is stale.
Every claim carries its evidence (a link and a number). If a number can't be verified, say "unverified" — never invent one.
Step 3 — Map the competition
This is where "refine" happens: the angle falls out of the gap. Look at the 3–6 videos that currently own this topic and note for each: what it covers, its format, its framework/tool versions, and what it skips or gets wrong. Then ask:
- What does every existing video miss that Dan can deliver — a newer version, a working build instead of slides, the Java/Spring angle on an AI-world topic?
- Is the topic saturated by big channels (hard to win on the same framing) or held by outdated/low-quality results (easy win)?
Propose 2–3 angle options, each stated as a promise ("who it's for and what they walk away with"), and recommend one with the reasoning attached.
Step 4 — Scope the video
Define the video concretely enough that Dan could start outlining:
- In: the 3–5 things the video covers, in rough order.
- Deliberately out: adjacent material that would bloat it — name it so it's a decision, not an accident. If something cut is strong on its own, flag it as a possible follow-up video.
- Demo plan: what's actually shown on camera — the build, the before/after, the result that proves the promise. Dan's outperformers are "show, don't slide" videos; a scope with nothing demonstrable is a warning sign.
- Assumed knowledge: what the viewer already knows, so the video starts at the right altitude.
- Rough shape: estimated runtime and effort (quick win vs. multi-day build).
Scope serves the chosen angle: everything "in" must support the one promise from Step 3. When a list grows past one promise, that's two videos.
Step 5 — Verdict and brief
Present the brief in the conversation, verdict first:
- Go — demand confirmed, angle chosen, scope defined. Lead with the one-line reason backed by the strongest evidence.
- Reframe — the topic has legs but a different angle is stronger. Show the original vs. the reframe and why the evidence favors it.
- Park — weak or falling demand, or a saturated field with no gap. Say so plainly, cite the evidence, and note what would change the verdict (an upcoming release, a rising trend to watch).
Then the supporting sections: the demand evidence, the competitive map with the gap, the angle options with the pick, and the scope. Close a go with the handoff: the promise sentence and target keyword feed straight into video-packaging, and video-project scaffolds the project folder when he's ready to film.
Principles
- Evidence over vibes. Same rule as ideation: every claim traces to a link and a number, or is labeled unverified.
- Park is a win. A well-evidenced "don't make this" is worth more than a polite green light. Dan brought the idea; he needs the truth about it, not validation.
- The angle comes from the gap. Refinement isn't wordsmithing the idea — it's finding what the existing coverage misses and aiming there.
- One video, one promise. Scoping is mostly deciding what to leave out. Cut material is future-video fuel, not waste.