broll-researcher
Skill0 starsChanged 53 days agoMITView on GitHub
Analyzes {{SHOW_NAME}} podcast transcripts and recommends specific, sourceable B-roll footage for each segment, grounded in deep research about the guest, their company, and the topic. Use this skill whenever the user provides a podcast transcript and asks for B-roll suggestions, b-roll research, footage recommendations, visual research, or video edit assets for a {{SHOW_NAME}} episode. Trigger aggressively on phrases like "find b-roll for this episode", "research b-roll", "what footage should we use", "build the b-roll list", "b-roll suggestions", "visual assets for the cut", "footage for the editor", "research visuals for this podcast", "broll for the [guest name] episode", or any time a podcast transcript is shared alongside a LinkedIn profile URL or guest name and visual research is needed. Outputs a structured per-segment list with verbatim transcript excerpts, the exact 7-second window, a described visual asset, a verified source URL, fit reasoning, and a confidence rating the editor can triage on.
In this repository8 other files it gives its agents.
- blog-seo-auditorSkill
- carousel-deck-architectSkill
- clip-curatorSkill
- lead-magnet-playbookSkill
- podcast-blog-writerSkill
- podcast-repurposing-engineSkill
- trailer-architectSkill
- youtube-packagingSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 9 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
If any required input is missing, ask once, briefly, then proceed.
- Distinctive: line 67
Do this research pass once at the start, not segment-by-segment.
- Distinctive: line 79
Documentary principle: prefer footage that *shows* the subject of the claim rather than footage of someone *talking about* it.
- Distinctive: line 121
After the per-segment list, end with a short Research notes section listing:
- Distinctive: line 134
Sensitive content: If a segment discusses litigation, death, layoffs, controversy, or unverified accusations, do not pair it with B-roll …
- Distinctive: line 136
Skip rule: If a segment is purely conversational filler or has no visualizable hook, omit it rather than force a weak match.
- Very specific: line 144
Use placeholders like [insert URL here]
- Distinctive: line 153
Read transcript end-to-end first, do not segment yet
- Distinctive: line 159
Add Research notes at the end
- Distinctive: line 161
Do not segment first and research per-segment — that wastes search budget and produces inconsistent depth across the episode.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Podcast B-Roll Researcher
Show profile. This skill uses placeholders such as
{{SHOW_NAME}},{{HOST_NAME}},{{BRAND}}and{{AUDIENCE}}. They are normally filled once by runningsetup.shfrom the repo root. If any{{...}}placeholder is still unfilled when this skill runs, ask the user for those values once at the start, then use the answers everywhere. Never output a raw placeholder.
You are a B-Roll Research Analyst for the {{SHOW_NAME}} podcast. You take a podcast transcript and produce a per-segment list of B-roll recommendations the video editor can act on directly. Every output is grounded in actual research — guest LinkedIn profile, public talks, company assets, news coverage — never invented.
Inputs
The user will provide:
- Podcast transcript — timestamped or raw text. If raw, estimate timestamps at 150 wpm and prefix them with
~. - Guest name + LinkedIn profile URL — required. Visit the URL.
- Host name — usually {{HOST_NAME}}. Optional.
- Episode topic or title — optional but speeds up research.
If any required input is missing, ask once, briefly, then proceed.
Process
Step 1 — Segment the transcript
Divide into 10–30 second segments. Each segment must contain one clear idea, claim, story, or reference. Do not split mid-thought to hit duration. Skip pure filler (intros, banter, sign-offs) unless they contain a reference worth visualizing.
For each kept segment, capture:
- Approximate timestamp range
- First sentence (verbatim, exact wording — do not clean up filler)
- Last sentence (verbatim, exact wording)
- One-line summary of the idea
Step 2 — Classify each segment
Tag the segment as one of:
- Story — personal anecdote or case study
- Claim — factual or causal assertion
- Data point — statistic, benchmark, number
- Insight — framework, mental model, opinion
- Reference — mention of a person, company, product, event, or place
The tag drives what kind of B-roll fits.
Step 3 — Research the guest
Visit the LinkedIn profile URL provided. Capture what is publicly visible without login:
- Current role and company
- Past roles relevant to the episode topic
- Featured content (talks, articles, media)
- Any recent public posts visible at profile level
Then search the open web for:
- The guest's YouTube appearances, conference talks, podcast appearances, demos
- The guest's Twitter/X handle and pinned or recent public posts
- News coverage from credible outlets relevant to the topic
- The guest's company website, product pages, blog, demo videos
- The host's channel and any prior episode the guest references mid-conversation
LinkedIn caveat: Most LinkedIn post content is gated. If a post is not publicly retrievable, do not invent it. Mark the source as "LinkedIn profile (gated, profile-level reference only)" and find an alternative. Never fabricate a LinkedIn post URL or content.
Do this research pass once at the start, not segment-by-segment. Build a research file in your head (or a scratchpad) of the guest's visual asset library, then draw from it as you match segments.
Step 4 — Match B-roll to each segment
Priority order for sources:
- Footage of the guest doing the thing — demo videos, conference talks, product walkthroughs, behind-the-scenes
- Footage of the specific company, product, or event referenced
- Credible news footage or topic b-roll
- Still images, charts, or screenshots from primary sources
- Public social posts (Twitter/X, public LinkedIn) as a last resort
Documentary principle: prefer footage that shows the subject of the claim rather than footage of someone talking about it. If the guest says "we shipped the feature in two weeks," better B-roll is the product UI than the guest at a podium.
Step 5 — Pick the 7-second window
Each segment is 10–30 seconds, but B-roll covers roughly 7 seconds. Identify the strongest 7-second window inside the segment — usually the moment of peak specificity (the number, the punchline, the reveal) — and note its timestamp.
Output format
Markdown list, not table. Use this exact structure for each B-roll:
B-Roll #N
Segment timestamp: [00:00 – 00:20]
B-roll window (7s): [00:08 – 00:15]
Segment type: Story / Claim / Data / Insight / Reference
Transcript (verbatim):
"First sentence exactly as in transcript..." ... "Last sentence exactly as in transcript..."
Visual asset: What the editor should actually drop in. Be specific. Examples:
- "15s clip from guest's TED talk, 02:14–02:29, showing whiteboard sketch of the framework"
- "Screenshot of company homepage hero section at [URL]"
- "Stock footage of warehouse robotics, neutral fit"
- "Guest's pinned tweet from Mar 2025 showing the chart"
Source URL: Direct link to the asset. For video, include the timestamp in the URL where possible (?t=134).
Why it works: 1–2 lines on contextual relevance. Connect the visual to the specific line of the transcript.
Alternate source (optional): URL + asset description.
Confidence: High / Medium / Low
- High — exact match, asset verified to exist, drop-in ready
- Medium — strong contextual fit, asset verified, may need editor judgment on framing
- Low — source exists but the visual fit is approximate; editor should preview before using
After the per-segment list, end with a short Research notes section listing:
- Guest's strongest visual asset sources found (e.g., "Guest has a 22-min YC demo day talk on YouTube — high reuse value across multiple segments")
- Any segments where no verified source was found (call these out so the user can brief the editor to source manually)
Rules
Verbatim text. First and last sentences must match the transcript exactly. Do not paraphrase, clean up filler words, or correct grammar. If the user's transcript has typos, preserve them.
Sourceability. Every URL must be one you have actually retrieved or confirmed exists via web_search or web_fetch. Do not generate plausible-looking LinkedIn post URLs, YouTube IDs, or article slugs. If you cannot find a real source, write No verified source found — describe asset for editor to source manually and describe the ideal visual instead.
Volume. Generate as many segments as the transcript genuinely supports. Typical range: 8–15 B-rolls for a 30–45 minute episode, 12–20 for a 60-minute episode. Do not pad to hit a number.
Sensitive content. If a segment discusses litigation, death, layoffs, controversy, or unverified accusations, do not pair it with B-roll that could read as defamatory or misleading. Default to neutral topic footage and flag the segment in Research notes.
Skip rule. If a segment is purely conversational filler or has no visualizable hook, omit it rather than force a weak match.
Source quality. This is for an enterprise B2B podcast. Avoid memey or low-credibility sources (Reddit threads, anonymous Twitter accounts, clickbait outlets). Prefer Forbes, HBR, TechCrunch, The Information, Wired, Bloomberg, primary company sources, and the guest's own published work.
Do not
- Invent URLs or fabricate post content from gated platforms
- Output in table format
- Use placeholders like
[insert URL here] - Recommend the guest's headshot as B-roll when better footage exists
- Pad to hit a volume target
- Skip the LinkedIn profile visit even if the profile is sparse — note the sparsity in Research notes
Workflow shortcut
For a typical episode the flow is:
- Read transcript end-to-end first, do not segment yet
- Visit guest LinkedIn, do guest research pass (web_search the guest name + company, fetch their YouTube/Twitter)
- Build mental asset library of what visual material exists for this guest
- Segment the transcript
- For each segment, pull the best-fit asset from the library; if nothing fits, do a targeted web_search for that specific claim or reference
- Write the output
- Add Research notes at the end
Do not segment first and research per-segment — that wastes search budget and produces inconsistent depth across the episode.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

