youtube-vid-research

Pack Availability Guard

Before telling the user to run a skill from another project-local pack, check .agents/project.json.enabled_packs. If the target pack is not enabled, recommend $pack install <pack> instead of the target skill. Global skills are always valid. Skills from this same pack are valid because the current skill is already running from that pack.

YouTube Video Research

Invoke as $youtube-vid-research.

Report-First Approval Gate

Default to report-only: present findings, evidence coverage, assumptions, recommended artifact path, and proposed file changes in a pre-approval alignment page plus a concise conversation summary for user approval before creating or updating canonical research, spec, or task files.

Do not write or overwrite synthesized deliverables until the user explicitly approves, unless the user invoked an explicit write/update/fix mode or clearly asked to write files upfront. Raw evidence capture may be persisted before analysis when reproducibility requires it; report those raw paths separately and still gate synthesized research/report writes.

When stopping for approval, build and attempt to open the alignment preview page first, then ask the user to review it and approve, question, or request adjustments. Do not include Recommended next skill, Recommended next command, or downstream routing language. The approval request itself is the next action. Only emit next-skill routing after the approved artifact has been written or updated.

Staged Research Workflow

Use this staged workflow for synthesized research or report outputs that would create or update canonical research, spec, or task files.

1. Stage 1 - Research and clarify. Perform the research, run required source/code checks, and ask any needed clarification questions. Write only a non-canonical working packet: flat mode uses research/_working/preliminary-<skill>-research.md; product-path mode uses research/{slug}/_working/preliminary-<skill>-research.md. Replace <skill> with this skill's name value. Do not create or update canonical research, spec, or task files in Stage 1. Raw evidence or search logs may remain as supporting evidence where this skill already requires them, but synthesized deliverables stay in the working packet.
2. Stage 2 - Review alignment. Consume the working packet and build the review HTML alignment page. The page must render the full preliminary packet, evidence matrix, assumptions/confidence register, source coverage gaps, proposed canonical file changes, and approval gates. Stop for either feedback-only YAML or final compiled YAML. Feedback-only YAML revises the working packet and page, then remains in Stage 2.
3. Stage 3 - Finalize approved artifacts. Consume final compiled YAML only when it has no unresolved needs-clarification, unresolved down feedback, or other unresolved negative feedback. Apply approved edits first, archive the working packet to docs/history/archive/YYYY-MM-DD/HHMMSS/<original-working-path>, remove the active working packet, write the approved canonical artifacts to the unchanged output paths below, and convert the alignment page to confirmed with the approval record preserved.

Canonical output paths remain unchanged. Search logs and other supporting evidence remain allowed only where this skill's output contract already requires them.

Research external YouTube videos so the agent can understand what the user is referring to and reuse that context in specs, strategy, writing, product work, or implementation. This is the general comprehension lane; use $youtube-video-audit for performance diagnosis, $youtube-competitive-research for why a competitor video worked, and the remotion pack's $youtube-format-research for production/style breakdowns.

Evidence And Feedback Handling

Treat user feedback as input to evaluate, not as automatic ground truth.

• For factual, evidentiary, technical, or source-backed claims: verify against available evidence. If the user appears to misunderstand the evidence or states something factually incorrect, push back clearly and cite the evidence. Do not rewrite findings merely to agree.
• For taste, brand, positioning preference, risk appetite, prioritization, or other subjective judgment calls: weigh user feedback heavily and adapt the recommendation unless it conflicts with verified evidence.
• When feedback mixes facts and preference, separate them explicitly: correct the factual part, then incorporate the preference where it is a legitimate judgment call.
• When uncertain, say what is known, what is inferred, and what would change the conclusion.

Inputs

• Required: one or more YouTube video URLs or video IDs.
• Optional --focus context|claims|summary|references: default context.
• Optional --compare-channel <slug>: reuse channel evidence under research/youtube/data/<slug>/ when the user asks how the reference relates to a known channel.

Process

1. Resolve every target into a video ID from watch URLs, Shorts URLs, youtu.be URLs, embed URLs, or raw 11-character IDs.

2. Require yt-dlp for public metadata:

   command -v yt-dlp
   

3. Select a transcript Python interpreter. Prefer a workspace-local .venv; create it if missing. Install youtube-transcript-api into .venv only when the import check fails and network access is available.

4. Persist raw evidence before analysis under research/youtube/data/<video-id>/:

   • metadata-YYYY-MM-DD.json: raw yt-dlp --dump-json "VIDEO_URL" output.
   • transcript/<video-id>.json: raw transcript JSON when available.
   • transcript/transcript-summary.json: transcript text or failure reason.

5. Extract public metadata fields when present: title, URL, channel, upload date, duration, description, chapters, tags, categories, thumbnails, view count, like count, and comment count.

6. Analyze the video for:

   • Main thesis, audience, context, and assumed prior knowledge.
   • Key claims, examples, frameworks, named entities, tools, sources, and references.
   • Narrative structure: hook, setup, sections, payoff, and CTA.
   • User-relevant takeaways: what matters for the current project, idea, spec, or decision.
   • Uncertainties and evidence gaps.

7. If multiple videos were supplied, synthesize common themes, contradictions, and how each video contributes distinct context.

Output

Create the research/youtube/ and research/youtube/data/<video-id>/ directories if they do not exist.

Write:

research/youtube/video-research-<video-id-or-slug>-YYYY-MM-DD.md

Use this structure:

# YouTube Video Research - [Title or Video Set]

> Videos: [URLs]
> Date captured: YYYY-MM-DD
> Evidence: [raw paths used]
> Focus: context / claims / summary / references

## Evidence Coverage

| Video | Public metadata | Transcript | Notes |
|---|---|---|---|
| [Title](URL) | Available / Missing | Available / Missing | ... |

## Context Brief

[What this video is about, who it is for, and why it matters to the user's request.]

## Key Claims And Examples

| Claim / Idea | Evidence | Confidence | Relevance |
|---|---|---|---|
| ... | metadata/transcript/chapter reference | High / Medium / Low | ... |

## Structure Notes

- **Hook**: ...
- **Sections**: ...
- **Payoff**: ...
- **CTA / next action**: ...

## Terms, References, And Source Leads

[Named people, companies, tools, papers, links, topics, or source leads surfaced by the video.]

## Transferable Takeaways

[What the user or downstream agent should carry forward.]

## Open Questions And Evidence Gaps

[Missing transcript, unclear claims, unavailable references, or claims needing independent verification.]

Constraints

• Do not invent transcript quotes, links, references, comments, metrics, or visual details.
• Mark missing transcript or metadata explicitly.
• Do not treat a single external video as proof of a market, trend, or channel pattern.
• When a claim affects factual, legal, medical, financial, or current-event accuracy, recommend independent source verification instead of relying only on the video.
• Do not bypass login walls, bot protections, access controls, or private YouTube Studio data.
• Archive existing canonical artifacts before replacing them under docs/history/archive/YYYY-MM-DD/HHMMSS/<original-relative-path>.

Approved Artifact Handoff

After an approved synthesized write, explicit write/update mode, or any direct artifact mutation:

• List every created or updated synthesized artifact path in the final response.
• State the verification performed, such as readback, schema/check command, or why no executable verification applies for a Markdown-only strategy artifact.
• Check and report the relevant git status for intended artifacts when the project is a git repository. If intended artifacts are modified or untracked, make the next action shipping, committing, or an explicit dirty-artifact handoff before recommending downstream strategy work.
• Do not imply the research workflow is complete while approved artifacts remain untracked or uncommitted unless the user explicitly asked not to ship.
• If stopping for approval before writing, the approval request remains the next action; do not include downstream routing.

Intent-Aware Routing

Before applying the default ## Next-Skill Routing sequence, classify the user's immediate intent and route to the missing action that best serves that intent:

• Strategy refresh: recommend the missing or stale positioning, programming, portfolio, metrics, or product-media artifact.
• Recording prep: recommend the missing series spec, script, build proof, walkthrough guide, or validation artifact needed before recording.
• Upload prep: recommend packaging, title/thumbnail, description, chapters, or final metadata work before broader strategy work.
• Performance review: recommend metrics, cadence, portfolio, peer benchmark, or owner-analytics export work before new content planning.
• Owner analytics or private/manual platform evidence: route to an explicit manual/guide handoff instead of inventing unavailable metrics.
• Dirty intended artifacts: route to shipping/commit/handoff first, not another creator strategy skill.

Use the default next-skill sequence only when no stronger user intent, missing artifact, manual blocker, or dirty-artifact handoff applies.

Alignment Page

When this skill produces durable deliverables (research, specs, plans, reports, prototypes, or any document output), build a full-depth HTML alignment page following ALIGNMENT-PAGE.md in this skill's directory. Output: alignment/youtube-vid-research-{topic}.html.

Default Shipping Contract

Follow the shared shipping contract convention in CLAUDE.md.