llms.txt Analysis & Generation Skill
Validates or generates llms.txt files so AI search engines can understand your site's structure and key facts at a glance.
Skill declares only Read/Grep/Glob/Bash/WebFetch/Write; operations are fetching public pages, validating, and writing Markdown reports — no sensitive-data collection or destructive defaults; repo provides uninstall.sh and an isolated venv. However, SKILL.md itself lacks confirmation gates, write scoping, and rollback notes; Bash grant is broad and data-flow disclosure incomplete, so mid-range score.
Steps are well-structured with HTTP status branches (200/404/403/301/302) and a validation checklist; the happy path is plausible. But no test or script evidence covers this skill's llms.txt validation/generation path (tests/ only cover fetch_page SSR heuristics); failure-feedback quality on abnormal input is unverified, so below the static cap of 10.
Scenario is clear (analyze or generate llms.txt), trigger semantics map to /geo llmstxt, and format rules are detailed. But non-fit boundaries (JS-rendered sites, login-walled pages) are undeclared, docs are English-only with no Chinese support; core function depends on fetching the target site, no fully unreachable overseas dependency found.
SKILL.md is well-layered (spec, analysis mode, generation mode, output templates, best practices) and MIT-licensed; however the skill has no version, changelog, or known-limitations section, the README is heavy on unsourced marketing metrics and Skool community promotion, and maintenance responsibility/update path is unclear — deducted.
Output templates (analysis report, generated llms.txt) are concrete and directly usable, with explicit scoring weights; but the actual benefit of llms.txt for AI citation is unverified ('fewer than 5% of sites' claims are unsourced), and under static review output correctness is unconfirmed by execution — deducted.
The spec and steps are auditable and partially manually checkable; the repo has one genuine test file but it covers only fetch_page SSR, not this skill's key paths; no third-party execution evidence or cross-source corroboration; the score weights (0.40/0.35/0.25) are author-defined, capped below 5 statically.
- Static review only; no code executed. The skill's key paths (llms.txt validation/generation) have no test coverage evidence.
- SKILL.md grants Bash and WebFetch with no confirmation requirement before writing files; review output locations before use.
- Market metrics in the docs (e.g. '+527% AI traffic', 'fewer than 5% of sites have llms.txt') are unsourced marketing claims and should not be treated as fact.
- Docs are English-only with no declared support for Chinese sites or content; llms.txt itself remains an emerging proposal and its real-world benefit on AI platforms is unverified.
- Publisher is unverified/unknown; the repo carries Skool community commercial promotion — be aware of incentive bias.
What does this skill do, and when should you use it?
geo-llmstxt is a sub-skill of the geo-seo-claude repository focused on the llms.txt standard (proposed by Jeremy Howard in September 2024). It can fetch and validate an existing llms.txt or llms-full.txt, or crawl a site's homepage, navigation and sitemap.xml to generate a new file from scratch. It ships with a full format specification, a weighted scoring model (Completeness 40%, Accuracy 35%, Usefulness 25%), and structured Markdown output reports. It is part of an MIT-licensed, free collection installed with a one-command script.
In analysis mode: uses WebFetch to retrieve [domain]/llms.txt and /llms-full.txt, records HTTP status codes (200/404/403/301/302), validates each structural element (H1 title, blockquote description, H2 sections, page entries, absolute URLs, Key Facts, Contact sections), and scores the file on three weighted dimensions. It also crawls the site's navigation and sitemap to flag important pages missing from llms.txt and stale entries. In generation mode: fetches the homepage to extract site name, description and navigation, reads sitemap.xml, selects and prioritizes 10–30 key pages, writes 10–30-word factual descriptions per page (fetched live), assembles Key Facts, and compiles a spec-compliant llms.txt, validating URL reachability and file length (50–150 lines) before output. Produces GEO-LLMSTXT-ANALYSIS.md or a ready-to-deploy llms.txt plus a generation report.
- A site owner who wants to verify an existing llms.txt is spec-compliant with valid URLs before or after deployment
- A GEO/SEO agency that needs a client-deliverable llms.txt analysis report as part of an AI visibility audit
- A SaaS or e-commerce site with no llms.txt that wants one auto-generated from a real crawl of the site
- A content team checking for missing or stale page entries after a site redesign
- A business that wants AI assistants like ChatGPT and Perplexity to cite the correct pages and reduce hallucinations about its facts
What are this skill's strengths and limitations?
- Extremely thorough coverage of the llms.txt / llms-full.txt spec, with per-element validation tables and a quantified scoring model
- Supports both analysis and generation workflows, and generates from real crawled content rather than guesswork
- Produces structured Markdown reports suitable for direct client delivery
- MIT-licensed and free; installs with the whole collection in one command, with isolated dependencies that don't touch system Python
- Real-world benefit depends on AI systems actually reading llms.txt — under 5% of sites had one as of early 2026, and citation gains lack third-party validation
- The scoring weights (40/35/25) are the author's own; no public test suite or benchmark evidence is provided
- Relies on page-by-page WebFetch for large sites, which can be slow and limited by target-site anti-crawler measures
- No documented standalone installation for this sub-skill; you must install the full 16-skill collection
- The collection's CRM/reporting features write data to ~/.geo-prospects/, which the uninstaller does not remove
How do you install this skill?
The skill installs with the full geo-seo-claude collection: on macOS/Linux run curl -fsSL https://raw.githubusercontent.com/zubair-trabzada/geo-seo-claude/main/install.sh | bash; on Windows use install-win.sh from Git Bash. Prerequisites: Python 3.8+, Claude Code CLI, Git. Dependencies go into an isolated venv at ~/.claude/skills/geo/.venv/. The skill itself lives at skills/geo-llmstxt/SKILL.md; the source does not document installing this sub-skill standalone.
How do you use this skill?
After installing the collection, run /geo llmstxt <url> in Claude Code against the target site: if llms.txt exists it enters analysis mode and outputs GEO-LLMSTXT-ANALYSIS.md; if not, it enters generation mode and outputs a new llms.txt. You can also just describe the task in conversation (e.g. "generate an llms.txt for example.com"). After generation, upload the file to the site root and verify https://yourdomain.com/llms.txt is reachable without redirects.
How does this skill compare with similar options?
The source names no direct competitors, but the skill sits in the emerging llms.txt tooling category; the repo positions itself as a GEO (generative engine optimization) complement to traditional SEO practices like robots.txt and schema markup.