Skillbook
Research · № R19
Pro
Preview cut from the real Pro body

Build a source dossier

Turn links, reports, and notes into a source-ranked dossier with claims, caveats, and next checks.

For
Researchers, writers
Time per use
18 min
Format
.md and .skill
How to use it
  1. 1.
    Open Claude or ChatGPT.
    Either works. The skill is just text.
  2. 2.
    Inspect the real preview, then unlock the full file.
    One click; no install, no setup.
  3. 3.
    Paste it as your first message.
    The assistant now knows how to do this one job.
  4. 4.
    Give it your specifics, get the result.
    Roughly 18 min, every time you need it.
Example output

Give it context. Get back a work product.

Illustrative sample using the same output shape. Verify live facts in the tool you run it in.

Give it

Assess whether the claim '60% of small businesses now use AI weekly' is credible before it goes into a founder memo.

Why it matters

Basing a memo, article, or sales narrative on a viral statistic that collapses once someone asks for the original source.

Get back

Source-ranked dossier

  • Best source to find first: the original survey or data release, not articles repeating the statistic.
  • Trust issue: sample size, respondent definition, survey sponsor, geography, and date matter more than the headline number.
  • Current confidence stays medium until the original questionnaire and respondent base are verified.
Get back

Claim map

  • Safe claim: 'A recent survey reported weekly AI usage among a sampled group of small businesses' if the source checks out.
  • Unsafe claim: 'Most small businesses use AI weekly' unless the sample is representative and the definition of use is clear.
  • Next check: retrieve the original report, methodology note, exact wording, and any sponsor incentives before publishing.
r19-build-a-source-dossier.skill.md2.1 KB
Run once

Fill the blanks first.

These fields update the skill preview and the Claude/ChatGPT buttons instantly.

IncludeWhat should this dossier help answer?
IncludePaste links, excerpts, documents, notes, or source names.
IncludeArticle, memo, investment, product decision, policy question, sales brief, or general research.
IncludeFormal citations, quick links, page numbers, source quality notes, or none.
IncludeDate range, geography, source types to prefer or avoid, time limit, or known bias.
Install as agent behavior

Permanent agent install needs the full body.

This page is only showing a preview. Unlock the full skill to install it in Claude Code, Claude Projects, or a Custom GPT.

Unlock full skill →
# Build a source dossier

You are a careful research librarian. I will give you links, excerpts, reports, notes, or claims. Build a source-ranked dossier I can use before writing, deciding, publishing, or briefing someone.

## Inputs

Research question: {{research_question||What should this dossier help answer?}}
Sources: {{sources||Paste links, excerpts, documents, notes, or source names.}}
Decision context: {{decision_context||Article, memo, investment, product decision, policy question, sales brief, or general research.}}
Citation needs: {{citation_needs||Formal citations, quick links, page numbers, source quality notes, or none.}}
Constraints: {{constraints||Date range, geography, source types to prefer or avoid, time limit, or known bias.}}

## Output

1. **Answer preview:** the current best answer and confidence level.
2. **Source table:** source, type, proximity to original, useful claims, limitations, and trust rating.
3. **Claim map:** which claims are well supported, weakly supported, or contradicted.
4. **Caveats:** incentives, outdated evidence, missing primary sources, or sampling issues.
5. **Next checks:** the exact sources, searches, or records to verify next.

## Quality standard

This should feel like the evidence pack behind a serious memo. The user should be able to see:

- Which sources are original versus derivative.
- Which claims are safe to use now.
- Which claims need one more verification step.
- Which sources are useful background but weak evidence.
- Where the current answer could still be wrong.

## Example output shape

For a question like "Is this statistic about AI adoption in small businesses credible?":

- Put the original survey or data release above blog posts that repeat it.
- Mark sample size, date, geography, and sponsor incentives.
- Separate "the statistic is accurately quoted" from "the statistic supports the argument."
- End with the exact original source or data table to retrieve before publishing.

## Rules

[Preview stops here. Unlock the Pro library for the full rules, guardrails, examples, and copyable file.]
The rest is in the Pro library.

This preview is cut from a real Pro workflow. Unlock the founding Pro library for the full file, rules, examples, and installable skill.

Full Pro file includes
  • Input checklist
  • Step-by-step workflow
  • Quality bar
  • Guardrails
  • Output format
  • Example run
  • Install formats
Unlock the library →