<?xml version="1.0" encoding="utf-8"?>
<?xml-stylesheet type="text/xsl" href="rfc2629.xslt" ?>
<!DOCTYPE rfc [
]>
<rfc
  xmlns:xi="http://www.w3.org/2001/XInclude"
  ipr="trust200902"
  docName="draft-han-ai-manifest-04"
  category="info"
  submissionType="independent"
  version="3">

  <front>
    <title abbrev="AI Manifest">
      AI Manifest: Site-Published Friction-Recovery Descriptors for AI Agents
    </title>
    <seriesInfo name="Internet-Draft" value="draft-han-ai-manifest-04"/>

    <author fullname="Won-pyo Han" initials="W." surname="Han">
      <organization>Individual</organization>
      <address>
        <postal>
          <country>KR</country>
        </postal>
        <email>pk102h@naver.com</email>
      </address>
    </author>

    <date year="2026" month="November" day="17"/>

    <area>General</area>
    <workgroup>Independent Submission</workgroup>

    <keyword>AI agents</keyword>
    <keyword>browser automation</keyword>
    <keyword>prompt injection</keyword>
    <keyword>agent security</keyword>
    <keyword>manifest</keyword>
    <keyword>AFRM</keyword>
    <keyword>friction recovery</keyword>
    <keyword>known traps</keyword>

    <abstract>
      <t>
        This document specifies the AI Manifest, a JSON format with
        which a website operator publishes an AI Friction-Recovery
        Manifest (AFRM): a description of the user-interface (UI)
        patterns on its own origin that predictably stall autonomous AI
        agents, together with constrained recovery steps and
        site-specific input hints. An agent that drives a browser on
        behalf of a user retrieves the manifest from a well-known URI
        and consults it when it encounters friction, instead of
        rediscovering the same site-specific behavior by trial and
        error.
      </t>
      <t>
        The manifest is untrusted input, interpreted by the agent's
        runtime and never shown to its language model. Its operative
        content is drawn from a closed vocabulary that cannot enter text,
        upload files, or reach another origin. Every action derived from
        a manifest passes through a policy enforcement point that is not
        a language model, and steps conventionally bound to a decision
        always require the user's confirmation; an agent's own policy
        can only deny them. Security controls
        such as access denials, rate limits, CAPTCHAs, authentication
        challenges, and confirmation dialogs are stop conditions, never
        traps to be recovered from. Optional publisher signatures and
        agent-configured reputation services provide authenticity but
        never imply safety and never relax these requirements.
      </t>
      <t>
        Preliminary measurements, including an adversarial evaluation of
        a reference implementation, are reported in an appendix.
      </t>
    </abstract>
  </front>

  <middle>

    <!-- ============================================================ -->
    <section anchor="introduction" numbered="true">
      <name>Introduction</name>

      <section anchor="problem">
        <name>Problem Statement</name>
        <t>
          AI agents increasingly operate websites on behalf of users by
          driving a browser through automation interfaces such as
          WebDriver, the Chrome DevTools Protocol, or tools exposed to a
          language model through the Model Context Protocol
          <xref target="MCP"/>. Such agents observe a page through its
          Document Object Model (DOM), its accessibility tree, or
          screenshots, and infer how to operate it.
        </t>
        <t>
          On many sites, and especially on enterprise transactional
          applications, that inference repeatedly fails at the same
          places: a cell edit that must be committed with a key gesture
          before the row unlocks, a list that renders only the visible
          window, a field whose valid values depend on configuration that
          is not visible on the page. Each agent rediscovers these
          patterns by trial and error, at a cost in actions, tokens, and
          failed tasks.
        </t>
        <t>
          The AI Manifest lets the operator of an origin describe these
          patterns once, in a form that an agent can consult and that an
          agent's safety machinery can check mechanically.
        </t>
      </section>

      <section anchor="applicability">
        <name>Applicability</name>
        <t>
          A manifest is appropriate only for friction that meets at
          least one of the following conditions:
        </t>
        <ol>
          <li>
            The UI cannot practically be changed by the party able to
            describe it, for example a deployed vendor application whose
            source is not accessible to the deploying organization.
          </li>
          <li>
            The friction remains valid for human users and exists for a
            reason, for example a commit gesture that protects data
            integrity, so that removing it would be wrong even though it
            stalls agents.
          </li>
          <li>
            Operating the UI correctly requires site-specific knowledge
            that is not derivable from public or training data, for
            example which code values are registered for a given
            organizational unit.
          </li>
        </ol>
        <t>
          Where a site can expose a structured interface to agents --
          an HTTP API, an MCP server <xref target="MCP"/>, page-registered
          tools <xref target="WEBMCP"/>, or an agent card
          <xref target="A2A"/> -- that interface is preferable and agents
          <bcp14>SHOULD</bcp14> use it instead of UI-level recovery. The
          AI Manifest addresses the residual UI that such interfaces do
          not cover. Friction that consists of a security control is
          never in scope (<xref target="controls"/>).
        </t>
        <t>
          In the first case the party that publishes the manifest is
          typically the deploying organization rather than the
          application vendor. Such an organization usually cannot change
          the application, but does control the web server or reverse
          proxy in front of it, and can serve a static file under
          /.well-known/ on that origin.
        </t>
      </section>

      <section anchor="related">
        <name>Relationship to Other Work</name>
        <ul>
          <li>
            robots.txt <xref target="RFC9309"/> and the vocabulary work of
            the IETF AIPREF working group <xref target="AIPREF"/> express
            whether and how automated clients may use content. The AI
            Manifest does not grant or express permission; an agent that
            consults it remains subject to those mechanisms.
          </li>
          <li>
            llms.txt <xref target="LLMSTXT"/> provides free-text
            documentation for language models. The AI Manifest is
            deliberately not free text: its operative content is a closed
            vocabulary that can be validated without a language model.
          </li>
          <li>
            Page-registered tools <xref target="WEBMCP"/> let a site
            execute functions on behalf of an agent. The AI Manifest
            never executes anything; it describes how the existing UI
            behaves.
          </li>
          <li>
            Agent authentication work such as the IETF Web Bot Auth
            working group <xref target="WEBBOTAUTH"/> lets a site
            identify agents. It is complementary
            (<xref target="sec-recon"/>).
          </li>
        </ul>
      </section>

      <section anchor="principles">
        <name>Design Principles</name>
        <t>
          The security model of this revision rests on six principles.
          Every normative requirement below derives from one of them.
        </t>
        <dl newline="false">
          <dt>P1. Untrusted by construction.</dt>
          <dd>
            A manifest is authored by the site, whose interests may
            differ from the user's. Its delivery channel, signature, or
            reputation status never makes its content trusted.
          </dd>
          <dt>P2. Data, not instructions.</dt>
          <dd>
            Operative content is expressed in a closed vocabulary of
            actions and categories, interpreted by the agent's runtime.
            No manifest string reaches the language model.
          </dd>
          <dt>P3. Origin-bound.</dt>
          <dd>
            A manifest describes only the origin that serves it, and
            actions derived from it can act only within that origin.
          </dd>
          <dt>P4. Least authority.</dt>
          <dd>
            A manifest can inform or narrow what an agent does; it can
            never make an agent do something it would not otherwise be
            permitted to do, and never removes a user confirmation.
          </dd>
          <dt>P5. Security controls are not friction.</dt>
          <dd>
            Mechanisms that deny, challenge, throttle, or ask for human
            confirmation are stop conditions, not traps to be escaped.
          </dd>
          <dt>P6. Mediated outside the model.</dt>
          <dd>
            Every manifest-derived action passes, before it has any
            effect, through a policy enforcement point whose decision a
            language model cannot influence, and is recorded.
          </dd>
        </dl>
        <t>
          P5 and P6 draw on the published design and defensive analysis
          of an open-source autonomous penetration-testing system
          <xref target="ARTEX"/>. Its defense guide identifies
          persistence with bypass variations after 401, 403, or 429
          responses as a behavioral indicator of autonomous attack, and
          its tool layer classifies such responses, including WAF blocks,
          rate limits, and CAPTCHAs, as "blocked" so that its planner
          changes strategy rather than stopping. A friction-recovery
          mechanism that did not exclude security controls would teach
          benign agents the same behavior. The same system passes every
          tool call through a hook that evaluates priority-ordered rules
          with allow, deny, and ask outcomes, optionally falls back to a
          language-model judge, and records an audit log.
          <xref target="pep"/> adopts that structure for manifest-derived
          actions, but excludes a language model from the decision.
        </t>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="terminology" numbered="true">
      <name>Conventions and Terminology</name>
      <t>
        The key words "<bcp14>MUST</bcp14>", "<bcp14>MUST NOT</bcp14>",
        "<bcp14>REQUIRED</bcp14>", "<bcp14>SHALL</bcp14>",
        "<bcp14>SHALL NOT</bcp14>", "<bcp14>SHOULD</bcp14>",
        "<bcp14>SHOULD NOT</bcp14>", "<bcp14>RECOMMENDED</bcp14>",
        "<bcp14>NOT RECOMMENDED</bcp14>", "<bcp14>MAY</bcp14>", and
        "<bcp14>OPTIONAL</bcp14>" in this document are to be interpreted
        as described in BCP 14 <xref target="RFC2119"/>
        <xref target="RFC8174"/> when, and only when, they appear in all
        capitals, as shown here.
      </t>
      <dl newline="true">
        <dt>AI Manifest; AI Friction-Recovery Manifest (AFRM)</dt>
        <dd>A JSON object <xref target="RFC8259"/> conforming to
          <xref target="format"/>. The two terms are used
          interchangeably.</dd>
        <dt>Publisher</dt>
        <dd>The origin <xref target="RFC6454"/> that serves a manifest,
          and the operator responsible for it.</dd>
        <dt>Agent</dt>
        <dd>Software, typically driven by a large language model (LLM),
          that operates web pages through a browser on behalf of a
          principal.</dd>
        <dt>Principal</dt>
        <dd>The human user on whose behalf the agent acts.</dd>
        <dt>Task</dt>
        <dd>One unit of work that the principal has asked the agent to
          perform, from the principal's request to its completion or
          abandonment.</dd>
        <dt>Manifest-Derived Action</dt>
        <dd>Any browser action that the agent takes (a) as a step of a
          recovery or commit in a manifest, (b) entering a value that the
          principal chose from a manifest's allowedValues, or (c)
          following a shortcut, together with the submission of any form
          that the manifest modified. A value that the agent takes from
          the principal's task and enters into a field it found itself is
          not manifest-derived, even if a field hint describes that
          field.</dd>
        <dt>Form Modified by the Manifest</dt>
        <dd>The form owner <xref target="HTML"/> of an element that was
          the target of a manifest-derived click or of a manifest-derived
          key step other than Tab, of an element whose value changed as a
          result of such a step, or of a field into which a value chosen
          from allowedValues was entered. For an element without a form
          owner, as in applications that save through script, the "form"
          is the document containing the element. Its submission or
          saving by any means, including by the step that modified it,
          is a sensitive action (<xref target="sensitive"/>).</dd>
        <dt>Policy Enforcement Point (PEP)</dt>
        <dd>The component of an agent that decides whether a
          manifest-derived action is allowed, denied, or requires the
          principal's confirmation (<xref target="pep"/>).</dd>
        <dt>Sensitive Action</dt>
        <dd>An action enumerated in <xref target="sensitive"/>.</dd>
        <dt>Security Control</dt>
        <dd>A mechanism enumerated in <xref target="controls"/>.</dd>
        <dt>Locator</dt>
        <dd>A description of a UI element within the publisher's
          documents (<xref target="locator"/>).</dd>
      </dl>
    </section>

    <!-- ============================================================ -->
    <section anchor="discovery" numbered="true">
      <name>Discovery</name>

      <section anchor="wellknown">
        <name>Well-Known URI</name>
        <t>
          A publisher that provides a manifest <bcp14>MUST</bcp14> serve
          it at the following well-known URI <xref target="RFC8615"/> on
          the origin it describes:
        </t>
        <artwork><![CDATA[
  /.well-known/ai-friction-manifest.json
]]></artwork>
        <t>
          The response <bcp14>MUST</bcp14> be served over HTTPS
          <xref target="RFC9110"/> with media type application/json.
          Agents <bcp14>MUST</bcp14> retrieve it without credentials and
          <bcp14>MUST NOT</bcp14> follow any redirect; a redirect
          response is treated as absence of a manifest, so that an open
          redirect on the origin cannot make another resource be read as
          its manifest. <xref target="sec-recon"/> describes an optional
          credentialed variant.
        </t>
        <t>
          Earlier revisions used the suffix "ai-manifest.json" and the
          link relation "ai-manifest". Those names described the purpose
          of the resource more broadly than its content, and are replaced
          by names specific to friction recovery. Agents
          <bcp14>MUST</bcp14> ignore any resource at
          /.well-known/ai-manifest.json. The earlier names were never
          registered with IANA.
        </t>
      </section>

      <section anchor="link">
        <name>Link Relation</name>
        <t>
          A publisher <bcp14>MAY</bcp14> additionally advertise the
          manifest with the "ai-friction-manifest" link relation
          <xref target="RFC8288"/>, either in an HTTP response header or
          in an HTML link element <xref target="HTML"/>:
        </t>
        <artwork><![CDATA[
  Link: </.well-known/ai-friction-manifest.json>;
        rel="ai-friction-manifest"

  <link rel="ai-friction-manifest"
        href="/.well-known/ai-friction-manifest.json">
]]></artwork>
        <t>
          The advertisement is a hint that a manifest exists, which
          addresses the observation that agents do not spontaneously
          look for one (<xref target="app-measure"/>). Agents
          <bcp14>MUST</bcp14> ignore a link target that does not resolve
          to the well-known URI on the same origin as the document that
          carries the link.
        </t>
      </section>

      <section anchor="removed-methods">
        <name>Removed Discovery Methods</name>
        <t>
          Earlier revisions defined an inline DOM element carrying the
          manifest in a data attribute, an "X-AI-Manifest" response
          header, and an HTML meta element pointing at an arbitrary
          path. They are removed:
        </t>
        <ul>
          <li>
            Inline manifests are indistinguishable from content injected
            by third-party scripts, advertisements, user-generated
            content, or cross-site scripting. Agents <bcp14>MUST
            NOT</bcp14> treat any element in a page as a manifest.
          </li>
          <li>
            The "X-" header prefix is deprecated <xref target="RFC6648"/>,
            and the Link header above carries the same information. A
            hash advertised by the same server that serves the manifest
            adds no integrity beyond HTTPS.
          </li>
          <li>
            A meta element with an arbitrary path weakens origin binding
            without adding function.
          </li>
        </ul>
      </section>

      <section anchor="local">
        <name>Agent-Local Manifests</name>
        <t>
          Earlier revisions defined a file-system path in a specific
          vendor's configuration directory from which an agent loads
          user-curated manifests, and gave such manifests precedence over
          and more trust than site-published ones. That mechanism is
          removed from the protocol: local configuration is an
          implementation matter, and a file in a user's configuration
          directory is frequently written by tools, plug-ins, or shared
          packages rather than by the user.
        </t>
        <t>
          An agent <bcp14>MAY</bcp14> accept local overrides, subject to
          all of the following:
        </t>
        <ul>
          <li>
            An override applies only to the site-published manifest that
            the agent has currently retrieved and validated for the same
            origin. If no such manifest exists, the override
            <bcp14>MUST</bcp14> be ignored.
          </li>
          <li>
            An override can only remove descriptors, remove values from
            allowedValues, shorten scope, or bring expires earlier. Any
            other content <bcp14>MUST</bcp14> be ignored.
          </li>
          <li>
            The resulting manifest is subject to every requirement of
            this document, without exemption.
          </li>
        </ul>
      </section>

      <section anchor="retrieval">
        <name>Retrieval and Caching</name>
        <ul>
          <li>
            An agent <bcp14>SHOULD</bcp14> retrieve the manifest at most
            once per origin per task, and <bcp14>MAY</bcp14> use it for
            the rest of that task until its "expires" member. Across
            tasks, an agent <bcp14>MUST</bcp14> stop reusing a stored
            manifest at the earliest of its HTTP freshness lifetime
            <xref target="RFC9111"/>, its "expires" member, and 90 days
            after retrieval.
          </li>
          <li>
            An agent <bcp14>MUST</bcp14> reject a manifest larger than
            262144 octets, a manifest that is not an I-JSON message
            <xref target="RFC7493"/>, a manifest whose version member is
            not "2.0", and a manifest whose objects and arrays are nested
            more deeply than the agent's implementation limit
            (<xref target="RFC8259" section="9"/>). Counting the
            top-level object as one level, that limit
            <bcp14>MUST</bcp14> be at least 64 levels, and publishers
            <bcp14>MUST NOT</bcp14> nest a manifest more deeply than 64
            levels. An agent <bcp14>MUST</bcp14> ensure that no manifest
            can end its process; enforcing the nesting limit before or
            while parsing is <bcp14>RECOMMENDED</bcp14>,
            because in many environments a parser that exhausts its stack
            ends the process rather than reporting an error.
          </li>
          <li>
            Failure to retrieve or validate a manifest <bcp14>MUST</bcp14>
            result in the agent's baseline behavior, as if no manifest
            existed. Exceeding the size or nesting-depth limit, and any
            error raised while parsing a manifest or while performing the
            checks in this list, including exhaustion of the parser's stack or
            memory, is such a failure: the task continues without the
            manifest. Invalid descriptors are handled as in
            <xref target="toplevel"/>.
          </li>
        </ul>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="format" numbered="true">
      <name>Manifest Format</name>

      <section anchor="toplevel">
        <name>Top-Level Members</name>
        <dl newline="true">
          <dt>version (string, REQUIRED)</dt>
          <dd>"2.0" for manifests conforming to this document. Manifests
            with any other value, including the "1.0" of earlier
            revisions, are ignored (<xref target="retrieval"/>).</dd>
          <dt>origin (string, REQUIRED)</dt>
          <dd>The ASCII serialization of the publisher's origin
            <xref target="RFC6454"/>, for example
            "https://erp.example.com". Agents <bcp14>MUST</bcp14>
            ignore a manifest whose origin member does not equal the
            origin of the URL from which it was retrieved.</dd>
          <dt>expires (string, REQUIRED)</dt>
          <dd>A date-time <xref target="RFC3339"/> after which the
            manifest <bcp14>MUST NOT</bcp14> be used. Publishers
            <bcp14>SHOULD</bcp14> keep this within 30 days of
            publication, so that stale descriptors expire after UI
            changes.</dd>
          <dt>scope (array of strings, OPTIONAL)</dt>
          <dd>Path prefixes, each beginning and ending with "/". A
            document's path is in scope if it starts with one of them.
            If absent, the manifest applies to all paths of the
            origin.</dd>
          <dt>frameworkHints (object, OPTIONAL)</dt>
          <dd>A "name" member (string) naming the UI framework family,
            and a "renderingModel" member (string) from
            <xref target="vocab-render"/>. Publishers
            <bcp14>SHOULD NOT</bcp14> include version numbers
            (<xref target="sec-recon"/>).</dd>
          <dt>knownTraps (array, REQUIRED)</dt>
          <dd>Trap descriptors (<xref target="trap"/>). May be empty.</dd>
          <dt>fieldHints (array, OPTIONAL)</dt>
          <dd>Field descriptors (<xref target="fieldhint"/>).</dd>
          <dt>shortcuts (array, OPTIONAL)</dt>
          <dd>Shortcut descriptors (<xref target="shortcut"/>).</dd>
          <dt>signatures (array, OPTIONAL)</dt>
          <dd>Publisher signatures (<xref target="signatures"/>).</dd>
        </dl>
        <t>
          Unknown top-level members are ignored. Within a descriptor, a
          locator, or a step, an unknown member or value makes the
          containing descriptor invalid, and the agent
          <bcp14>MUST</bcp14> ignore that descriptor while it
          <bcp14>MAY</bcp14> use the rest of the manifest. Ignoring a
          member of a locator would make it match more elements, so
          locators always fail closed.
        </t>
        <t>
          The members "publisher", "manifestId", and "registry_url" of
          earlier revisions are removed. In particular, a manifest can no
          longer name the service that vouches for it
          (<xref target="reputation"/>).
        </t>
      </section>

      <section anchor="locator">
        <name>Locator Object</name>
        <t>
          A locator identifies a UI element. It contains one or more of
          the following members; an element matches only if it satisfies
          all members present.
        </t>
        <dl newline="false">
          <dt>role (string):</dt>
          <dd>A non-abstract role defined in WAI-ARIA 1.2
            <xref target="WAI-ARIA"/>. Other role values make the
            descriptor invalid.</dd>
          <dt>name (string):</dt>
          <dd>The element's accessible name <xref target="ACCNAME"/>,
            compared exactly after collapsing whitespace.</dd>
          <dt>column (string):</dt>
          <dd>For an element in a grid or table, the accessible name of
            its column header.</dd>
          <dt>focused (boolean true):</dt>
          <dd>The element that currently has focus.</dd>
          <dt>within (locator):</dt>
          <dd>An ancestor that the element must have, itself resolved
            under the same rules.</dd>
          <dt>testId (string):</dt>
          <dd>The value of the element's "data-testid" attribute.</dd>
          <dt>css (string):</dt>
          <dd>A selector <xref target="SELECTORS4"/>, as a fallback.</dd>
        </dl>
        <t>
          Publishers <bcp14>SHOULD</bcp14> prefer role, name, and column,
          which survive build-time class-name mangling and are what
          accessibility-tree-based agents observe, and
          <bcp14>SHOULD</bcp14> use css only when no accessible name
          exists. On pages that contain user-generated or third-party
          content, publishers <bcp14>SHOULD</bcp14> anchor every step
          locator with "within" to an application container identified
          by testId, so that injected elements outside that container
          cannot match (<xref target="sec-bait"/>).
        </t>
        <t>
          An agent <bcp14>MUST</bcp14> resolve a locator against the
          live document at the time of use, only within documents of the
          manifest's origin whose path is in scope, and never inside a
          cross-origin frame. If the locator matches no element, more
          than one element, or an element that is not being rendered
          <xref target="HTML"/>, the agent <bcp14>MUST NOT</bcp14>
          perform the associated action.
        </t>
        <t>
          The locator of a field hint is different: it identifies a
          class of elements and <bcp14>MAY</bcp14> match more than one.
          It is used only to recognize that an element the agent is
          already editing belongs to the field, never to choose a target,
          and the exactly-one rule does not apply to it.
        </t>
      </section>

      <section anchor="trap">
        <name>Trap Descriptor</name>
        <dl newline="true">
          <dt>trapId (string, REQUIRED)</dt>
          <dd>Unique within the manifest.</dd>
          <dt>category (string, REQUIRED)</dt>
          <dd>A value from <xref target="vocab-trap"/>.</dd>
          <dt>locator (object, conditional)</dt>
          <dd>The element at which the trap manifests. REQUIRED except
            for the category native-alert, which has no element.</dd>
          <dt>condition (object, conditional)</dt>
          <dd>Exactly one of "visible" (a locator that must match),
            "absent" (a locator that must not match), or "dialog" (the
            value "alert"). REQUIRED if recovery contains a wait step or
            the category is native-alert.</dd>
          <dt>recovery (array, REQUIRED)</dt>
          <dd>An ordered list of one to eight step objects
            (<xref target="step"/>). Steps are identified by trapId and
            zero-based position.</dd>
          <dt>description (string, OPTIONAL)</dt>
          <dd>Display text, at most 280 characters
            (<xref target="freetext"/>).</dd>
        </dl>
      </section>

      <section anchor="step">
        <name>Step Object</name>
        <dl newline="true">
          <dt>action (string, REQUIRED)</dt>
          <dd>A value from <xref target="vocab-action"/>.</dd>
          <dt>locator (object, conditional)</dt>
          <dd>REQUIRED for click, focus, key, scroll, and assert.</dd>
          <dt>key (string, conditional)</dt>
          <dd>REQUIRED for key. One of "Enter", "Tab", "Escape",
            "ArrowUp", "ArrowDown", "PageUp", "PageDown".</dd>
          <dt>target (string, conditional)</dt>
          <dd>REQUIRED for navigate. A path-absolute URI reference
            <xref target="RFC3986"/> with neither a query nor a fragment
            component.</dd>
          <dt>timeoutMs (integer, conditional)</dt>
          <dd>REQUIRED for wait; from 1 to 30000.</dd>
        </dl>
        <t>
          An agent <bcp14>MUST</bcp14> resolve a navigate target against
          the manifest's origin with the URL parser of
          <xref target="URL"/>, and <bcp14>MUST</bcp14> treat the step as
          invalid if the resulting URL's origin differs from the
          manifest's origin or its path is not in scope. The same rules
          apply to every redirect: before following one, the agent
          <bcp14>MUST</bcp14> check its target, and <bcp14>MUST</bcp14>
          abort the navigation if the target has a different origin, a
          query or fragment component, or a path outside scope. Because
          browsers follow redirects automatically, this requires
          intercepting navigation requests, for example through the
          request interception facility of the automation interface.
        </t>
        <t>
          There is no step that enters text, selects an option, or
          supplies any value. Steps cannot carry data.
        </t>
      </section>

      <section anchor="fieldhint">
        <name>Field Hint Descriptor</name>
        <t>
          Field hints convey site-specific knowledge about valid input,
          the third case of <xref target="applicability"/>. Their design
          is motivated by the measurement in
          <xref target="app-measure"/>, in which the information the
          agent lacked was of this kind.
        </t>
        <dl newline="true">
          <dt>fieldId (string, REQUIRED)</dt>
          <dd>Unique within the manifest. Several field hints
            <bcp14>MAY</bcp14> have the same locator with different
            "when" members.</dd>
          <dt>locator (object, REQUIRED)</dt>
          <dd>The input element.</dd>
          <dt>allowedValues (array of strings, OPTIONAL)</dt>
          <dd>Values the site will accept in this field, at most 64
            entries, each of at most 32 characters drawn from ASCII
            letters, digits, ".", "_", and "-".</dd>
          <dt>when (object, OPTIONAL)</dt>
          <dd>A map from the fieldId of another field hint to a value.
            The field hint applies only if, for every entry, the current
            value of that other field, as exposed in its accessibility
            tree, equals the given value. If the other field's locator
            matches more than one element, the agent uses the one that
            shares the nearest ancestor with role row with the element
            being edited; if there is no such element or more than one,
            the entry is unsatisfied.</dd>
          <dt>commit (object, OPTIONAL)</dt>
          <dd>A single step object with action "key", identified by
            fieldId, required to commit an entry in this field.</dd>
          <dt>description (string, OPTIONAL)</dt>
          <dd>Display text, at most 280 characters.</dd>
        </dl>
        <t>
          Even short tokens can spell instructions, so allowedValues are
          never shown to a language model (<xref target="freetext"/>).
          They are used as follows:
        </t>
        <ol>
          <li>When the model proposes a value for a field that an
            applicable field hint describes, the PEP checks the value
            against allowedValues and returns to the model only whether
            it matched.</li>
          <li>If it did not match, the agent <bcp14>MUST NOT</bcp14>
            substitute a listed value. It <bcp14>MAY</bcp14> show the
            listed values to the principal, who may choose one or
            abandon the entry; otherwise it reports the mismatch.</li>
          <li>A value that the principal chose from the list is entered
            by the agent's runtime, not by the model, so that the model
            does not see it. The choice is the principal's confirmation,
            and the form that contains the field counts as modified
            (<xref target="sensitive"/>).</li>
        </ol>
        <t>
          A field hint therefore constrains which value is acceptable and
          can inform the principal; it never chooses a value, and never
          informs the model's reasoning beyond a match result.
        </t>
      </section>

      <section anchor="shortcut">
        <name>Shortcut Descriptor</name>
        <dl newline="true">
          <dt>shortcutId (string, REQUIRED)</dt>
          <dd>Unique within the manifest.</dd>
          <dt>target (string, REQUIRED)</dt>
          <dd>As for the navigate action (<xref target="step"/>).</dd>
          <dt>label (string, REQUIRED)</dt>
          <dd>Display text naming the destination, at most 80
            characters.</dd>
        </dl>
        <t>
          A shortcut replaces a sequence of menu clicks with a GET
          request. Shortcuts are offered to the principal, not to the
          language model: an agent <bcp14>MAY</bcp14> list shortcut labels
          to the principal, who may choose one. Because a server can
          perform state changes on GET, following a shortcut is still
          subject to the PEP like any other navigation
          (<xref target="pep"/>). A site that wants agents to
          perform operations directly <bcp14>SHOULD</bcp14> expose a
          structured interface (<xref target="applicability"/>).
        </t>
      </section>

      <section anchor="freetext">
        <name>Manifest Strings and the Language Model</name>
        <t>
          No string taken from a manifest reaches a language model. This
          covers description, label, allowedValues, trapId, fieldId,
          shortcutId, locator members, and navigation targets. An agent
          <bcp14>MUST NOT</bcp14> place any of them in a model's context,
          quoted or otherwise.
        </t>
        <t>
          What an agent <bcp14>MAY</bcp14> give a model about a manifest
          is limited to values from the closed vocabularies of
          <xref target="vocab"/> (trap categories, action names, and
          rendering models), small integers that the agent itself
          assigns to descriptors, and the match results of
          <xref target="fieldhint"/>. A model can thus learn, for example,
          "trap 2, category commit-gesture, is active; recovery is
          available", and can ask the agent to run recovery 2, but cannot
          read anything the publisher wrote. Description and label are
          shown only to the principal and recorded in logs. No PEP
          decision depends on their content.
        </t>
        <t>
          The same limit applies to everything the agent returns to the
          model about manifest-derived actions, including PEP outcomes,
          reasons, error messages, and audit records: these
          <bcp14>MUST</bcp14> contain only the items permitted above. An
          agent that feeds tool output or logs back to its model
          <bcp14>MUST</bcp14> filter them accordingly.
        </t>
      </section>

      <section anchor="vocab">
        <name>Vocabularies</name>
        <t>
          The vocabularies below are closed for version "2.0". A future
          version of this document may extend them; values unknown to an
          agent are handled as specified in <xref target="toplevel"/>.
        </t>

        <section anchor="vocab-action">
          <name>Actions</name>
          <dl newline="false">
            <dt>click:</dt><dd>Activate the located element.</dd>
            <dt>focus:</dt><dd>Move focus to the located element.</dd>
            <dt>key:</dt><dd>Press one key while the located element has
              focus.</dd>
            <dt>scroll:</dt><dd>Scroll the located container by at most
              one viewport height.</dd>
            <dt>wait:</dt><dd>Wait up to timeoutMs for the trap's
              condition to stop holding.</dd>
            <dt>assert:</dt><dd>Verify that the located element is
              present; otherwise stop the recovery.</dd>
            <dt>navigate:</dt><dd>Load a same-origin target with GET.</dd>
            <dt>dismissAlert:</dt><dd>Dismiss a window.alert() dialog
              <xref target="HTML"/>, which offers no choice.</dd>
          </dl>
          <t>
            The actions "fill", "select", and "upload" of earlier
            revisions are removed. None of the trap categories requires a
            manifest to supply a value, and without them a manifest has no
            means to place data into a page or send it anywhere. There is
            deliberately no action that accepts a confirmation dialog or
            answers a prompt dialog.
          </t>
        </section>

        <section anchor="vocab-trap">
          <name>Trap Categories</name>
          <dl newline="false">
            <dt>commit-gesture:</dt><dd>An entry must be confirmed with a
              key gesture before the UI proceeds or unlocks.</dd>
            <dt>shadow-dom:</dt><dd>The target is rendered inside a
              shadow tree.</dd>
            <dt>virtual-scroll:</dt><dd>Only visible items are
              rendered.</dd>
            <dt>same-origin-frame:</dt><dd>The target is inside a frame
              of the same origin.</dd>
            <dt>delayed-render:</dt><dd>The target appears only after an
              asynchronous operation.</dd>
            <dt>hidden-required:</dt><dd>A required input is not visible
              until another control is operated.</dd>
            <dt>native-alert:</dt><dd>A window.alert() dialog blocks the
              page.</dd>
          </dl>
          <t>
            Compared with earlier revisions, the cross-origin case of the
            iframe category is removed (<xref target="locator"/>), and the
            native dialog category is narrowed to alert(), because
            confirm() and prompt() exist to obtain a human decision
            (<xref target="controls"/>).
          </t>
        </section>

        <section anchor="vocab-render">
          <name>Rendering Models</name>
          <t>
            "server-rendered", "client-rendered", "hybrid", "canvas".
          </t>
        </section>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="processing" numbered="true">
      <name>Agent Processing Requirements</name>

      <section anchor="consult">
        <name>Consulting the Manifest</name>
        <t>
          A manifest is reference data, consulted by the agent's runtime
          rather than read by its model. The runtime evaluates trap
          conditions against the live document; when one holds, it
          reports the trap to the model as described in
          <xref target="freetext"/>, and runs the recovery through the PEP
          when the model asks for it. Field hints are applied as in
          <xref target="fieldhint"/>, and shortcuts are offered to the
          principal (<xref target="shortcut"/>). The agent remains free to
          use its own inference at any time, and a manifest never
          overrides the principal's instructions or the agent's own
          policy.
        </t>
        <t>
          An agent <bcp14>MUST</bcp14> perform manifest-derived actions
          as trusted input through the browser's own input pipeline, for
          example WebDriver actions, as a user would. It
          <bcp14>MUST NOT</bcp14> dispatch events from script, modify
          page script state, or remove or disable page elements in order
          to recover from a trap.
        </t>
        <t>
          Trusted input reaches whatever element the browser finds at the
          position of the input, which need not be the element that the
          locator resolved: an element that is transparent, or otherwise
          hard to see, and placed over the resolved element receives a
          click aimed at it, while an ask names the element underneath
          (<xref target="pep"/>). Immediately before it performs a
          manifest-derived click, and after any ask, an agent
          <bcp14>MUST</bcp14> verify that the element that would receive
          the click at the position it uses, taken as the innermost
          element at that position without retargeting across shadow
          boundaries <xref target="DOM"/>, is one of the following, and
          <bcp14>MUST NOT</bcp14> perform the click otherwise:
        </t>
        <ul>
          <li>the resolved element;</li>
          <li>a shadow-including descendant <xref target="DOM"/> of the
            resolved element, provided that no element on the path from
            it up to the resolved element, other than the resolved
            element itself, is activatable (<xref target="pep"/>);
            or</li>
          <li>a label element whose labeled control
            <xref target="HTML"/> is the resolved element.</li>
        </ul>
        <t>
          The element that would receive the click is determined in the
          document that contains the resolved element; for a document in
          a frame, the element that would receive the click in each
          ancestor document <bcp14>MUST</bcp14> be the element of the
          frame that leads to it. An agent <bcp14>SHOULD</bcp14> also
          decline a manifest-derived click or key step on an element that
          the principal cannot see at the position of the input, or for a
          key step at the element's in-view position, for example because
          the element is fully transparent or clipped away at that
          position, or because that position is outside the viewport
          after the element has been scrolled into view. This document does not specify how
          these checks are made. The Element Click command of WebDriver
          <xref target="WEBDRIVER"/> returns an error when the element
          is obscured by another element, but its Perform Actions
          command, which dispatches input at a position, makes no such
          check. If a check fails, the agent <bcp14>MAY</bcp14> repeat
          it for up to 5000 ms, without acting on the page in between
          other than scrolling the element into view; this is not a
          variant of the action (<xref target="pep"/>). An action that
          these checks still decline is treated as denied by the PEP and
          recorded as such (<xref target="audit"/>).
        </t>
      </section>

      <section anchor="pep">
        <name>Policy Enforcement Point</name>
        <t>
          An agent that consumes manifests <bcp14>MUST</bcp14> track
          which of its actions are manifest-derived, and
          <bcp14>MUST</bcp14> route every manifest-derived action through
          a PEP before execution. A language model <bcp14>MUST
          NOT</bcp14> take part in the PEP's decision, so that manifest
          content cannot influence it. For each action the PEP:
        </t>
        <ol>
          <li>validates the descriptor against <xref target="format"/>;</li>
          <li>re-resolves any locator against the live document
            (<xref target="locator"/>);</li>
          <li>denies the action if it is directed at a security control
            that the agent has detected (<xref target="controls"/>);</li>
          <li>assigns the default outcome below;</li>
          <li>applies the agent's own policy, which can only make the
            outcome stricter; and</li>
          <li>records the outcome (<xref target="audit"/>).</li>
        </ol>
        <t>
          In this document, an activatable element is a button, an "a" or "area" element with an href attribute, a
          summary
          element, an input element whose type is submit, image, reset,
          button, checkbox, or radio, or an element whose role is button,
          link, menuitem, menuitemcheckbox, menuitemradio, option, tab,
          switch, checkbox, or radio. A value-bearing control is a select
          element, an input or textarea element, an editing host
          (contenteditable), or an element whose role
          is combobox, listbox, radiogroup, slider, spinbutton, textbox,
          or tablist, or that is a descendant of one. The default
          outcomes are:
        </t>
        <ul>
          <li>allow: focus, scroll, assert, and dismissAlert;</li>
          <li>allow: wait with timeoutMs of at most 5000; otherwise
            ask;</li>
          <li>allow: key Tab;</li>
          <li>allow: key ArrowUp, ArrowDown, PageUp, or PageDown on an
            element that is neither activatable nor a value-bearing
            control; otherwise ask, and the control counts as modified
            for <xref target="sensitive"/>;</li>
          <li>ask: key Enter and key Escape, on any element;</li>
          <li>allow: entering a value that the principal has just chosen
            from allowedValues (<xref target="fieldhint"/>);</li>
          <li>ask: click, navigate, and following a shortcut;</li>
          <li>ask: any sensitive action (<xref target="sensitive"/>);</li>
          <li>ask or deny: anything the PEP cannot classify.</li>
        </ul>
        <t>
          Enter and Escape always ask because page script can bind any
          effect to a key event on any element, whatever its tag or role,
          and these two keys are the ones conventionally bound to commit
          and dismiss. The same is true in principle of every input
          event; the actions allowed by default are those whose
          conventional meaning is navigation within the page rather than
          a decision, and their residual risk is discussed in
          <xref target="sec-bait"/>.
        </t>
        <t>
          Key Enter on an element whose form owner could be submitted by
          implicit submission <xref target="HTML"/> would submit a form
          that the step itself has modified
          (<xref target="terminology"/>); <xref target="sensitive"/>
          therefore requires the PEP to determine it to be a sensitive
          action.
        </t>
        <t>
          An ask <bcp14>MUST</bcp14> show the principal at least the
          following. The agent's runtime derives them from the live page,
          from the structural values of the validated descriptor, and
          from the PEP's own classification, never from description or
          label text or from a language model. Values that the publisher
          controls, such as accessible names and destination paths, are
          shown as quoted data, distinct from the agent's own wording.
        </t>
        <dl newline="false">
          <dt>Target:</dt>
          <dd>the action, with its key or its wait duration where it has
            one; for actions on an element, the role and accessible name
            of the element actually resolved on the live page; and for
            navigate and shortcuts, the origin and path of the
            destination.</dd>
          <dt>Source:</dt>
          <dd>that the action comes from the manifest of the page's
            origin, as a recovery step, a field-hint commit, a shortcut,
            or the submission of a form modified by the manifest,
            together with the page's origin and path.</dd>
          <dt>Effect:</dt>
          <dd>first, every sensitive action (<xref target="sensitive"/>)
            that the PEP has determined the action can perform or
            trigger; then the conventional effect of the action, such as
            activating the element, loading the destination with GET,
            committing or dismissing with Enter or Escape, changing the
            value of a control, or letting time pass; and, for every
            action other than wait, a statement that the page decides the
            actual effect, which the agent cannot determine and which can
            include a purchase, a deletion, or another sensitive
            action.</dd>
        </dl>
        <t>
          That statement discloses that the effect is unknown; it does
          not identify the effect, which the PEP cannot determine
          (<xref target="sensitive"/>). An agent <bcp14>SHOULD</bcp14>
          warn the principal when the text rendered for an activatable
          element is not contained in its accessible name. Any
          description or label shown with an ask <bcp14>MUST</bcp14> be
          visibly marked as written by the publisher, so that publisher
          text cannot pass for the agent's own explanation. The same
          marking applies when shortcut labels or allowedValues are shown
          to the principal. An ask is shown only to the principal; what
          the model receives about it remains limited as specified in
          <xref target="freetext"/>.
        </t>
        <t>
          When the principal declines, or the PEP denies, the agent
          <bcp14>MUST</bcp14> abandon the descriptor rather than attempt
          a variant of it. Repeating the checks of
          <xref target="consult"/> within their time limit, as described
          there, is not a variant.
        </t>
      </section>

      <section anchor="sensitive">
        <name>Sensitive Actions</name>
        <t>
          The following are sensitive actions. An agent
          <bcp14>MUST</bcp14> obtain the principal's confirmation before
          a manifest-derived action that the PEP determines can perform
          or trigger one, regardless of any manifest content, signature,
          or reputation status:
        </t>
        <ul>
          <li>submitting, by any means, a form modified by the manifest
            (<xref target="terminology"/>), or a form that contains
            personal data or credentials;</li>
          <li>a purchase, payment, transfer, or other financial
            commitment;</li>
          <li>deleting data, or any operation the site presents as
            irreversible;</li>
          <li>accepting terms, consents, cookie or tracking choices, or
            subscriptions;</li>
          <li>sending a message or publishing content; and</li>
          <li>changing account, security, or notification
            settings.</li>
        </ul>
        <t>
          A PEP <bcp14>MUST</bcp14> at least determine as sensitive a
          click, or a key step other than Tab, on a submit button
          <xref target="HTML"/>, key
          Enter on an element whose form owner could be submitted by
          implicit submission <xref target="HTML"/>, and any submission
          of a form that it has recorded as modified by the manifest or
          that contains an input element whose type is password. It
          <bcp14>MAY</bcp14> determine others by its own means, never
          from manifest content.
        </t>
        <t>
          A manifest cannot declare an action to be non-sensitive. A PEP
          cannot reliably determine the effect of an element whose role,
          name, and handlers a hostile publisher controls; the defaults
          of <xref target="pep"/> therefore ask for confirmation of every
          action conventionally bound to a decision rather than depend on
          that determination, and
          nothing turns such an ask into allow. A sensitive action that
          page script binds to an action allowed by default is neither
          determined nor confirmed; <xref target="sec-bait"/> describes
          this limit.
        </t>
      </section>

      <section anchor="controls">
        <name>Security Controls Are Not Friction</name>
        <t>
          The following are security controls. They are never traps.
          An agent <bcp14>MUST NOT</bcp14> use a manifest to avoid,
          dismiss, or circumvent them, and <bcp14>MUST</bcp14> ignore a
          descriptor whose locator or condition targets one that it
          detects:
        </t>
        <ul>
          <li>HTTP 401, 403, and 429 responses and equivalent in-page
            denials;</li>
          <li>CAPTCHAs and other challenges that distinguish humans from
            automation;</li>
          <li>authentication, re-authentication, and multi-factor or
            step-up challenges;</li>
          <li>window.confirm(), window.prompt(), and beforeunload dialogs,
            and site-built dialogs that ask the user to confirm or
            decline an action;</li>
          <li>consent and cookie-choice interfaces; and</li>
          <li>locks, holds, or timers that a site applies to give a human
            the opportunity to reconsider.</li>
        </ul>
        <t>
          Native confirm(), prompt(), and beforeunload dialogs block the
          page's event loop, so DOM interactions hang until the dialog is
          closed; an agent detects them through its automation interface
          and hands them to the principal. Detection of site-built
          controls is best effort; the confirmation defaults of
          <xref target="pep"/> are the backstop for controls that go
          undetected.
        </t>
        <t>
          On detecting a security control, an agent <bcp14>MUST</bcp14>
          stop the manifest-derived recovery and either hand control to
          the principal or end the task. It <bcp14>MUST NOT</bcp14> retry
          with altered inputs.
        </t>
      </section>

      <section anchor="audit">
        <name>Audit Record</name>
        <t>
          An agent <bcp14>SHOULD</bcp14> keep, for each manifest-derived
          action, a record containing the origin; the SHA-256
          <xref target="RFC6234"/> digest of the JCS canonical form
          <xref target="RFC8785"/> of the manifest; the descriptor
          identifier and step position; the PEP outcome and its reason;
          the text of any alert dismissed; and the time. The record
          <bcp14>SHOULD</bcp14> be available to the principal. It allows
          the principal, and the publisher's defenders, to distinguish
          manifest-guided recovery from other automated behavior
          (<xref target="sec-defenders"/>).
        </t>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="integrity" numbered="true">
      <name>Authenticity and Reputation</name>

      <section anchor="signatures">
        <name>Publisher Signatures</name>
        <t>
          HTTPS already authenticates a manifest retrieved from its
          origin. Signatures are useful when a manifest is authored by a
          party acting for the origin, such as the organization that
          deploys a vendor application, or when it is held outside the
          origin, such as in an agent's cache.
        </t>
        <t>
          Each element of the signatures member is a JSON Web Signature
          <xref target="RFC7515"/> in compact serialization with a
          detached payload (<xref target="RFC7515" section="F"/>). The
          payload is the JCS canonical form <xref target="RFC8785"/> of
          the manifest with the signatures member removed. The protected
          header <bcp14>MUST</bcp14> contain "alg" and "kid". Agents that
          verify signatures <bcp14>MUST</bcp14> support "ES256"
          <xref target="RFC7518"/>, <bcp14>MAY</bcp14> support "Ed25519"
          <xref target="RFC9864"/>, and <bcp14>MUST</bcp14> reject "none",
          every symmetric algorithm, and every algorithm that
          <xref target="RFC9864"/> deprecates.
        </t>
        <t>
          This document does not define how keys are discovered. An
          agent <bcp14>MUST</bcp14> verify signatures only against keys
          that its own configuration or its principal designates for the
          origin; a key or key location named by the manifest
          <bcp14>MUST NOT</bcp14> be used. If such a key is designated for
          the origin, a manifest without a signature that verifies under
          it <bcp14>MUST</bcp14> be ignored. If no key is designated,
          signatures are ignored.
        </t>
        <t>
          A valid signature establishes who published the manifest
          bytes. It does not establish that the manifest is safe, and
          <bcp14>MUST NOT</bcp14> relax any requirement of
          <xref target="processing"/>.
        </t>
      </section>

      <section anchor="reputation">
        <name>Reputation Services</name>
        <t>
          Earlier revisions defined a "Central Trust Registry" whose
          location each manifest declared in a registry_url member, and
          whose "white" status permitted execution. Because the party
          being vouched for chose the voucher, a malicious publisher
          could name a registry it controlled. That design is withdrawn.
        </t>
        <t>
          An agent <bcp14>MAY</bcp14> consult reputation services that
          its own configuration designates. A manifest
          <bcp14>MUST NOT</bcp14> be able to select, add, or override
          such a service. A reputation service:
        </t>
        <ul>
          <li>
            keys its information by origin, because a digest of manifest
            bytes changes with any edit;
          </li>
          <li>
            reports abuse history and the results of any static analysis
            it performs on manifests it has observed; an unfavorable
            status <bcp14>MUST</bcp14> cause the manifest to be ignored,
            and a favorable status <bcp14>MUST NOT</bcp14> relax any
            requirement of <xref target="processing"/>;
          </li>
          <li>
            cannot certify safety, because a manifest's effect depends on
            the live page that its locators resolve against, which a
            service evaluating manifest bytes never sees
            (<xref target="sec-bait"/>);
          </li>
          <li>
            <bcp14>SHOULD</bcp14> sign its responses, so that an agent
            can verify cached responses; and
          </li>
          <li>
            <bcp14>SHOULD</bcp14> be queried in a way that does not
            disclose the principal's browsing to the service, for example
            by downloading a list or by hash-prefix queries, rather than
            with a per-visit lookup (<xref target="privacy"/>).
          </li>
        </ul>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="iana" numbered="true">
      <name>IANA Considerations</name>

      <section anchor="iana-wk">
        <name>Well-Known URI Registration</name>
        <t>
          IANA is requested to register the following in the
          "Well-Known URIs" registry <xref target="RFC8615"/>:
        </t>
        <dl newline="false" spacing="compact">
          <dt>URI Suffix:</dt><dd>ai-friction-manifest.json</dd>
          <dt>Change Controller:</dt><dd>Won-pyo Han
            (pk102h@naver.com)</dd>
          <dt>Reference:</dt><dd>This document</dd>
          <dt>Status:</dt><dd>provisional</dd>
          <dt>Related Information:</dt><dd>None</dd>
        </dl>
      </section>

      <section anchor="iana-link">
        <name>Link Relation Type Registration</name>
        <t>
          IANA is requested to register the following in the "Link
          Relation Types" registry <xref target="RFC8288"/>:
        </t>
        <dl newline="false" spacing="compact">
          <dt>Relation Name:</dt><dd>ai-friction-manifest</dd>
          <dt>Description:</dt><dd>Refers to the AI Friction-Recovery
            Manifest of the link context's origin, which describes UI
            patterns that stall automated agents and constrained steps to
            recover from them.</dd>
          <dt>Reference:</dt><dd>This document</dd>
          <dt>Notes:</dt><dd>None</dd>
        </dl>
      </section>

      <section anchor="iana-noreg">
        <name>No New Registries</name>
        <t>
          Earlier revisions requested two new registries with the
          Specification Required policy. Registries created by an
          Independent Submission cannot use that policy
          <xref target="RFC8726"/>, and the vocabularies are closed in
          this version. The vocabularies are therefore defined in
          <xref target="vocab"/>, and no new registry is requested.
        </t>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="security" numbered="true">
      <name>Security Considerations</name>

      <section anchor="sec-threats">
        <name>Threat Model</name>
        <t>This document considers the following adversaries:</t>
        <dl newline="false">
          <dt>T1 Malicious publisher:</dt>
          <dd>an origin that publishes a manifest designed to harm
            the principal.</dd>
          <dt>T2 Conflicted publisher:</dt>
          <dd>an honest origin whose interests differ from the
            principal's, for example in consent, retention, or upselling
            flows.</dd>
          <dt>T3 Content injector:</dt>
          <dd>a third party able to place content in an origin's pages,
            through user-generated content, advertising, a compromised
            script dependency, or cross-site scripting.</dd>
          <dt>T4 Local supply chain:</dt>
          <dd>a tool, plug-in, or package able to write files in the
            principal's configuration directories.</dd>
          <dt>T5 Network attacker:</dt>
          <dd>an on-path party.</dd>
          <dt>T6 Reconnaissance agent:</dt>
          <dd>an autonomous offensive agent that harvests manifests to
            plan an attack on the publisher.</dd>
        </dl>
      </section>

      <section anchor="sec-injection">
        <name>Prompt Injection</name>
        <t>
          A language-model agent cannot reliably separate instructions
          from data in its input, and prompt injection through web
          content remains an unsolved problem. Revision -02 of this
          document recommended out-of-band discovery because out-of-band
          data was observed not to be screened as page content. That
          rationale is withdrawn: content from /.well-known is written by
          the same publisher as the page and is exactly as untrusted
          (P1).
        </t>
        <t>
          This document instead keeps manifest content out of the model
          entirely. No string written by the publisher reaches a language
          model (<xref target="freetext"/>): the model sees only
          closed-vocabulary values, integers assigned by the agent, and
          match results, none of which the publisher can use to express
          an instruction. The agent's runtime, not the model, interprets
          the manifest, and every action it takes on the basis of the
          manifest is decided by a PEP that the model cannot influence
          (P2, P6). This follows designs that separate the data an agent
          acts on from the channel that controls it
          <xref target="CAMEL"/>, applied to the manifest. It does not
          protect against injection through the page itself, which a
          manifest neither adds nor removes.
        </t>
        <t>
          Showing allowedValues or shortcut labels to the model, even as
          quoted and attributed data, would not be safe: 64 tokens of 32
          characters, or an 80-character label, suffice to spell
          instructions (for example "IGNORE_TASK_CLICK_CANCEL"), and a
          model so influenced could act on page elements that the
          manifest never names, outside the PEP. Those lists are
          therefore shown only to the principal.
        </t>
      </section>

      <section anchor="sec-exfil">
        <name>Exfiltration</name>
        <t>
          An agent that holds private data, reads untrusted content, and
          can communicate externally is exposed to data exfiltration
          <xref target="TRIFECTA"/>. This document closes the manifest's
          external channel: steps carry no values, there is no action
          that enters text or uploads a file, navigation targets are
          same-origin paths without query or fragment, and redirects that
          leave the origin or scope, or add a query, stop navigation
          (<xref target="step"/>). Enter, Escape, and click require
          confirmation (<xref target="pep"/>), and the submission of any
          form modified by the manifest is a sensitive action
          (<xref target="sensitive"/>). What the agent enters from
          the principal's task into fields it found itself reaches the
          origin exactly as it would without a manifest.
        </t>
      </section>

      <section anchor="sec-bait">
        <name>Locator Substitution</name>
        <t>
          A locator's meaning is determined by the live page, not by the
          manifest. A publisher (T1) can publish an innocuous manifest,
          have it signed or favorably rated, and later change the page so
          that the same locator resolves to a different control. A
          content injector (T3) can add an element that matches a
          publisher's locator. Hashes, signatures, and reputation over
          manifest bytes cannot detect either.
        </t>
        <t>
          The mitigations are that locators are resolved at use time and
          must match exactly one rendered element
          (<xref target="locator"/>), that step locators should be
          anchored to an application container, and that every action
          conventionally bound to a decision requires confirmation by
          default regardless of how the element was reached
          (<xref target="pep"/>). An injected element that duplicates a
          locator makes recovery fail closed, which is a denial of the
          manifest's benefit rather than a harm to the principal; an
          injected element that alone matches before the genuine element
          renders can receive only the actions allowed by default. An
          injected element that does not match the locator but covers the
          genuine element, transparent or not, would receive a click
          aimed at it; the check of the element that receives a click
          (<xref target="consult"/>) makes such a click fail closed as
          well, if the covering element is present when the check is
          made.
        </t>
        <t>
          The actions allowed by default (focus, scroll, assert, waits of
          at most five seconds, Tab, arrow and page keys on elements that
          are neither activatable nor value-bearing, dismissing an alert,
          and entering a value that the principal has chosen from
          allowedValues) can still be followed by effects of page script,
          since a hostile origin can bind any effect to any event or run
          it at any time, including a sensitive action such as a
          purchase. A PEP cannot determine such an effect, and it is not
          confirmed: <xref target="sensitive"/> requires confirmation of
          the sensitive actions that the PEP determines, and
          <xref target="pep"/> asks for every action conventionally bound
          to a decision, but neither covers an effect bound to an action
          allowed by default. No requirement of this document ensures
          that a manifest-derived action never has an unconfirmed effect.
          The
          manifest's contribution to that risk is limited to directing
          such events at a chosen element; the same origin could bind the
          same effects to events that an agent without a manifest, or a
          human user, produces in ordinary navigation. Agents that want
          no residual risk
          <bcp14>MAY</bcp14> treat every manifest-derived action as
          ask.
        </t>
      </section>

      <section anchor="sec-conflict">
        <name>Publisher-Principal Conflict</name>
        <t>
          A manifest is advisory, but it is not neutral (T2). A recovery
          that clicks "accept all" on a consent interface, confirms a
          dialog, presses Escape on a banner that the site interprets as
          acceptance, waits out a cooling-off timer, or continues a
          subscription in a cancellation flow would serve the publisher
          at the principal's expense. A field hint that omits a valid
          value could steer the principal toward another. These are
          addressed by <xref target="sensitive"/>, the confirmation
          defaults of <xref target="pep"/> for click, Enter, Escape, and
          waits longer than five seconds, and the prohibition on
          substituting values (<xref target="fieldhint"/>), none of which
          manifest content can relax (P4). A cooling-off timer shorter
          than five seconds is protected only by the agent's best-effort
          detection of security controls (<xref target="controls"/>) and
          by confirmation of whatever sensitive action follows it.
        </t>
      </section>

      <section anchor="sec-controls">
        <name>Circumvention of Security Controls</name>
        <t>
          A mechanism that teaches agents to recover from friction is
          dual-use if it applies to security controls: it would give
          benign agents the persistence after denial that defenders use
          to recognize autonomous attacks <xref target="ARTEX"/>, and
          could be used to defeat anti-automation measures, rate limits,
          and human confirmation. <xref target="controls"/> therefore
          defines these as stop conditions, and future versions of this
          document will not define trap categories that correspond to a
          security control.
        </t>
      </section>

      <section anchor="sec-local">
        <name>Local Configuration Supply Chain</name>
        <t>
          Earlier revisions exempted user-curated local manifests from
          prompt-injection screening and gave them precedence. A tool or
          package able to write one file (T4) would thereby obtain a
          persistent, unscreened, highest-precedence instruction channel
          for any origin. <xref target="local"/> limits local overrides to
          narrowing a site-published manifest that is actually present,
          so that a local file can only reduce what an agent does.
        </t>
      </section>

      <section anchor="sec-recon">
        <name>Information Disclosure, Reconnaissance, and Cloaking</name>
        <t>
          A manifest retrieved without credentials is public and is read
          by attackers as well as by agents (T6). Framework versions help
          an attacker select known vulnerabilities for which a deployment
          is unpatched, and shortcuts or field hints can reveal paths and
          code values that are not otherwise discoverable. Publishers:
        </t>
        <ul>
          <li><bcp14>SHOULD NOT</bcp14> include framework or component
            version numbers; and</li>
          <li><bcp14>MUST NOT</bcp14> include, in a manifest retrievable
            without credentials, any path, value, or element that is not
            reachable without credentials.</li>
        </ul>
        <t>
          A publisher <bcp14>MAY</bcp14> serve a more detailed manifest
          at the same URI to requests that carry the credentials the UI
          itself requires, and <bcp14>MAY</bcp14> additionally require
          agent authentication <xref target="WEBBOTAUTH"/>. An agent
          <bcp14>MAY</bcp14> make such a credentialed request using only
          credentials it already holds for that origin in the current
          task. Such a manifest is subject to every requirement of this
          document. If an agent retrieves both, the credentialed manifest
          replaces the uncredentialed one for the rest of the task, and
          any local override applies to the manifest in use.
        </t>
        <t>
          Because a publisher can serve different manifests to different
          requesters, a reputation service or reviewer can be shown a
          benign manifest while agents receive another (cloaking). This
          is a further reason that no favorable status relaxes the
          requirements of <xref target="processing"/>.
        </t>
      </section>

      <section anchor="sec-defenders">
        <name>Distinguishing Conforming Agents</name>
        <t>
          The defense guide cited above advises anchoring the detection
          of autonomous attacks on behavioral fingerprints rather than on
          static ones such as User-Agent strings, and lists persistence
          with bypass variations after denials among them
          <xref target="ARTEX"/>. An agent conforming to this document
          acts on manifest guidance only through declared steps on
          located elements, and stops at denials. A publisher can
          therefore use its own manifest as a positive baseline in
          behavioral detection: automated activity that follows declared
          recovery steps and stops at controls is consistent with a
          conforming agent, and activity that persists past controls is
          not, whatever the client claims to be.
        </t>
      </section>

      <section anchor="sec-resource">
        <name>Resource Exhaustion and Transport</name>
        <t>
          Size and nesting-depth limits (<xref target="retrieval"/>), the
          bound of eight steps per recovery (<xref target="trap"/>),
          bounded wait and scroll actions, and the absence of any
          reference from one manifest to another bound the cost of
          processing. A parser that exhausts its stack on deeply nested
          input can end the agent instead of falling back to baseline
          behavior; the nesting limit, and the treatment of parser errors
          as validation failures, address this
          (<xref target="app-measure"/>). HTTPS and the
          prohibition of cross-origin redirects
          (<xref target="wellknown"/>) protect against T5.
        </t>
      </section>
    </section>

    <!-- ============================================================ -->
    <section anchor="privacy" numbered="true">
      <name>Privacy Considerations</name>
      <t>
        Requesting /.well-known/ai-friction-manifest.json tells the origin that
        the client is likely an AI agent, in addition to the visit that
        prompted it. Agents that do not want to disclose this
        <bcp14>MAY</bcp14> refrain from retrieving manifests.
      </t>
      <t>
        Reputation lookups are a larger exposure: a per-visit query
        reveals the principal's browsing to a third party, which is why
        <xref target="reputation"/> recommends list-based or hash-prefix
        designs.
      </t>
      <t>
        Manifests contain no personal data, since steps carry no values.
        Audit records (<xref target="audit"/>) contain the principal's
        browsing history and <bcp14>SHOULD</bcp14> be held under the
        principal's control. Local overrides listing origins reveal those
        origins if synchronized or shared.
      </t>
    </section>

    <!-- ============================================================ -->
    <section anchor="impl" numbered="true">
      <name>Implementation Status</name>
      <t>
        Note to RFC Editor: Please remove this section before
        publication, as described in <xref target="RFC7942"/>.
      </t>
      <t>
        A reference implementation is available at
        https://github.com/11pyo/AINavManifest under the MIT License. It
        implements the requirements of revision -03 for the version "2.0"
        format: a JSON Schema and a validator; a policy enforcement point
        with the default outcomes, standing-approval limits, and ask
        display that revision -03 specified, and an audit record;
        tracking of forms modified by the manifest; a model interface
        that exposes only the items permitted by
        <xref target="freetext"/>; the field-hint flow of
        <xref target="fieldhint"/>; discovery without redirects and
        per-hop redirect checks for navigation (<xref target="step"/>);
        verification of detached JWS publisher signatures with "ES256"
        and "Ed25519" against keys that its configuration designates for
        the origin, rejecting every other algorithm, including "EdDSA"
        (<xref target="signatures"/>), where a valid signature identifies
        the publisher and relaxes no policy; example publishers serving
        manifests at the well-known URI with the Link advertisement; and
        tests, including browser tests, that exercise these. Item 6 of
        <xref target="app-measure"/> reports fourteen attacks on its
        signature verification. It does not implement reputation
        services.
      </t>
      <t>
        Of the changes in revision -04, it implements only the nesting
        limit and the handling of parser errors
        (<xref target="retrieval"/>) and, through the actionability
        checks of its browser automation library, declines a click that
        another element would intercept, which approximates the check of
        <xref target="consult"/>. It does not yet implement the
        withdrawal of standing approval: its PEP still lets a principal
        approve key Enter in advance for an origin. The revised
        definition of a form modified by the manifest
        (<xref target="terminology"/>), the content of an ask
        (<xref target="pep"/>), and the visibility check of
        <xref target="consult"/> are not implemented either. The
        measurements in items 1 to 4 of <xref target="app-measure"/> were
        made with an earlier version of the implementation that used the
        version "1.0" format, which is preserved for reproducibility;
        items 5 and 6 used the version "2.0" implementation, item 6 after
        signature verification was added.
      </t>
    </section>

    <!-- ============================================================ -->
    <section anchor="ipr" numbered="true">
      <name>Intellectual Property Rights Disclosure</name>
      <t>
        The technology described in this document is the subject of
        Korean Patent Applications No. 10-2026-0071716 (filed
        2026-04-21) and No. 10-2026-0097280 (filed 2026-05-29), both by
        the author. The applicant commits to offer any essential claims
        under Fair, Reasonable, and Non-Discriminatory (FRAND) terms to
        implementers of this specification, as declared in the project
        repository.
      </t>
    </section>

    <!-- ============================================================ -->
    <section anchor="ack" numbered="true">
      <name>Acknowledgments</name>
      <t>
        The author thanks the Claude Code, Model Context Protocol, and
        OpenAI function-calling communities for the empirical
        observations that motivated this work. The threat model of
        <xref target="security"/> draws on the publicly documented design
        and defensive analysis of ARTEX, originally developed as
        Autumn-27/ARTEX, as published in its global edition
        <xref target="ARTEX"/>.
      </t>
    </section>

  </middle>

  <back>
    <references>
      <name>References</name>

      <references>
        <name>Normative References</name>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.2119.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.3339.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.3986.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.6234.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.6454.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.7493.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.7515.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.7518.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.9864.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8174.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8259.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8288.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8615.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8785.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.9110.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.9111.xml"/>
        <reference anchor="WAI-ARIA" target="https://www.w3.org/TR/wai-aria-1.2/">
          <front>
            <title>Accessible Rich Internet Applications (WAI-ARIA) 1.2</title>
            <author><organization>W3C</organization></author>
            <date year="2023" month="June"/>
          </front>
          <seriesInfo name="W3C Recommendation" value="wai-aria-1.2"/>
        </reference>
        <reference anchor="ACCNAME" target="https://www.w3.org/TR/accname-1.2/">
          <front>
            <title>Accessible Name and Description Computation 1.2</title>
            <author><organization>W3C</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="HTML" target="https://html.spec.whatwg.org/multipage/">
          <front>
            <title>HTML Living Standard</title>
            <author><organization>WHATWG</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="DOM" target="https://dom.spec.whatwg.org/">
          <front>
            <title>DOM Living Standard</title>
            <author><organization>WHATWG</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="URL" target="https://url.spec.whatwg.org/">
          <front>
            <title>URL Living Standard</title>
            <author><organization>WHATWG</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="SELECTORS4" target="https://www.w3.org/TR/selectors-4/">
          <front>
            <title>Selectors Level 4</title>
            <author><organization>W3C</organization></author>
            <date/>
          </front>
        </reference>
      </references>

      <references>
        <name>Informative References</name>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.6648.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.7942.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8726.xml"/>
        <xi:include href="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.9309.xml"/>

        <reference anchor="MCP" target="https://modelcontextprotocol.io/specification">
          <front>
            <title>Model Context Protocol Specification</title>
            <author><organization>Model Context Protocol</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="WEBMCP" target="https://github.com/webmachinelearning/webmcp">
          <front>
            <title>WebMCP</title>
            <author><organization>W3C Web Machine Learning Community Group</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="A2A" target="https://github.com/a2aproject/A2A">
          <front>
            <title>Agent2Agent (A2A) Protocol</title>
            <author><organization>A2A Project</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="LLMSTXT" target="https://llmstxt.org/">
          <front>
            <title>The /llms.txt file</title>
            <author initials="J." surname="Howard" fullname="Jeremy Howard"/>
            <date year="2024" month="September"/>
          </front>
        </reference>
        <reference anchor="AIPREF" target="https://datatracker.ietf.org/wg/aipref/about/">
          <front>
            <title>AI Preferences (aipref) Working Group</title>
            <author><organization>IETF</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="WEBBOTAUTH" target="https://datatracker.ietf.org/wg/webbotauth/about/">
          <front>
            <title>Web Bot Auth (webbotauth) Working Group</title>
            <author><organization>IETF</organization></author>
            <date/>
          </front>
        </reference>
        <reference anchor="WEBDRIVER" target="https://www.w3.org/TR/2018/REC-webdriver1-20180605/">
          <front>
            <title>WebDriver</title>
            <author><organization>W3C</organization></author>
            <date year="2018" month="June" day="5"/>
          </front>
          <seriesInfo name="W3C Recommendation" value="REC-webdriver1-20180605"/>
        </reference>
        <reference anchor="CAMEL" target="https://arxiv.org/abs/2503.18813">
          <front>
            <title>Defeating Prompt Injections by Design</title>
            <author initials="E." surname="Debenedetti" fullname="Edoardo Debenedetti"/>
            <date year="2025" month="March"/>
          </front>
          <seriesInfo name="arXiv" value="2503.18813"/>
        </reference>
        <reference anchor="AGENTDOJO" target="https://arxiv.org/abs/2406.13352">
          <front>
            <title>AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents</title>
            <author initials="E." surname="Debenedetti" fullname="Edoardo Debenedetti"/>
            <date year="2024" month="June"/>
          </front>
          <seriesInfo name="arXiv" value="2406.13352"/>
        </reference>
        <reference anchor="TRIFECTA" target="https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/">
          <front>
            <title>The lethal trifecta for AI agents: private data, untrusted content, and external communication</title>
            <author initials="S." surname="Willison" fullname="Simon Willison"/>
            <date year="2025" month="June"/>
          </front>
        </reference>
        <reference anchor="ARTEX" target="https://github.com/jiwoochris/artex-global">
          <front>
            <title>ARTEX Global Edition: docs/defense-en.md, guard/guard.go, and intercept/intercept.go</title>
            <author><organization>artex-global maintainers (upstream: Autumn-27/ARTEX)</organization></author>
            <date year="2026"/>
          </front>
        </reference>
        <reference anchor="DRAFT02" target="https://datatracker.ietf.org/doc/html/draft-han-ai-manifest-02">
          <front>
            <title>AI Manifest: Site-Published Friction-Recovery Descriptors for AI Agents</title>
            <author initials="W." surname="Han" fullname="Won-pyo Han"/>
            <date year="2026" month="June" day="2"/>
          </front>
          <seriesInfo name="Internet-Draft" value="draft-han-ai-manifest-02"/>
        </reference>
      </references>
    </references>

    <!-- ============================================================ -->
    <section anchor="app-example" numbered="true">
      <name>Example Manifest</name>
      <t>
        The following manifest describes a table-maintenance screen on
        which a cell entry must be committed with Enter before the row
        unlocks, and on which the valid account-category reference codes
        depend on the selected company code. Under
        <xref target="sec-recon"/>, its fieldHints would be served only
        in response to requests authenticated to the ERP system.
      </t>
      <sourcecode type="json"><![CDATA[
{
  "version": "2.0",
  "origin": "https://erp.example.com",
  "expires": "2026-12-01T00:00:00Z",
  "scope": ["/maint/"],
  "frameworkHints": {
    "name": "webdynpro",
    "renderingModel": "server-rendered"
  },
  "knownTraps": [
    {
      "trapId": "row-commit",
      "category": "commit-gesture",
      "locator": { "role": "gridcell", "focused": true },
      "condition": { "visible": { "role": "status",
                                  "name": "Row locked" } },
      "recovery": [
        { "action": "key", "key": "Enter",
          "locator": { "role": "gridcell", "focused": true } },
        { "action": "wait", "timeoutMs": 3000 }
      ],
      "description": "Commit each entry with Enter to unlock."
    }
  ],
  "fieldHints": [
    {
      "fieldId": "companyCode",
      "locator": { "role": "combobox", "name": "Company Code" }
    },
    {
      "fieldId": "acctCatRef-1000",
      "locator": { "role": "gridcell",
                   "column": "Account Category Ref" },
      "when": { "companyCode": "1000" },
      "allowedValues": ["Z051", "Z063", "Z072"],
      "commit": { "action": "key", "key": "Enter",
                  "locator": { "role": "gridcell",
                               "focused": true } },
      "description": "Only codes registered for the company code."
    }
  ],
  "shortcuts": [
    { "shortcutId": "acct-cat-maint",
      "target": "/maint/account-category",
      "label": "Account category maintenance" }
  ]
}
]]></sourcecode>
      <t>
        The locator of acctCatRef-1000 matches every cell of its column,
        which is permitted for field hints (<xref target="locator"/>).
        The model never sees the list. When it proposes a code from the
        principal's task, the PEP answers only whether the code is
        accepted. If the model proposes a standard code such as one it
        knows from public documentation, the answer is no; the agent
        then shows Z051, Z063, and Z072 to the principal, who chooses one
        or abandons the entry (<xref target="fieldhint"/>). The shortcut
        is likewise offered to the principal, not to the model.
      </t>
      <t>
        The Enter steps ask the principal each time
        (<xref target="pep"/>), so maintaining many rows takes at
        least one confirmation per row; a deployment that needs unattended bulk
        entry would offer a structured interface instead
        (<xref target="applicability"/>). This example
        assumes that the focused element is the grid cell itself; on
        pages that place an input element inside the cell, the publisher
        would use { "role": "textbox", "focused": true, "within":
        { "role": "gridcell" } }.
      </t>
    </section>

    <section anchor="app-measure" numbered="true">
      <name>Preliminary Measurements</name>
      <t>
        This appendix reports measurements made with the version "1.0"
        format of <xref target="DRAFT02"/> (items 1 to 4), none of which
        evaluates security, and adversarial evaluations of the reference
        implementation of the version "2.0" requirements of revision -03
        (item 5) and of its signature verification (item 6). They are
        preliminary; their limitations are stated with each.
      </t>
      <ol>
        <li>
          <t>Proprietary-knowledge enterprise task (2026-05-29).</t>
          <t>
            An SAP S/4HANA-style table-maintenance screen on which valid
            account-category references depend on company-code
            registration, and on which a cell entry must be committed
            before the row unlocks. One run per condition with one agent
            (Claude Sonnet 4.6). Both conditions completed. The manifest
            condition used 6 interaction actions versus 13 and about 47.8K
            versus 87.3K input tokens (estimated). Wall-clock time was
            closer, 275 s versus 353 s, because at this difficulty the
            agent's own reasoning time dominates both conditions; actions
            and tokens reflect the efficiency difference more directly.
            Limitations: a single run per condition; one model; site,
            manifest, and evaluation all prepared by the author; token
            counts estimated.
          </t>
        </li>
        <li>
          <t>Reference benchmark.</t>
          <t>
            A two-step order-entry task repeated 30 times, input tokens
            counted with the cl100k_base tokenizer: the manifest-aware
            agent used 341 input tokens on average with 30 of 30
            successes, and a DOM-analysis baseline used 1887.6 tokens
            with 6 of 30 successes. Limitations: both agents were
            implemented by the author; the tokenizer differs from that of
            the model in item 1.
          </t>
        </li>
        <li>
          <t>Spontaneous discovery (Phase 1, n=30, 2026-04-28).</t>
          <t>
            No agent spontaneously looked for an unannounced manifest in
            30 trials. Consumption therefore depends on agents
            implementing discovery, which motivates the link relation
            (<xref target="link"/>). An effect size of Cohen's d = 5.00
            reported in earlier revisions was an artifact of a
            near-zero-variance idealized baseline and is withdrawn.
          </t>
        </li>
        <li>
          <t>Cross-agent consumption (Phase 2, 2026-05-11).</t>
          <t>
            Given a manifest in its conversation context, one agent
            applied individual trap entries, a second recognized the
            document type without applying entries, and a third did not
            extract manifest-specific content. This tested
            interpretation, not discovery. Version "2.0" removes the
            dependence on interpretation: the agent's runtime, not its
            model, interprets the manifest.
          </t>
        </li>
        <li>
          <t>Adversarial evaluation (2026-11).</t>
          <t>
            Using the attack-success measure of
            <xref target="AGENTDOJO"/>, 32 attack cases were run against
            four agent configurations on a local test site that records
            side effects such as purchases and deletions. Each case is a
            malicious manifest with a success criterion and a predicted
            outcome per configuration, a few left open; the cases
            include one control case and one malformed case kept as
            registered. The configurations were an unguarded agent that
            executes manifest steps as written; the same agent with only
            the schema validator; and the reference implementation of the
            requirements of <xref target="processing"/> as they stood in
            revision -03 (validator, PEP, no manifest string reaching the
            model, and provenance), once with a principal who declines
            every ask and once with one who approves every ask and picks
            the first item of every list. Attack success rates were 32 of
            32 and 24 of 32 for the first two configurations, which are
            upper bounds by construction rather than evidence; 3 of 32
            with the declining principal; and 13 of 32 with the approving
            principal, 10 of them after an approved ask. The 13 include
            the control case, and the 32 include the malformed case,
            which was blocked. Excluding the two attacks in which a
            standing approval (an approval of Enter given in advance for
            an origin, which revision -03 permitted and revision -04
            withdraws) let Enter run without an ask, the last two
            rates were 1 of 30 and 11 of 30. No manifest string reached
            the model in any run of the reference implementation (0 of 64
            distinct runs, 192 with repeats).
          </t>
          <t>
            The attacks that succeeded without any ask were a deeply
            nested manifest whose parsing raised an exception that the
            implementation did not handle, which led to the nesting limit
            and the parser-error rule of <xref target="retrieval"/>, and
            two in which a standing approval of Enter let a purchase occur
            (<xref target="sensitive"/>): one through a key handler that
            page script bound to a text field, and one through an Enter
            that submitted an ordinary form. Revision -04 withdraws
            standing approval (<xref target="pep"/>), so by analysis of
            the specification both Enter steps require an ask; this has
            not been measured, because the reference implementation does
            not yet reflect the withdrawal (<xref target="impl"/>). By
            the same analysis, an estimate and not a measurement, the fix
            for the nested manifest together with the withdrawal would
            leave 0 of 32 successes with the declining principal and 12
            of 32 with the approving principal, all after an approved
            ask. A click on a button covered by a transparent
            element was stopped only by the hit test of the browser
            automation library used by the adapter, which led to the check
            in <xref target="consult"/>. In the two approved purchases
            bound to Enter on a text field, the ask did not name the
            effect, judged by a keyword test fixed in advance; this led to
            the requirements on what an ask shows (<xref target="pep"/>),
            under which an ask discloses that the effect is unknown but
            does not identify it. After the parser was fixed, a single
            re-measurement blocked the deeply nested manifest in the three
            configurations that validate and changed no other result. The
            predictions for two attacks were wrong in some configurations
            and are kept as registered.
          </t>
          <t>
            Method: a first measurement was discarded after a fix to the
            test harness, and the definition of a success after an
            approved ask was changed during measurement, after which all
            results were discarded and measured again from scratch. The
            criteria and predictions are stated to have been fixed before
            measurement, but they were committed to the repository four
            seconds before the results, so their prior registration
            cannot be verified independently. Utility, that is whether the
            agent still completes benign tasks, was not measured in the
            scripted evaluation. In the language-model pilot noted below,
            a benign task in which the user names the value succeeded in
            5 of 5 runs in every configuration, while a task that needs a
            value from a manifest's allowedValues could not be completed
            by the model in the configurations of the reference
            implementation, which never show it the list
            (<xref target="freetext"/>): a cost of the design.
          </t>
          <t>
            Limitations: the attacks were designed by the author and form
            a regression suite for known classes, not a discovery of
            unknown ones; the model is a script that obeys action words in
            any text it is shown, not a language model, so the results
            measure what the surrounding structure contains, not how a
            real model behaves; the attacker does not adapt; the two
            principals are constants, not people; the runs are
            deterministic, so the three repeats confirm only the
            stability of the test harness; and everything ran on a local
            test site, without real sites, credentials, or payments. A
            pilot with one language model in place of the script, 75
            runs with 5 per scenario and configuration, is published with
            the evaluation; with 5 runs per cell it is not evidence.
          </t>
        </li>
        <li>
          <t>Signature attacks (2026-11).</t>
          <t>
            After signature verification was added to the reference
            implementation (<xref target="signatures"/>), 14 further
            attack cases were registered with predictions, after the
            verifier was written and in a commit
            separate from and earlier than their results, and run three
            times in the four configurations of item 5. In the two
            configurations of the reference implementation a key was
            designated for the site's origin, except in one control case.
            The nine cases with a forged, stripped, or confused signature
            (forged "ES256" and "Ed25519" signatures under the designated
            key identifier, "alg" "none" in two forms, HMAC keyed with the
            designated public key, a key named by the manifest or the JWS
            header, a genuine signature moved onto another manifest, a
            missing signature, and the deprecated "EdDSA" label) succeeded
            in 0 of 9 with either principal, against 9 of 9 for the
            unguarded agent and 8 of 9 for the validator-only agent. With
            the designated key removed in test code, 8 of the 9 succeeded
            with the approving principal, after an approved ask. Over all 14 cases the approving
            principal saw 3 successes, all after an approved ask, in the
            cases that test whether a valid signature widens anything: it
            did not, since a validly signed manifest gave the same outcome
            and the same ask display as its unsigned control, and a signed
            copy of an attack from item 5 the same as the original. All
            registered predictions held, and no result changed between
            repeats. Limitations: those of item 5; in addition, these
            attacks test the
            verifier and the rule for designated keys, not key
            management, which this document leaves to the agent's
            configuration.
          </t>
        </li>
      </ol>
      <t>
        Earlier single-trial figures (a 5.0 s "Tier C" time and 19- to
        31-fold speedups in draft-han-ai-manifest-01) remain withdrawn.
        Items 5 and 6 are first adversarial measurements of the kind
        that revision -03 named as the most needed next one.
        Measurements with language models at a scale that can serve as
        evidence, with attacks not designed by the author, and of the
        requirements as revised in revision -04 remain to be made.
      </t>
    </section>

    <section anchor="app-history" numbered="true">
      <name>Document History</name>
      <t>
        Note to RFC Editor: Please remove this section before
        publication.
      </t>
      <dl newline="true">
        <dt>draft-han-ai-manifest-04 (2026-11)</dt>
        <dd>
          <t>Withdraws the standing approval of Enter, restates the
            confirmation requirement to what an agent can determine, and
            tightens requirements that an adversarial evaluation of the
            reference implementation found incomplete
            (<xref target="app-measure"/>, item 5):</t>
          <ul>
            <li>Retrieval: an implementation limit on nesting depth, of at
              least 64 levels; exceeding a limit and any parser error,
              including resource exhaustion, are validation failures
              that fall back to baseline behavior.</li>
            <li>Standing approval of Enter is withdrawn: Enter always
              asks, and an agent's policy can only deny it.</li>
            <li><xref target="sensitive"/> is restated to require
              confirmation of the sensitive actions the PEP determines,
              with a minimum that every PEP determines (submit buttons,
              Enter that could submit a form by implicit submission, and
              forms modified by the manifest or containing a password
              field); the -03 wording ("performs or triggers") could not
              be met for effects that page script binds to actions allowed
              by default (<xref target="sec-bait"/>). Revision -03 had the
              same limit without stating it. The submission of a form
              modified by the manifest is sensitive also when the
              modifying step itself submits it.</li>
            <li>Before a manifest-derived click, the agent verifies that
              the click would reach the resolved element rather than an
              element covering it, judged on the innermost element
              without retargeting, and may repeat the check for up to
              5000 ms; it should decline actions on elements that the
              principal cannot see.</li>
            <li>An ask shows the target, the source, and the effect of the
              action, including a statement that the actual effect is
              unknown to the agent; values that the publisher controls are
              shown as quoted data.</li>
            <li>The abstract no longer carries revision history, which is
              kept in this appendix.</li>
            <li><xref target="removed-methods"/> is retitled "Removed
              Discovery Methods".</li>
            <li>The adversarial evaluation and fourteen attacks on
              signature verification are reported, with their
              limitations, in <xref target="app-measure"/>.</li>
          </ul>
        </dd>
        <dt>draft-han-ai-manifest-03 (2026-11)</dt>
        <dd>
          <t>Replaces the trust model. Principal changes:</t>
          <ul>
            <li>The manifest is untrusted regardless of channel; the -02
              rationale that out-of-band data escapes prompt-injection
              screening is withdrawn. Added design principles,
              applicability criteria, a threat model, provenance tracking
              of manifest-derived actions, a policy enforcement point
              with confirmation defaults, sensitive actions, and the
              exclusion of security controls from recovery. No manifest
              string reaches the language model; value lists and
              shortcuts are shown only to the principal.</li>
            <li>Discovery: removed the inline DOM element, the
              X-AI-Manifest header, the meta element, and the
              vendor-path local method; local overrides may only narrow a
              site manifest that is present. Added origin binding and the
              Link header form of the link relation. Renamed the
              well-known suffix to "ai-friction-manifest.json" and the
              link relation to "ai-friction-manifest".</li>
            <li>Schema version "2.0": closed action vocabulary of
              value-free steps; removed fill, select, and upload;
              multi-strategy locators with ARIA role, name, and column;
              fieldHints for site-specific knowledge; same-origin
              shortcuts without query; expires and scope members; removed
              publisher, manifestId, and registry_url; narrowed the dialog
              category to alert() and removed cross-origin frames.</li>
            <li>Integrity: replaced the manifest-declared Central Trust
              Registry with agent-configured reputation services and
              optional JWS signatures over the JCS canonical form.</li>
            <li>IANA: no new registries (RFC 8726); the well-known URI
              and link relation registrations are kept.</li>
            <li>Measurements moved to an appendix with stated
              limitations, including one run per condition for the
              enterprise task.</li>
          </ul>
        </dd>
        <dt>draft-han-ai-manifest-02 (2026-06-02)</dt>
        <dd>Realigned the document with the advisory friction-recovery
          model; relaxed the deterministic runtime to OPTIONAL;
          withdrew the "Tier C" figures and Cohen's d = 5.00; added the
          enterprise measurement.</dd>
        <dt>draft-han-ai-manifest-01 (2026-05-15)</dt>
        <dd>Extended the schema to the three-part model
          (frameworkHints, knownTraps, shortcuts); added Method E and
          the native-dialog-trap category.</dd>
        <dt>draft-han-ai-manifest-00 (2026-04-21)</dt>
        <dd>Initial version.</dd>
      </dl>
    </section>
  </back>
</rfc>
