Skip to main content
thl-open is the open-source skill suite from Tech Horizon Labs, the team behind Are you found by AI?. It ships 17 skills that run inside Claude Code, Codex, Cursor, Gemini CLI or any agent that reads SKILL.md. This guide covers the two you are most likely to start with, found-by-ai and geo-audit, plus the fix-and-re-measure loop for a monitored site.
Only the readiness method and the audit suite are open source. The measurement engine behind Are you found by AI?, including its sampling and regression evaluation set, is not in the repo. The found-by-ai skill calls the hosted scan; it does not contain it.

​
Install the skill suite

Run this command in your terminal to install the full suite:
This installs all 17 skills. To install the suite as a Claude Code plugin instead, run these as two separate commands (paste the first, press Enter, then paste the second):
The skills are open source under MIT. Read the source for every skill at github.com/techhorizonlabs/thl-open before running it: skills can run code and fetch the open web.

​
The found-by-ai skill

The found-by-ai skill connects your agent to the free scan at areyoufoundbyai.com. It runs a live scan for any URL and returns the result, with no account or authentication required. What it does: Calls POST https://areyoufoundbyai.com/api/scan with your URL, waits for the engines to answer (about 60 seconds), and returns structured JSON your agent can read and act on. How to use it: ask in plain language, for example:
What it returns (main fields):
Always share the report URL with your client or stakeholder. It opens the readable report, with the answers behind the result, without needing any technical context.
The free scan asks 2 buyer questions on 2 engines (ChatGPT and Gemini). The scan API is rate-limited per IP, and fair use is up to 15 free scans per browser per day. For seven engines, Pro runs checks of up to 25 buyer questions on the schedule you choose, spending from a shared token balance (14,000 tokens a month for US$79; a 25-question check plus a fresh site read uses 3,150). See the found-by-ai skill page for the paid per-call endpoints and MCP.

​
The geo-audit skill

The geo-audit skill runs a parallel audit of your site across six GEO dimensions, using five subagents, and produces a composite GEO Score (0-100) with per-dimension findings.
The geo-audit composite is thl-open’s own audit score. It is not the AI Readiness score in your Are you found by AI? report, which is composed of five site signals with Footprint scored separately. The two use different inputs and weights, so do not compare them number for number.
What it does: Launches five subagents to evaluate six dimensions. Subagent 1 returns both AI Citability and Brand Authority, while subagents 2 to 5 handle Platform Optimization, Technical GEO, Content E-E-A-T, and Schema and Structured Data respectively. Each returns a score with a provenance tag and evidence-backed findings, and the orchestrator assembles them with a fixed formula. How to use it:
Output: A file named GEO-AUDIT-REPORT.md in your working directory, containing:
  • The composite GEO Score with its rating band
  • Per-dimension scores with provenance tags and the specific issues found
  • Prioritised issues by severity, quick wins and a 30-day action plan
  • An appendix of the pages analysed
The report is written to a Markdown file so you can commit it to your repository, share it with teammates, or feed it to audit-report-kit for a branded PDF.

​
The fix-and-re-measure loop

The loop is not one of the 17 thl-open skills. It is a separate skill file published at areyoufoundbyai.com/guides/ai-loop (agent-readable at /guides/ai-loop/skill.md), written for a site on Pro, where the deep measurement supplies the fix plan. It automates the fix, deploy, re-measure cycle over your site’s MCP connection, with you approving every change. What it does: Reads get_fix_plan, classifies each fix as in-repo (it applies these) or yours to do (DNS, directory listings, Google Business Profile), applies in-repo fixes on a branch, shows you the diff, deploys after your approval, calls request_rescan, polls get_visibility until the new measurement lands, and records the measured change for each fix shipped. How to use it: save the skill file in your site’s repository at .claude/skills/found-by-ai-loop/SKILL.md, connect your site’s MCP endpoint (see MCP Integration), then ask:
Before starting, the skill:
  1. Calls get_visibility to record your baseline scores and the time of the last full scan.
  2. Confirms protected-pages.txt exists beside the skill file, listing paths it must never edit (pricing, legal, checkout). It refuses to start without it.
  3. Agrees run parameters with you: target score, iteration cap and rescan budget.
Every change requires your approval before it deploys. This is a hard rule in the skill, with no exceptions, including changes the agent judges trivial. You review the full diff, approve, and only then does deployment proceed.
Re-measures are capped: on a subscription that keeps its earlier terms, request_rescan allows 5 per rolling 7 days, is refused when the freshest scan is under 20 minutes old, and queues one at a time. On a token account, request_rescan queues nothing and asks you to confirm the token price in your workspace, so agree the token budget for the run with the loop as well. The loop stops when any one of these conditions is met:
  • Your target score is reached
  • Two consecutive iterations move the targeted score by nothing
  • The iteration cap is hit
  • The rescan budget is spent
  • All remaining fix plan items are yours to do (off-repo work only)
At the end of each run, the skill produces a run report showing your baseline scores, final scores, each fix shipped with its measured change, the items on your list with reasons, and the rescan budget used.
See the full skill reference at /skills/overview for documentation on all 17 skills.