evidence-led-research

name: evidence-led-research description: Build an evidence log for a public figure, topic, or draft before writing. Use when researching personality analyses, verifying factual claims, collecting quote support, or checking whether a draft has enough source backing. argument-hint: '<person, topic, or file path>' context: fork agent: general-purpose disable-model-invocation: true

Evidence-Led Research

Use this skill when the job is to gather and validate evidence, not to draft the final piece.

If $ARGUMENTS is empty, ask for one of:

1. A person name
2. A topic
3. A file path

Examples:

/evidence-led-research Taylor Swift
/evidence-led-research src/blog/people/drafts/Paris-Hilton.md
/evidence-led-research enneagram and ADHD

Read First

Load the minimum relevant context first:

• docs/blogs-famous-people/prompts/research-prompt.md
• docs/content-analysis/fact-check-priority-2026-02-20.md
• docs/brand/README.md

If the target is a people draft, also search and load:

• src/blog/people/drafts/[Person].md
• youtube-transcripts-people/*review.md
• youtube-transcripts-people/*
• youtube-transcript-research/*

If the user passed a file path, read that file before searching elsewhere.

For the exact output shape, use:

• template.md
• example-output.md

Workflow

1. Resolve the target

• Direct file path: use it
• Person name: check src/blog/people/drafts/, transcript review docs, and related research docs
• Topic: search the repo for drafts, notes, transcripts, and prior analysis

2. Build a source stack

Prefer sources in this order:

1. Direct transcripts or primary interviews
2. Official statements, websites, filings, or published first-party materials
3. High-quality reporting with dates and attribution
4. Existing 9takes research docs

For every source, capture:

• source type
• date
• why it is credible
• what it is useful for

3. Separate claim types

Label findings before judging them:

• hard_fact: dates, numbers, awards, rankings, timelines, business facts
• direct_quote: exact words attributed to a person
• observed_behavior: specific public actions, clips, patterns, choices
• interpretive_bridge: the psychological meaning inferred from the evidence

Do not treat interpretive_bridge as fact.

4. Verify in priority order

Verify these first:

• unusual numbers
• dates and timelines
• direct quotes
• controversy claims
• rankings, records, or "first/most/biggest" language
• medical, legal, or diagnosis-adjacent claims

If a claim is weak, say so directly. If sources conflict, preserve the conflict instead of smoothing it away.

5. Build the evidence log

Use the supporting template and follow its section order exactly.

Rules

• Never invent citations, dates, or quote wording.
• Keep exact quotes exact. If unsure, paraphrase and mark it as paraphrase.
• Separate fact-checking from Enneagram interpretation.
• Treat transcript-backed evidence as stronger than generic profile summaries.
• If the evidence is thin, say the draft is under-supported.
• ultrathink when sources conflict or the evidence chain is weak.

Save Behavior

Do not write a file unless the user asks or the workflow clearly calls for it.

If saving is useful, default to:

• docs/content-research/[slug]-evidence-log.md