Showcase

An agent record, and what it may assert

A SemText document can carry its own agent: the agent element holds a name, a one-line bio and the instructions the model must follow, and the document body holds exactly what those instructions may draw on. The contract is auditable by a human — the instructions say “cite fact ids”, and the facts are right there with their ids.

The document

The agent element is inert to the browser and a record to the extractor — it never renders, which is why the visible page reads as the answerable material alone.

agent example

agent.md · authored SemText

source

      
      <sem-enhanced-document>
    
        <agent>
    
          <name>Auth Desk</name>
    
          <bio>Answers authentication questions from the facts on this page.</bio>
    
          <instructions>Answer only from sem-fact ids; cite the id with every answer. If no fact covers the question, say so.</instructions>
    
        </agent>
    
        <h1 data-kind="title">Auth desk</h1>
    
        <p>The records this agent answers from:</p>
    
        <sem-facts view-as="list">
    
          <sem-fact id="a-hdr" kind="concept">
    
            <statement>Where does the OIDC token travel?</statement>
    
            <conclusion>In the Authorization header, as a bearer token.</conclusion>
    
          </sem-fact>
    
          <sem-fact id="a-store" kind="anti-pattern">
    
            <statement>Where should a browser keep tokens?</statement>
    
            <conclusion>Memory or an httpOnly cookie — localStorage is readable by any script on the page.</conclusion>
    
          </sem-fact>
    
        </sem-facts>
    
        <sem-reveal summary="Which fact covers token storage?" collapsed>
    
          a-store — the anti-pattern record above; cite it for any question about where a browser keeps tokens.
    
        </sem-reveal>
    
      </sem-enhanced-document>

render

Render view: this block wires to the SemText runtime bundle in the full example viewer (port-reference lane) — the source beside it is the same file, byte for byte.

Why this works

statement and conclusion children make every fact an assertable pair — the conclusion is the citable takeaway.
The id attribute is a stable anchor: a #a-store deep link reaches the fact through any closed reveal.
sem-reveal is the non-assertive counterpart to the fact pair — a disclosure, not a claim extraction would return.

Elements used: agent, sem-facts, sem-reveal. Their contracts live on the reference pages .