agentleFS
frontmost-aifrontmost-ai/aeo-skills/skills/answer-accuracy/SKILL.md

answer-accuracy

Skill1 starsChanged 3 days agoMITView on GitHub

Build a spreadsheet of each AI answer that names the business, with the exact words, tone, other businesses named, and cited pages, so that a person can mark what is wrong. Use for "what does ChatGPT say about me" or a wrong description. Needs a Frontmost AI project.

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. About 0 are common, 9 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this guidance to list each saved answer that names the business, so that a person can read what AI apps say and mark what is wrong.

    line 8
  2. Very specific:

    If a tool returns setup_required, give the person the url from the result.

    line 14
  3. Very specific:

    If a tool returns plan_required, say that this data needs a paid Frontmost AI plan, and stop.

    line 15
  4. Very specific:

    If the Frontmost AI tools are not available, use the mention-check skill.

    line 16
  5. Distinctive:

    Never invent the data, and never fill a gap with a guess.

    line 17
  6. Very specific:

    Call get_scan for the project and the scan in the request.

    line 21
  7. Very specific:

    Call get_answers with the project_id, the scan_id, named: true, and limit: 20.

    line 22
  8. Very specific:

    Do step 2 again with questions: "unbranded".

    line 23
  9. Distinctive:

    Add the columns that the person fills in: Accurate or Wrong, What Is Wrong, Page That Should Correct It.

    line 41
  10. Distinctive:

    Write a short note only when the excerpt states a fact that the person can check: a price, a customer type, a product type, a location, a …

    line 46
  11. Distinctive:

    If the person gave you their site or their facts in this conversation, you can say which fact differs and quote both.

    line 48
  12. Distinctive:

    Leave the cell empty for an answer with no checkable fact.

    line 49
  13. Distinctive:

    Do steps 1 to 3 again for the latest scan.

    line 53
  14. Distinctive:

    Add the new rows above the old rows.

    line 55
  15. Distinctive:

    Ask before you change anything in Frontmost AI.

    line 64

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

Answer accuracy

Use this guidance to list each saved answer that names the business, so that a person can read what AI apps say and mark what is wrong. A chart cannot show a wrong fact. A person must read the words.

Before you start

This skill reads a Frontmost AI project through the Frontmost AI tools that this plugin connects.

  • If a tool returns setup_required, give the person the url from the result. Stop until they say that set-up is done.
  • If a tool returns plan_required, say that this data needs a paid Frontmost AI plan, and stop.
  • If the Frontmost AI tools are not available, use the mention-check skill. It needs no account, and it records the facts to check in a small set of answers. To read each recorded answer, the person connects Frontmost AI: https://frontmost.ai/connect
  • Never invent the data, and never fill a gap with a guess.

Build the sheet

  1. Call get_scan for the project and the scan in the request. Read metric_breakdown.
  2. Call get_answers with the project_id, the scan_id, named: true, and limit: 20. Call again with each next_cursor until it is null.
  3. Do step 2 again with questions: "unbranded". An answer id in this second list is from an unbranded question.
  4. Make one row for each answer. The row key is the answer id. Put it in the last column and never change it.
  5. Fill the Frontmost AI columns:
ColumnValue
Scan Datethe date of the scan
AI Appapp
Questionquestion.text
Branded"Unbranded" or "Branded", from step 3
What It Saysexcerpt, copied word for word
Sentimentsentiment
Positionposition
Others Namedthe names in mentions where you is false
Cited Pagesthe url values in citations. Mark the ones where own_site is true
Check Thissee below
Linkurl
  1. Add the columns that the person fills in: Accurate or Wrong, What Is Wrong, Page That Should Correct It. Leave them empty.
  2. Above the table, write three lines from metric_breakdown: all questions, branded, and unbranded. Each line gives visibility.named of visibility.answers.

The Check This column

  • Write a short note only when the excerpt states a fact that the person can check: a price, a customer type, a product type, a location, a feature, or a different business with a similar name.
  • Quote the words that hold the fact. Do not say that the fact is wrong. You do not know the business as the person does.
  • If the person gave you their site or their facts in this conversation, you can say which fact differs and quote both.
  • Leave the cell empty for an answer with no checkable fact.

Refresh the sheet

  1. Do steps 1 to 3 again for the latest scan.
  2. Match each answer to a sheet row by the answer id. An answer id belongs to one scan, so most rows of a new scan are new.
  3. Add the new rows above the old rows. Never delete an old row or its typed cells.
  4. Rewrite the three count lines.

Rules

  • Copy the excerpt. Never rewrite it, and never join two answers.
  • To read a full answer, call get_answers with answer_id.
  • Report each rate with its counts. Keep branded and unbranded apart.
  • Never tell the person to ask an AI app to change an answer, to post fake reviews, or to expect a correction by a date. The usual correction is a clear statement of the fact on the person's own page and on the pages that the answer cites.
  • Ask before you change anything in Frontmost AI. This sheet reads only.

Output

A spreadsheet file with one table, the three count lines above it, and the scan date. Then list the answers with a Check This note, in the order of the sheet. If the AI app cannot make a file, return the table as CSV.

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.