Every claim keeps its evidence
Checks link conclusions to headers, markup, protocol documents, browser observations, or task citations.
How evidence flows →Website readiness / measured
Can an agent
actually use your site?
Agentlint scans the public evidence, delegates only bounded judgment, and leaves a fix prompt your coding agent can execute.
npx @timbenniks/agentlint init https://example.comSCAN / 7F2APUBLIC · READ ONLY
98
overall readinessevidence-backed
PASS- 01Discovery surface99
- 02Bounded missions94
- 03Docs prerequisitesWARN
- 04Safe tool plan98
30 pass2 warnings0 fail7 n/a
01 / METHOD
A scanner and a reasoner,
with a hard boundary between them.
The CLI owns collection and scoring. Your agent handles only the structured tasks that need meaning.
- ObserveHTTP, HTML, protocols, accessibility
- ReasonSchema-bound tasks using supplied evidence
- TestFour safe, non-executed missions
- FixA prioritized prompt for the site repository
02 / PROPERTIES
The useful constraints are the product.
Only applicable checks count
A portfolio is not punished for lacking an API. N/A is explicit and never quietly scored as failure.
Read the scoring model →The agent gets a closed world
Reasoning instructions, evidence, and output schema travel together. Invented sources and mutating plans are rejected.
Inspect reasoning tasks →03 / ROUTES
Choose your working mode.
One command to expose the gap between
machine-readable and agent-usable.