Llms Txt Audit

llms.txt Audit

Audit a website's /llms.txt file and produce an actionable report for AI/LLM discoverability.

When to use this skill

Use this skill when the user asks to:

• Audit, validate, or check an llms.txt or llms-full.txt file
• Make a site discoverable and usable by AI assistants, chatbots, and LLMs
• Find broken, redirecting, or blocked links inside llms.txt
• Confirm llms.txt follows the llmstxt.org structure (H1 title, summary, link sections)
• Decide what should go into llms.txt vs. llms-full.txt

Background: what llms.txt is

llms.txt is a proposed standard (https://llmstxt.org) for a markdown file at the site root (/llms.txt) that gives LLMs a concise, curated map of the site's most useful content. Unlike robots.txt (which controls crawling) or sitemap.xml (which lists every URL), llms.txt is hand-curated and human-readable. Its structure:

# Project Name              ← required: the only mandatory element

> A short summary.          ← recommended blockquote summary

Optional free-form markdown with key context (no headings).

## Docs                     ← H2 section of curated links
- [Quickstart](https://example.com/quickstart): How to get started.
- [API reference](https://example.com/api): Full API docs.

## Optional                 ← an "Optional" section may be skipped by LLMs with limited context
- [Changelog](https://example.com/changelog): Release history.

A companion /llms-full.txt may contain the full expanded content in one file. This skill audits structure and link health against that spec.

Inputs to collect

Before starting, confirm with the user:

1. Site URL or llms.txt URL — a live site root (e.g., https://example.com, which fetches /llms.txt) or a specific llms.txt URL (--llms).
2. Scope — how many referenced links to probe (default 300, capped with --max-urls).
3. Probing — whether to fetch each link to check status/redirect/noindex (default on). Use --no-probe for a fast structure-only pass.

Workflow

Step 1: Fetch and parse llms.txt

Use scripts/collect_llms.py:

python3 scripts/collect_llms.py https://example.com --max-urls 300 --output llms_inventory.json

Point directly at an llms.txt with --llms:

python3 scripts/collect_llms.py https://example.com/llms.txt --llms --output llms_inventory.json

The script fetches /llms.txt without auto-following redirects (so a redirected or HTML-served file is visible), records status / content type / size, parses the H1 title, blockquote summary, and H2 sections with their list items, then probes every referenced link (status, redirect target, canonical, meta robots, X-Robots-Tag, robots.txt disallow). It also checks whether /llms-full.txt exists.

Step 2: Run the audit checks

Use scripts/audit_llms.py:

python3 scripts/audit_llms.py llms_inventory.json --output audit_report.json

Step 3: Evaluate the audit checks

Evaluate each check defined in references/audit-checks.md. The core checks are:

| Check | Severity | |---|---| | /llms.txt missing (non-200) | High | | Served as HTML, not markdown/plain text | High | | Missing required H1 title | High | | Referenced link returns 4xx/5xx/error | High | | /llms.txt itself redirects | Medium | | Referenced link redirects (not a final 200) | Medium | | Missing blockquote summary | Medium | | Link list item missing a description | Medium | | Malformed list item (not a markdown link) | Medium | | Link on the wrong host | Medium | | Linked page is noindex or robots-blocked | Medium | | Wrong content type | Medium | | Relative URL instead of absolute | Low | | No sections / no links | Low | | Duplicate link | Low | | /llms-full.txt missing | Low |

Step 4: Produce the report

Write a report following the structure in references/report-template.md:

1. Summary — file status, H1, sections, links found/probed, issue counts by severity
2. File structure — H1, summary present, section list, llms-full.txt presence
3. Issues — one section per failing check, listing the link/item, the problem, and the fix
4. Prioritized action list — ordered by severity and impact

Step 5: Recommend fixes

For each issue, give a concrete fix:

• Missing file: create /llms.txt with at minimum an H1 title, a one-line > summary, and one or more ## sections of curated - [name](url): description links. Serve it as text/markdown or text/plain at the root with a 200.
• Served as HTML / wrong content type: serve the raw markdown file, not a rendered page; set Content-Type: text/markdown (or text/plain).
• Missing H1: add a single top-level # Project Name heading — it is the only required element.
• Broken / redirecting / blocked links: replace each with its final, canonical, indexable 200 URL (or remove it). Links in llms.txt should point straight to the live destination.
• Missing descriptions / malformed items: rewrite each item as - [Descriptive title](absolute-url): one-line description so an LLM can tell what each link is for.
• Relative URLs: use absolute https:// URLs so the file works when fetched out of context.
• llms-full.txt: if the user wants full-content ingestion, generate /llms-full.txt with the expanded page contents.
• Base every recommendation on the observed file; quote the exact line to change and do not invent URLs.

Optional: export the report

As a Word document (.docx)

If the user wants the report as a .docx (for example, to share with stakeholders or attach to a ticket), save the Markdown report to a file and convert it:

python3 scripts/md_to_docx.py report.md --output report.docx

scripts/md_to_docx.py uses only the Python standard library (no pip install) and renders headings, tables, lists, links, bold/italic, and code blocks. Offer this whenever a user asks for a Word doc, a .docx, or a shareable/downloadable report.

As a CSV of findings (.csv)

If the user wants the raw findings as a spreadsheet (for filtering, sorting, or triage in Sheets or Excel), convert the audit's audit_report.json directly:

python3 scripts/findings_to_csv.py audit_report.json --output findings.csv

scripts/findings_to_csv.py is also standard-library only. It writes one row per finding, with check and severity columns prepended and list fields (e.g. the pages sharing a duplicate value) joined with ; . Unlike the .docx, which reformats the written report, the CSV is a direct dump of the structured findings — offer it whenever a user wants the data itself, a spreadsheet, or to slice findings by check or severity.

Resources

• scripts/md_to_docx.py — convert the Markdown report into a Word (.docx) document (standard library only)
• scripts/findings_to_csv.py — flatten the audit findings JSON into a CSV, one row per finding (standard library only)
• references/audit-checks.md — full definitions and rationale for every audit check
• references/report-template.md — report output structure
• scripts/collect_llms.py — fetch, parse, and probe a site's llms.txt
• scripts/audit_llms.py — run audit checks against an llms.txt inventory JSON file

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: JakeLabate
• Source: JakeLabate/Claude-SEO-Skills
• License: MIT
• Homepage: https://www.jakelabate.com/claude-seo-skills/

Install and usage instructions live in the source repository linked above.