agentleFS
aya-talebaya-taleb/video-boost-skills/skills/source-check/SKILL.md

source-check

Skill1 starsChanged 22 days agoMITView on GitHub

Checks research notes before they become rules - every finding needs a working link, a publish date inside the time window (12 months by default, 6 with recent), the right market and language (US and English by default), and numbers that match their source. Use after shorts-research, reels-research or tiktok-research, or when the user says check these sources, verify this research, or are these sources recent.

7 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

8 rules. 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Open the link: Use the web fetch tool. If the page will not load, try once more.

    line 18
  2. Very specific:

    If the page does not support the claim, mark it not_supported and write what the page actually says.

    line 27
  3. Distinctive:

    No sources left: drop the finding.

    line 32
  4. Very specific:

    Only example sources: confidence low, and it can only become a nice rule.

    line 33
  5. Distinctive:

    Never fill in a missing date with a guess.

    line 40
  6. Distinctive:

    Never keep a number you could not find on the page.

    line 41
  7. Distinctive:

    Never upgrade a source type to make a finding pass.

    line 42
  8. Distinctive:

    If a whole platform ends up with fewer than 5 findings, say so in the report as a gap.

    line 43

“Common” is measured: kept by at least 6 in 10 of the 7 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 is only useful if it is true now, for this market and this platform. This skill checks every finding in a research notes file and drops or downgrades what fails.

Input: one or more notes-<platform>.json files (format in shared/checklist-format.md), plus the settings they were made with: market, language, window in months.

Output: the same files, cleaned, plus source-report.md that lists what was dropped and why.

Steps

For every finding, for every source in it:

  1. Open the link. Use the web fetch tool. If the page will not load, try once more. If it still fails, mark the source unreachable.
  2. Find the publish date on the page. Look for the article date, the post date, or "updated on". Use the date printed on the page, not the date in the search result. If there is no date anywhere, mark it undated. One exception: a platform's own live help or policy page (for example support.google.com/youtube, help.instagram.com, support.tiktok.com) shows the rule as it is today. Keep it with "published": null, "accessed": "<today>", "type": "official".
  3. Check the window. Today minus the window (12 months by default, 6 if the user asked for recent). A source older than that is too_old. An "updated on" date counts only if the part you quote was clearly part of the update.
  4. Check the scope. Does the source talk about this platform? This market (country)? This language? A global platform announcement is fine and is marked "scope": "global". A study of another country is wrong_market unless the finding says so.
  5. Check the claim against the page. The claim and every number in it must be on the page. Read the sentence around it. Watch for:
    • a number from a different year or a different platform on the same page
    • "up to" or "as much as" turned into an average
    • a creator survey reported as if it were platform data
    • an ad (paid) result reported as organic If the page does not support the claim, mark it not_supported and write what the page actually says.
  6. Rate the source type. official (the platform's own help pages, newsroom, blog, creator accounts, policy pages), data (a study or dataset that states its sample size and dates), analysis (an expert article), example (a video). Marketing pages that sell a tool count as analysis at best.

Then for every finding:

  • No sources left: drop the finding.
  • Only example sources: confidence low, and it can only become a nice rule.
  • One official source, or two independent non-official sources that agree: confidence high.
  • Otherwise: medium.
  • Two sources that disagree: keep the finding, set confidence medium at most, and add a conflict note with both sides.

Rules that never bend

  • Never fill in a missing date with a guess.
  • Never keep a number you could not find on the page.
  • Never upgrade a source type to make a finding pass.
  • If a whole platform ends up with fewer than 5 findings, say so in the report as a gap. Do not pad it with old or off-market sources.

Report format (source-report.md)

Source check: tiktok, travel tips, US, en, last 12 months (checked 2026-10-23)

Kept: 21 findings (9 high, 8 medium, 4 low)
Dropped: 6
  tt-03  too_old         creatoracademy page last updated 2024-11-02
  tt-09  not_supported   page says "up to 34% more reach", note said "34% more reach"
  tt-12  wrong_market    study of UK creators only
Conflicts: 1
  tt-15  ideal length: 21-34 s (source A, 2026-05) vs 60 s+ (source B, 2026-08)
Gaps: no 2026 official data on posting times for the US

Plain words in the report. Short lines. No source is "trusted" just because it ranks high in search.

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.