lumeierecollection-blip

deep-research

Use before writing any video script or picking a topic. Runs multiple search passes across general web, news, and social/community sources (including what other YouTube automation creators recommend), then returns a findings summary for human review — never a finished script.

lumeierecollection-blip 0 Actualitzat fa 2 mesos
GitHub

Instal·la

npx skillscat add lumeierecollection-blip/youtube/deep-research

Instal·la des del registre de SkillsCat.

SKILL.md

Deep Research Skill

Purpose

Every topic must be backed by real, current, multi-source research before
any script is written. This skill never invents facts and never writes the
final script — it gathers and summarizes, then hands off for human
clarification.

Process (minimum 3-4 distinct search passes per topic)

  1. Pick a topic that has not been used before. Before searching, check
    the channel's dedup log:

    node src/utils/topic-log.cjs <channel-id> list
    node src/utils/topic-log.cjs <channel-id> pick "<candidate topic>"

    If data/research/<channel-id>/next-topic.json exists, that topic was
    already reserved by the pipeline — research THAT one. Never research or
    recommend a topic already in used_topics (the pick command exits
    non-zero for duplicates). After the topic is confirmed, reserve it:

    node src/utils/topic-log.cjs <channel-id> reserve "<topic>"
  2. Pass 1 — broad web pass. General search on the topic/niche to
    understand the current landscape and any recent news.

  3. Pass 2 — narrow/factual pass. Re-search with more specific terms
    surfaced from pass 1 (names, numbers, events) to verify and deepen facts.

  4. Pass 3 — practitioner pass. Search specifically for what other
    people already running YouTube channels or automation in this niche are
    doing/recommending — creator forums, YouTube-automation communities,
    niche-specific "how I grew this channel" posts.

  5. Pass 4 — social/community pass. Search Reddit, X/Twitter, or niche
    forums for what people are actually discussing about this topic right
    now — this is where a genuine "why would a stranger care" hook usually
    comes from.

Each pass should use different search terms than the last — don't repeat
the same query. If a pass turns up a contradiction or gap, run an extra
pass to resolve it before moving on.

Output format (every time)

Return to the user:

  • Topic
  • Key facts found (each tagged with its source)
  • Strongest angle/hook based on what's actually trending or discussed
  • What practitioners in this niche are doing (if relevant)
  • Open questions or gaps — call these out explicitly rather than
    filling them in

Do not proceed to script-writing until the user replies with edits or
confirmation.

Rules

  • Never state a fact that wasn't found in an actual search result this
    session.
  • Never combine research passes into a single generic search — the whole
    point is different angles catching different things.
  • If sources conflict, say so; don't silently pick one.