agentleFS
frankxaifrankxai/geninvestor-skills/skills/source-check/SKILL.md

source-check

Skill0 starsChanged 10 days agoApache-2.0View on GitHub

Trace every number in a piece of research back to a primary source and check it. Use when reviewing a memo, an AI-written summary, a screen result, or any claim that contains figures, dates or quotes.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Prefer the original: a filing (annual and quarterly reports, ESEF or XBRL data), a central bank or statistics office, an exchange, the …

    line 16
  2. Distinctive:

    Never fill a gap from memory.

    line 52
  3. Distinctive:

    Never round or adjust to make a number match.

    line 53

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Source check

Research goes wrong at the number level: a stale figure, a wrong unit, a restated value, a quote that was paraphrased. This skill checks each one against where it came from.

Workflow

1. List every claim that can be checked

Pull out each number, date, ratio and quotation, with the sentence it sits in.

2. Find the primary source

Prefer the original: a filing (annual and quarterly reports, ESEF or XBRL data), a central bank or statistics office, an exchange, the company's own investor pages. An article about a figure is a pointer to the source, not the source.

3. Record the evidence

For each claim: the value, the unit, the period the value describes (as of), when you retrieved it, and exactly where it sits (URL, page, table, field).

4. Compare

Mark each claim match, mismatch, cannot find or derived. Look for the usual traps:

  • percent versus percentage points;
  • thousands versus millions, and the currency;
  • fiscal year versus calendar year;
  • restated figures: for a historical question, use the value that was known at that time;
  • adjusted versus unadjusted prices;
  • delayed data presented as current.

5. Recompute what was derived

Growth rates, margins, ratios, per-share figures: recompute them from the inputs and compare.

6. Check quotations

A quote must match the source word for word. A paraphrase is not a quote and must not sit inside quotation marks.

7. Give a verdict

The research passes only if every number is matched or correctly derived. Otherwise list what failed and what was not found.

Output

| # | Claim | Value in text | Value in source | Unit | As of | Source location | Result |
|---|---|---|---|---|---|---|---|

**Verdict:** pass | fail
**Not found:** <claims with no primary source>
**Data age:** <oldest as-of date and what it feeds>

Rules

  • Never fill a gap from memory. If you cannot find the source, write "cannot find".
  • Never round or adjust to make a number match.
  • Secondary sources may point you to the primary one but never settle a claim.
  • Say how old the data is. Stale data that feeds a conclusion is a finding, not a footnote.
  • One unsourced number is enough to fail the check.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.