agentleFS
namedtoasternamedtoaster/The-Last-Mile/skills/fact-check/SKILL.md

fact-check

Skill39 starsChanged 23 days agoMITView on GitHub

Re-check stated facts about the App Store, Google Play and app tools against the current official pages, update each fact's status and date, fix broken source links, and report what changed. Works on a project's rules file and docs, or on The Last Mile's own traps and facts (one topic or all of them). Use when the user says "check the facts", "is this still true", "re-verify the traps", "store rules changed", or before a store submission.

38 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use deep-research, seo and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • NVIDIASkill

    deep-research

    Queue deep, multi-step research and analysis tasks to the DeepAgents worker. Supports execution depth presets, request-specific rubrics, SubAgent delegation, and RubricMiddleware cross-validation.

    Also in 4 projects that use this

  • nilssonrSkill

    seo

    Full SEO workflow for auditing pages and creating SEO-optimized content. Provides checklists with educational explanations aligned with Google Search Central documentation.

    Also in 3 projects that use this

  • BlockSkill

    frontend-design

    Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 9 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 31 edited versions we read keep these.

  • 9

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

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

    In The Last Mile repo or a fork: every file in skills/ship/references/ (traps, facts, stacks, domains, payments, levels, rulebooks) and the …

    line 17
  2. Distinctive:

    Use the same three statuses as the traps reference:

    line 27
  3. Distinctive:

    If all of these fail, leave the status as it was.

    line 58
  4. Distinctive:

    No official page states it: keep or set From experience.

    line 68
  5. Very specific:

    Fix the link: If the link is broken or redirects, replace it with the current official URL.

    line 70
  6. Distinctive:

    Read the saved copy of the vendor's page for that screen, and give the steps in its words.

    line 80
  7. Distinctive:

    Never upgrade a status from memory.

    line 90
  8. Distinctive:

    Never set a date on an entry you did not open.

    line 91
  9. Distinctive:

    Do not soften an entry that the page contradicts.

    line 92
  10. Distinctive:

    Keep it short. Rows where nothing changed can be grouped into one line if there are many.

    line 113
  11. Distinctive:

    Every entry in scope has been opened or reported as not loadable, every status and date in the files matches what was found today, and the …

    line 129

“Common” is measured: kept by at least 6 in 10 of the 31 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

Fact check

Facts about stores and tools go stale quietly. This skill re-reads the official page behind each fact and records what it found, so every status means what it says.

Pick the target

  • Inside a user's project: the facts their rules file and docs state about the stores and their tools (fees, tester counts, target API levels, review guidelines, console steps). Ask which files if it is not obvious. Only check statements that claim an outside rule or number; skip the project's own decisions.
  • In The Last Mile repo or a fork: every file in skills/ship/references/ (traps, facts, stacks, domains, payments, levels, rulebooks) and the factual claims in skills/ship/SKILL.md, including console steps. The user may name one topic ("Google Play testing", "account deletion") or ask for everything. An open "Re-check facts" issue names the files whose sources changed; start there.

List the entries you will check and how many there are before starting.

Status meanings

Use the same three statuses as the traps reference:

  • Checked YYYY-MM-DD (official page opened and read): the page was opened today and a sentence on it supports the entry.
  • From experience: it happened to a real app, but no official page states it.
  • Needs a human: the answer sits behind a console login no automation can read.
  • Seen on screen YYYY-MM-DD: the user's own console screen showed it (a screenshot or pasted text). The live screen beats a saved page: when they disagree, record what the screen showed, note that the official page lags, and keep the page as the source for anything it still states correctly.

In facts.md the status lives in the Last checked column: a date means Checked, otherwise write "From experience" or "Needs a human".

Check each entry

  1. Open the source page. Fetch the page itself. A search snippet or summary is not enough. If the page will not load (blocked, login wall, timeout), use these in order and say which one you used:
    • The Last Mile's saved copy. A weekly job saves a text copy of every page the ship references cite. Read sources/index.json in The Last Mile (https://raw.githubusercontent.com/namedtoaster/the-last-mile/snapshots/sources/index.json, on the snapshots branch), find the URL, and read sources/<file> from the same branch. GitHub is reachable from cloud sessions by default, so this needs no settings. The copy counts as the page opened on its fetched date: set Checked <fetched date> (saved copy of the official page). If fetched is more than a month old or the entry has an error, say so.
    • Another route to the same page: a different fetch tool, or the page in another language on the same official site.
    • A search with results limited to the official domain. That is a lead: an entry read only this way is Not yet checked, per CONTRIBUTING.md. If all of these fail, leave the status as it was. If the block comes from the session's network rules, name the host in one line at the end of the report, as an optional fix; never make the user change settings to get an answer.
  2. Find the supporting sentence. Quote it word for word in your notes. If the entry has no source, search the official docs (Apple, Google, Expo, the tool's own site) for one. Blogs, forums and other repos are leads, never sources.
  3. Decide:
    • The page supports the entry: set Checked with today's date.
    • The page contradicts it: write the corrected wording, show the old and new text with the quoted sentence, and wait for the user to agree before changing it.
    • No official page states it: keep or set From experience.
    • It sits behind a console login: set Needs a human, with what to look for.
  4. Fix the link. If the link is broken or redirects, replace it with the current official URL. Point it at the exact rule: use the anchor for the guideline section (for example #5.1.1v) or the specific help article, not a home page. Open the new link to confirm it lands on the rule.

Console steps

Steps through a console (App Store Connect, Play Console, RevenueCat, a registrar) are facts too, and the most likely to drift. Before giving or writing any:

  • Read the saved copy of the vendor's page for that screen, and give the steps in its words. Say which steps come from the page and which from memory.
  • Name the fields the user will see in order, so a missing or extra field shows up at once. When the user reports a different screen, ask for the exact text or a screenshot, correct the reference the same day, and mark it Seen on screen.
  • One wrong step usually means the page is old: re-check the other steps from the same page before carrying on.

Hard rules

  • Never upgrade a status from memory. Checked needs a page opened today.
  • Never set a date on an entry you did not open.
  • Do not soften an entry that the page contradicts; correct it or flag it.
  • Change only the status, date, link and, once agreed, the wording. Keep each entry's shape (see CONTRIBUTING.md, "Adding a trap", in The Last Mile).
  • Plain words in any wording you write. No long dashes.

Report

Reply in chat with a short table, one row per entry checked:

EntryOld statusNew statusWhat changed
Testers for a new personal accountChecked 2026-07-02Checked 2026-10-11Nothing
Account deletion in the appChecked 2026-07-02Checked 2026-10-11Link now points at the guideline anchor

Then:

  • Proposed rewrites, each with the old text, the new text and the quoted sentence.
  • Pages that would not load, and entries marked Needs a human, so the user knows what is still unconfirmed.
  • The edits made, as a list of files.

Keep it short. Rows where nothing changed can be grouped into one line if there are many.

When to run it

  • Before a submission: the ship skill should run this on the traps that apply to the release (iOS, Android, or both) before preparing the submission.
  • When a rule seems to have moved: a rejection names a guideline the traps do not match, or the user hears a store rule changed.
  • When a saved copy changes: the weekly snapshot job (.github/workflows/snapshot-sources.yml) commits every page whose text changed. A changed or failed page opens a "Re-check facts" issue naming the reference files that cite it; run this skill on those files and close the issue with the result.

Done when

Every entry in scope has been opened or reported as not loadable, every status and date in the files matches what was found today, and the user has seen the table.

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.