geoharness.toml reference
GeoHarness reads one file, geoharness.toml, from your site directory. Keys are camelCase, unknown keys are errors, and policy choices such as Content Signals are never defaulted.
Which keys does geoharness.toml accept?
| Key | Type | Required | Default (reason) | Validation |
|---|---|---|---|---|
site.origin | string | yes | — | https://host[:port], no path, query or trailing slash |
site.name | string | yes | — | non-empty, single line (it is the llms.txt H1) |
site.language | string | yes | — | BCP 47 tag (en, en-US) |
identity.kind | "person" | "organization" | yes | — | |
identity.name | string | yes | — | non-empty, single line |
identity.url | string | yes | — | absolute http(s) URL |
identity.sameAs | string[] | no | [] (no profiles is a valid statement, not an assumption) | absolute http(s) URLs; injected as rel=me and JSON-LD sameAs |
identity.organization | table { name, url } | no | absent (a person need not belong to one) | both keys required when present |
contentSignal.search | bool | yes | — (a policy choice; never defaulted) | |
contentSignal.aiInput | bool | yes | — | |
contentSignal.aiTrain | bool | yes | — | |
bots | array of tables | yes | — | at least one; a token may appear in only one policy (case-insensitive) |
bots[].tokens | string[] | yes | — | user-agent tokens, no whitespace or :; * allowed |
bots[].group | "search" | "user" | "training" | "other" | yes | — | search/user tokens are checked by the audit's blocking robots check |
bots[].allow | string[] | no | [] (a group may be disallow-only) | each starts with / or * |
bots[].disallow | string[] | no | [] (a group may be allow-only) | each starts with / or * |
content.mainSelector | string | yes | — | valid CSS selector; when a page has no match its <body> is used with a page.no_main warning |
content.exclude | string[] | no | [] (excluding nothing changes nothing) | route globs starting with / |
llms.summary | string | yes | — | single line (the llms.txt blockquote) |
llms.sections | array of tables | yes | — | may be empty ([]) for an llms.txt with no sections |
llms.sections[].title | string | yes | — | single line (an H2) |
llms.sections[].routes | string[] | yes | — | route globs starting with / |
env | string[] | no | [] (no template may read the environment unless a name is declared) | [A-Z_][A-Z0-9_]*, unique |
headers.immutable | string[] | no | [] (no asset is assumed to be content-hashed) | Cloudflare _headers URL patterns starting with / |
prerender | table | no | absent (only sites built with a prerender adapter need it) | |
prerender.routes | string[] | yes, when [prerender] is present | — | literal absolute routes (no globs, query or fragment); read by the host's vite-prerender adapter, validated by the core |
Globs
content.exclude and llms.sections[].routes are matched against page routes: * matches within one path segment, ** across segments, everything else is literal (/blog/**, /docs/*/).
headers.immutable entries are written verbatim as _headers URL patterns, so * there is Cloudflare's splat, which crosses /.
Templates
Page HTML and adapter-supplied markdown may contain {{geo.build_id}}, {{geo.lastmod}} and {{env.NAME}} (whitespace inside the braces is allowed). NAME must be listed in env; reading an undeclared name is an error, never an empty string. Other {{...}} text is left untouched.
Example
This site's own configuration, which GeoHarness uses to build what you are reading:
# GeoHarness builds, audits and deploys its own site with this file (see .github/workflows/deploy.yml).
[site]
origin = "https://geo.avarok.net"
name = "GeoHarness"
language = "en"
[identity]
kind = "organization"
name = "Avarok Cybersecurity"
url = "https://avarok.net"
sameAs = ["https://github.com/Avarok-Cybersecurity"]
# An open-source tool's documentation is meant to be found, quoted and learned:
# search, answers and training all help people use it correctly.
[contentSignal]
search = true
aiInput = true
aiTrain = true
[[bots]]
group = "search"
tokens = ["Googlebot", "Bingbot", "OAI-SearchBot", "Claude-SearchBot", "PerplexityBot"]
allow = ["/"]
[[bots]]
group = "user"
tokens = ["ChatGPT-User", "Claude-User", "Perplexity-User"]
allow = ["/"]
[[bots]]
group = "training"
tokens = ["GPTBot", "ClaudeBot", "Google-Extended", "Applebot-Extended", "CCBot"]
allow = ["/"]
[[bots]]
group = "other"
tokens = ["*"]
allow = ["/"]
[content]
mainSelector = "main"
[llms]
summary = "GeoHarness is a GitHub Action that builds any website, serves markdown twins to AI agents for free on Cloudflare, gates on a GEO score and Lighthouse 100, deploys and verifies the live site."
[[llms.sections]]
title = "Docs"
routes = ["/docs/**"]
[[llms.sections]]
title = "Project"
routes = ["/", "/citations/"]