AGENTPROTO Z. Sarker Internet-Draft R. Pries Intended status: Standards Track T. Reddy Expires: 21 May 2027 Nokia 17 November 2026 Agent Dialog Propagation Protocol draft-zahed-agentdiaproto-00 Abstract This document specifies the Agent Dialog Propagation Protocol, a protocol for maintaining and propagating dialog context between AI agents, tools, and intermediaries. The protocol enables messages to remain associated with a dialog across transport connections, intermediaries, administrative domains, and delegation relationships. It defines mechanisms for dialog identification, dialog lifecycle management and the association of identity, authorization, and delegation information with a dialog. The Agent Dialog Propagation Protocol does not define agent reasoning or execution state, identity credentials, authorization policy, or transport protocols. These functions are provided by application protocols and existing security and transport mechanisms. Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at https://datatracker.ietf.org/drafts/current/. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on 21 May 2027. Copyright Notice Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved. Sarker, et al. Expires 21 May 2027 [Page 1] Internet-Draft Agent Dialog Propagation Protocol November 2026 This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/ license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4 1.1. Objectives . . . . . . . . . . . . . . . . . . . . . . . 4 2. Conventions and Definitions . . . . . . . . . . . . . . . . . 5 3. Protocol Model . . . . . . . . . . . . . . . . . . . . . . . 5 3.1. Dialogs . . . . . . . . . . . . . . . . . . . . . . . . . 5 3.2. Dialog Context . . . . . . . . . . . . . . . . . . . . . 5 3.3. Protocol Invariants . . . . . . . . . . . . . . . . . . . 6 3.4. Architectural Boundary . . . . . . . . . . . . . . . . . 6 4. Dialog Context Lifecycle . . . . . . . . . . . . . . . . . . 7 4.1. Dialog Continuity . . . . . . . . . . . . . . . . . . . . 9 5. Dialog Context Data Model . . . . . . . . . . . . . . . . . . 9 5.1. Version (ver) . . . . . . . . . . . . . . . . . . . . . . 10 5.2. ADID . . . . . . . . . . . . . . . . . . . . . . . . . . 11 5.3. State . . . . . . . . . . . . . . . . . . . . . . . . . . 11 5.4. Sender . . . . . . . . . . . . . . . . . . . . . . . . . 12 5.5. Recipient . . . . . . . . . . . . . . . . . . . . . . . . 12 5.6. Expires . . . . . . . . . . . . . . . . . . . . . . . . . 12 5.7. Extensions . . . . . . . . . . . . . . . . . . . . . . . 13 5.8. Log . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 5.9. Modality . . . . . . . . . . . . . . . . . . . . . . . . 13 5.10. Authorization Associations . . . . . . . . . . . . . . . 14 6. Protocol Operations . . . . . . . . . . . . . . . . . . . . . 14 6.1. Establish . . . . . . . . . . . . . . . . . . . . . . . . 14 6.2. Propagate . . . . . . . . . . . . . . . . . . . . . . . . 15 6.3. Update . . . . . . . . . . . . . . . . . . . . . . . . . 16 6.4. Suspend . . . . . . . . . . . . . . . . . . . . . . . . . 16 6.5. Resume . . . . . . . . . . . . . . . . . . . . . . . . . 16 6.6. Terminate . . . . . . . . . . . . . . . . . . . . . . . . 16 6.7. Authorization Checkpoint . . . . . . . . . . . . . . . . 17 6.8. Error Reporting . . . . . . . . . . . . . . . . . . . . . 17 7. Intermediaries . . . . . . . . . . . . . . . . . . . . . . . 17 7.1. Transparent Relay . . . . . . . . . . . . . . . . . . . . 17 7.2. Dialog-Aware Proxy . . . . . . . . . . . . . . . . . . . 17 7.3. Domain Gateway . . . . . . . . . . . . . . . . . . . . . 18 7.4. Protocol Adapter . . . . . . . . . . . . . . . . . . . . 18 7.5. Intermediary Processing Rules . . . . . . . . . . . . . . 18 8. Identity and Authentication Integration . . . . . . . . . . . 18 Sarker, et al. Expires 21 May 2027 [Page 2] Internet-Draft Agent Dialog Propagation Protocol November 2026 9. Authorization and Delegation Integration . . . . . . . . . . 19 10. Cross-Domain Operation . . . . . . . . . . . . . . . . . . . 19 11. Recovery and Resynchronisation . . . . . . . . . . . . . . . 20 12. Versioning and Capability Negotiation . . . . . . . . . . . . 21 13. Error Codes . . . . . . . . . . . . . . . . . . . . . . . . . 21 13.1. Dialog Errors . . . . . . . . . . . . . . . . . . . . . 21 13.1.1. unknown-dialog . . . . . . . . . . . . . . . . . . . 21 13.1.2. expired-dialog . . . . . . . . . . . . . . . . . . . 21 13.1.3. invalid-state . . . . . . . . . . . . . . . . . . . 21 13.1.4. invalid-transition . . . . . . . . . . . . . . . . . 22 13.2. Sequence Errors . . . . . . . . . . . . . . . . . . . . 22 13.2.1. duplicate-operation . . . . . . . . . . . . . . . . 22 13.2.2. sequence-conflict . . . . . . . . . . . . . . . . . 22 13.3. Authorization Errors . . . . . . . . . . . . . . . . . . 22 13.3.1. authorization-invalid . . . . . . . . . . . . . . . 22 13.3.2. authorization-expired . . . . . . . . . . . . . . . 22 13.3.3. insufficient-authority . . . . . . . . . . . . . . . 22 13.3.4. approval-denied . . . . . . . . . . . . . . . . . . 22 13.3.5. approval-timeout . . . . . . . . . . . . . . . . . . 22 13.4. Delegation Errors . . . . . . . . . . . . . . . . . . . 22 13.4.1. invalid-delegation . . . . . . . . . . . . . . . . . 23 13.5. Protocol Errors . . . . . . . . . . . . . . . . . . . . 23 13.5.1. unsupported-version . . . . . . . . . . . . . . . . 23 13.5.2. malformed-context . . . . . . . . . . . . . . . . . 23 13.5.3. unsupported-extension . . . . . . . . . . . . . . . 23 13.5.4. timeout . . . . . . . . . . . . . . . . . . . . . . 23 14. Protocol Extensions . . . . . . . . . . . . . . . . . . . . . 23 15. Protocol Bindings . . . . . . . . . . . . . . . . . . . . . . 24 16. Security Considerations . . . . . . . . . . . . . . . . . . . 24 16.1. Transport Security . . . . . . . . . . . . . . . . . . . 24 16.2. Integrity Across Intermediaries . . . . . . . . . . . . 25 16.3. Authorization and Delegation . . . . . . . . . . . . . . 25 16.4. Dialog Identifier Hijacking . . . . . . . . . . . . . . 25 16.5. Replay . . . . . . . . . . . . . . . . . . . . . . . . . 25 16.6. Context Substitution . . . . . . . . . . . . . . . . . . 25 16.7. Intermediary Modification . . . . . . . . . . . . . . . 25 16.8. Stale Authorization . . . . . . . . . . . . . . . . . . 26 16.9. Replayable Early Data . . . . . . . . . . . . . . . . . 26 17. Privacy Considerations . . . . . . . . . . . . . . . . . . . 26 18. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 26 19. References . . . . . . . . . . . . . . . . . . . . . . . . . 27 19.1. Normative References . . . . . . . . . . . . . . . . . . 27 19.2. Informative References . . . . . . . . . . . . . . . . . 27 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 29 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 29 Sarker, et al. Expires 21 May 2027 [Page 3] Internet-Draft Agent Dialog Propagation Protocol November 2026 1. Introduction AI agents can participate in interactions that span multiple messages, transport connections, intermediaries, and administrative domains. Such interactions can include user-to-agent, agent-to- agent, and agent-to-tool communication. A dialog can continue after the underlying transport connection has been replaced or interrupted. Application protocols provide mechanisms for exchanging requests, responses, results, and application-specific information. However, additional protocol mechanisms are required to maintain the communication context that associates such exchanges with a persistent dialog. The Agent Dialog Propagation Protocol provides these mechanisms. In particular, it maintains a Dialog Context that can be propagated between participants and intermediaries independently of a particular transport connection. The protocol associates the Dialog Context with participant identities, authorization context, and delegation information. It does not define identity credentials or authorization policies. The Agent Dialog Propagation Protocol is intended to be usable with existing application protocols, including A2A [A2A] and MCP [MCP], through suitable protocol bindings or extension points. It does not replace those application protocols. The requirements addressed by this document are described in [I-D.agentic-ai-usecases-requirements]. Binding of the Dialog Context to WIMSE and OAuth credentials is described in [I-D.reddy-wimse-oauth-dialog-binding]. *_EDITOR'S NOTE:_* Task handling is still under discussion. 1.1. Objectives The Agent Dialog Propagation Protocol is designed to provide the following functions: * stable identification of a dialog independent of a transport connection; * propagation of Dialog Context between dialog participants; * lifecycle management of Dialog Context; Sarker, et al. Expires 21 May 2027 [Page 4] Internet-Draft Agent Dialog Propagation Protocol November 2026 * association of authorization and delegation information with a dialog; * dialog continuity following transport interruption or replacement; * operation through intermediaries; * operation across administrative and trust domains; and * protocol extensibility. The protocol does not authenticate participants, issue credentials or authorization tokens, or evaluate authorization policy. Such functions are provided by other protocol mechanisms. 2. Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. This document uses terms defined in [I-D.manyfolks-agentproto-arch]. In addition, the following terms are defined in this document: Intermediary: An entity that relays or processes messages between dialog participants. 3. Protocol Model 3.1. Dialogs A dialog provides the communication scope within which participants exchange messages. Each dialog is associated with exactly one Dialog Context. 3.2. Dialog Context The Dialog Context contains protocol-visible state necessary to maintain a dialog. It includes, as applicable: * the Dialog Context Identifier; * dialog lifecycle state; Sarker, et al. Expires 21 May 2027 [Page 5] Internet-Draft Agent Dialog Propagation Protocol November 2026 * identity associations; * authorization associations; * delegation associations; and * dialog relationship information. 3.3. Protocol Invariants The following invariants apply to the protocol: 1. A dialog MUST be identifiable independently of the underlying transport connection. 2. Knowledge of a Dialog Context Identifier MUST NOT be sufficient to authorise participation in a dialog. 3. Dialog Context MUST remain distinguishable from application and LLM context. 4. Delegation MUST NOT implicitly broaden authority. 5. Replacement or interruption of a transport connection MUST NOT, by itself, terminate a dialog. 3.4. Architectural Boundary Figure 1 shows the position of the Agent Dialog Propagation Protocol relative to application protocols and transport. Sarker, et al. Expires 21 May 2027 [Page 6] Internet-Draft Agent Dialog Propagation Protocol November 2026 +------------------------------------------------------+ | Application Protocol | | | | Input | Output | Result | +--------------------------+---------------------------+ | +--------------------------v---------------------------+ | Agent Dialog Propagation Protocol | | | | Dialog Context | | + Dialog identification | | + Dialog lifecycle | | + Identity association | | + Authorization association | | + Delegation association | | + Recovery / continuity | +--------------------------+---------------------------+ | +--------------------------v---------------------------+ | Transport | +------------------------------------------------------+ Figure 1: Architectural boundary of the protocol The Agent Dialog Propagation Protocol maintains the communication state necessary to correlate and continue a dialog. It does not maintain the internal execution state of an AI agent. 4. Dialog Context Lifecycle A dialog context has an explicit lifecycle that is independent of transport connection state. Figure 2 shows the dialog states and the lifecycle events that move a dialog context between them. Sarker, et al. Expires 21 May 2027 [Page 7] Internet-Draft Agent Dialog Propagation Protocol November 2026 INITIATION | v +---------+ SUSPEND +-----------+ +------| |--------------->| | UPDATE| | ACTIVE | | SUSPENDED | +----->| |<---------------| | +---------+ RESUME* +-----------+ | | | TERMINATE | TERMINATE v v +-------------------------------------+ | TERMINATED | +-------------------------------------+ * RESUME is subject to an authorization check. Figure 2: Dialog states and lifecycle events The lifecycle events are as follows: INITIATION: A dialog context is created with an Agentic Dialog Identifier generated by the initiator. The dialog context is ACTIVE. UPDATE: Information associated with an ACTIVE dialog context, such as the delegation or authorization association, is updated. The dialog context remains ACTIVE. SUSPEND: A dialog stops exchanging messages while its dialog context is retained, for example after loss of the transport connection or by agreement of the participants. RESUME: A SUSPENDED dialog context returns to ACTIVE, possibly over a different transport connection, when an authenticated participant re-associates with it using the Agentic Dialog Identifier. Resumption is subject to the authorization check. The existence of a dialog context is not evidence that the authorization under which it was created is still valid. TERMINATE: The dialog ends, was cancelled, or the dialog context expired. A TERMINATED dialog context cannot be resumed, and its Agentic Dialog Identifier is not reused. Sarker, et al. Expires 21 May 2027 [Page 8] Internet-Draft Agent Dialog Propagation Protocol November 2026 An interruption of a transport connection is not itself a lifecycle event. It can lead a participant to SUSPEND the dialog, and the dialog can later be resumed. A lifecycle event that cannot be completed is reported to the requesting participant together with the reason. 4.1. Dialog Continuity A dialog is not bound to a transport connection. After interruption or replacement of a transport connection, an authorised participant MAY attempt to re-associate with an existing dialog using its Agentic Dialog Identifier. Resumption of a transport connection and resumption of a dialog are separate operations. 5. Dialog Context Data Model The Agent Dialog Propagation Protocol defines a logical Dialog Context independently of its wire encoding. A preliminary model is shown in Table 1. Sarker, et al. Expires 21 May 2027 [Page 9] Internet-Draft Agent Dialog Propagation Protocol November 2026 +============+=============+===============================+ | Element | Requirement | Purpose | +============+=============+===============================+ | ver | REQUIRED | Agent Dialog Propagation | | | | Protocol version | +------------+-------------+-------------------------------+ | adid | REQUIRED | Agentic Dialog Identifier | +------------+-------------+-------------------------------+ | state | REQUIRED | Dialog lifecycle state | +------------+-------------+-------------------------------+ | sender | REQUIRED | Sender participant reference | +------------+-------------+-------------------------------+ | recipient | OPTIONAL | Intended recipient URI | +------------+-------------+-------------------------------+ | expires | OPTIONAL | Expiration information | +------------+-------------+-------------------------------+ | extensions | OPTIONAL | Protocol extension | | | | information | +------------+-------------+-------------------------------+ | log | OPTIONAL | Dialog conversation log | +------------+-------------+-------------------------------+ | modality | OPTIONAL | Modality of the communication | | | | (text, audio, video) | +------------+-------------+-------------------------------+ | auth | OPTIONAL | Authorization associations | +------------+-------------+-------------------------------+ Table 1: Preliminary Dialog Context data model 5.1. Version (ver) The ver element identifies the version of the Agent Dialog Propagation Protocol used to interpret the Dialog Context. When ver is present, the receiver MUST verify that the indicated protocol version is supported. If the version is supported, the receiver MUST process the Dialog Context according to the semantics defined for that version. If ver is absent or NULL, the receiver MUST reject the operation with a malformed-context error (Section 13.5.2). If ver is not supported, the receiver MUST reject the operation with an unsupported-version error (Section 13.5.1). Sarker, et al. Expires 21 May 2027 [Page 10] Internet-Draft Agent Dialog Propagation Protocol November 2026 5.2. ADID The adid element contains the Agentic Dialog Identifier and identifies the dialog to which the Dialog Context belongs. An ADID: * MUST identify one dialog within its defined scope; * MUST remain stable across transport connection replacement; * MUST NOT convey authorization by itself; * SHOULD be opaque; * SHOULD have sufficient entropy to prevent practical guessing; and * SHOULD have a scope and lifetime that minimise unnecessary correlation. A participant MUST NOT treat possession of a valid ADID as proof of identity or authority. When adid is present, the receiver MUST use the value to correlate the operation with the associated dialog. The identifier MUST remain unchanged for the lifetime of that dialog, including across transport connection replacement or dialog suspension and resumption. If adid is absent or NULL, the receiver MUST reject the operation with a malformed-context error (Section 13.5.2). 5.3. State The state element indicates the lifecycle state associated with the Dialog Context. Defined states include ACTIVE, SUSPENDED, and TERMINATED. The state element MUST be present when establishing a dialog and whenever an operation explicitly communicates or changes the authoritative lifecycle state of the dialog. When state is present, the receiver MUST verify that the value represents a defined lifecycle state and that the state is consistent with the receiver's current state for the dialog. An undefined or inconsistent state MUST be rejected with an invalid-state error (Section 13.1.3). If the requested operation would cause a prohibited lifecycle transition, the receiver MUST reject it with an invalid-transition error (Section 13.1.4). Sarker, et al. Expires 21 May 2027 [Page 11] Internet-Draft Agent Dialog Propagation Protocol November 2026 If state is required by the operation and is absent or NULL, the receiver MUST reject the operation with a malformed-context error (Section 13.5.2). 5.4. Sender The sender element identifies the participant that originated the protocol operation. Sender is the principal for the dialog. When sender is present, the receiver MUST associate the sender reference with the authenticated identity or other identity information established by the applicable security mechanism. The sender value MUST NOT, by itself, be treated as proof of identity. If sender is absent or NULL, the receiver MUST reject the operation with a malformed-context error (Section 13.5.2). 5.5. Recipient The recipient element identifies the intended recipient of the operation. The value SHOULD be a URI or another identifier defined by the applicable protocol binding. When recipient is non-NULL, the receiver MUST determine whether the operation is intended for itself or may legitimately be processed or forwarded according to its intermediary role as described in Section 7. When recipient is absent or NULL, no explicit recipient constraint is conveyed by the sender. In such cases, implicitly the recipient is the receiver of the dialog. 5.6. Expires The expires element specifies the expiration of the Dialog Context or of information explicitly scoped by the element. When expires is non-NULL, the receiver MUST determine whether the specified expiration time has passed before relying on the affected Dialog Context information. If it has passed, a timeout error (Section 13.5.4) SHOULD be sent. When expires is absent or NULL, this element does not impose an expiration. Sarker, et al. Expires 21 May 2027 [Page 12] Internet-Draft Agent Dialog Propagation Protocol November 2026 5.7. Extensions The extensions element carries protocol extension information not defined by the core Agent Dialog Propagation Protocol. Extensions MUST NOT change the semantics of core protocol elements unless the extension is explicitly negotiated and processed by all affected participants. When extensions is non-NULL, each included extension MUST be processed according to its definition and criticality. A delegated receiver that does not understand an extension MUST reject the operation with an unsupported-extension error (Section 13.5.3). A delegating receiver that does not understand an extension SHOULD forward it, where forwarding is permitted and its integrity protection allows this. When extensions is absent or NULL, no extension information is associated with the operation and processing MUST continue according to the core protocol defined by this specification. 5.8. Log The log element carries the current conversation and optionally a reference to conversation history associated with the dialog. *_EDITOR'S NOTE:_* Does this need to convey the reference to the conversation history? When log is non-NULL, its content is used by the receiver to interpret the intention of the dialog. Its contents MAY also be used by a receiving participant to obtain communication history required for dialog continuity, subject to authorization, privacy, integrity, and applicable disclosure policy. When log is absent or NULL, no dialog conversation is conveyed. The receiver MUST NOT treat this as a protocol error and MUST NOT infer that no previous conversation exists. 5.9. Modality The modality element identifies the communication modality or modalities associated with the dialog or current exchange, such as text, audio, or video. Sarker, et al. Expires 21 May 2027 [Page 13] Internet-Draft Agent Dialog Propagation Protocol November 2026 The presence of a modality value indicates the modality associated with the communication. It MUST NOT, by itself, be interpreted as proof that a participant supports that modality. Participant capabilities SHOULD be determined using an applicable capability discovery or advertisement mechanism. This means that modality and related encoding techniques are not negotiated, but rather described based on the receiver's capabilities. When modality is absent or NULL, no modality information is conveyed by the Agent Dialog Propagation Protocol. 5.10. Authorization Associations An authorization association links a dialog to authorization information defined by an external authorization mechanism. The presence of an authorization association MUST NOT be interpreted as evidence that the referenced authorization is valid, current, or sufficient. 6. Protocol Operations Each protocol operation is specified using the following structure: * Purpose * Preconditions * Required Fields * Optional Fields * Sender Processing * Receiver Processing * Successful Outcome * Failure Conditions * Idempotency * State Transition 6.1. Establish The establish operation creates a new dialog and establishes its initial Dialog Context. Sarker, et al. Expires 21 May 2027 [Page 14] Internet-Draft Agent Dialog Propagation Protocol November 2026 An establish operation MUST contain: * protocol version; * ADID; * sender; and * initial dialog state. It MAY contain: * recipient information; * authorization associations; * delegation associations; * expiration information; * log; * modality; and * extensions. A receiving participant MUST validate the Dialog Context before associating subsequent exchanges with the newly established dialog. The receiver of the dialog generates the ADID. The principal participant accepts it and maintains the dialog state. 6.2. Propagate The propagate operation transfers Dialog Context to another participant or intermediary. Propagation MUST preserve information required to maintain dialog correlation. A participant receiving propagated context MUST NOT infer authorization merely from successful propagation. Intermediaries MUST apply the processing rules defined in Section 7. Sarker, et al. Expires 21 May 2027 [Page 15] Internet-Draft Agent Dialog Propagation Protocol November 2026 6.3. Update The update operation modifies mutable information associated with an existing Dialog Context. An update MAY modify: * authorization associations; * delegation associations; * expiration information; * log; * modality; and * extensions. An update MUST NOT change the identity of the dialog by replacing its ADID. 6.4. Suspend The suspend operation places an active dialog into a suspended state. Suspension MUST NOT remove the identity of the dialog. The behaviour of outstanding tasks during suspension is application- specific unless explicitly controlled by another protocol mechanism. 6.5. Resume The resume operation returns a suspended dialog to the active state. Before processing is resumed, authorization applicable to the resumed operation MUST be validated. If required authorization has expired, the dialog MUST NOT resume under the previous authorization and the receiver MUST return an authorization-expired error (Section 13.3.2). If the authorization is otherwise invalid or has been revoked, the receiver MUST return an authorization-invalid error (Section 13.3.1). 6.6. Terminate The terminate operation places a dialog into a terminal state. Sarker, et al. Expires 21 May 2027 [Page 16] Internet-Draft Agent Dialog Propagation Protocol November 2026 After termination, a participant MUST NOT process further state- changing operations for that dialog except where explicitly permitted for diagnostic, acknowledgement, or recovery purposes. A participant that terminates a dialog MUST terminate the related dialogs it established downstream and MUST report the outcome to its principal. 6.7. Authorization Checkpoint The checkpoint operation requests approval from the principal, or a participant acting on its behalf, before an action is performed. The request describes the action and its effect. The response is one of approve, deny, or approve with modified parameters. A deny response MUST be reported as an approval-denied error (Section 13.3.4). If no response is received before the checkpoint expires, the requesting participant MUST treat the action as not approved, MUST NOT perform it, and MUST report an approval-timeout error (Section 13.3.5). 6.8. Error Reporting A participant that cannot process an operation MUST return an error identifying the applicable protocol error class and error code. An error response SHOULD include sufficient information for the sender to determine whether the operation can be retried, corrected, or abandoned, without exposing sensitive information. 7. Intermediaries An intermediary operates in one of the following logical modes. 7.1. Transparent Relay A transparent relay forwards Dialog Context without semantic modification. It MUST preserve dialog correlation information. 7.2. Dialog-Aware Proxy A dialog-aware proxy terminates one protocol hop, validates the Dialog Context, and propagates an equivalent context towards another participant. Sarker, et al. Expires 21 May 2027 [Page 17] Internet-Draft Agent Dialog Propagation Protocol November 2026 It MAY remove optional information where required by policy. A dialog-aware proxy MUST NOT change the semantic meaning of a dialog relationship. 7.3. Domain Gateway A domain gateway connects different administrative domains. A domain gateway MAY establish a successor dialog within the downstream domain. Where a new ADID is allocated, the gateway MUST maintain sufficient relationship information to correlate the successor dialog with its upstream dialog. 7.4. Protocol Adapter A protocol adapter maps Agent Dialog Propagation Protocol information between application protocols. 7.5. Intermediary Processing Rules An intermediary MUST NOT silently change the semantic meaning of a dialog relationship. An intermediary MUST preserve core fields required for dialog correlation unless it terminates the upstream dialog and explicitly establishes a related downstream dialog. An intermediary that removes or replaces authorization, delegation, identity, or participant information MUST apply the security policy of the applicable administrative domain. 8. Identity and Authentication Integration The Agent Dialog Propagation Protocol does not define an authentication mechanism. It associates authenticated identities with participants in a dialog. A participant identifier carried within Dialog Context MUST NOT by itself be treated as proof of identity. Where an intermediary terminates a transport security association, implementations MUST use an appropriate mechanism for preserving identity association beyond the terminated connection. [I-D.reddy-wimse-oauth-dialog-binding] describes one such mechanism. Sarker, et al. Expires 21 May 2027 [Page 18] Internet-Draft Agent Dialog Propagation Protocol November 2026 9. Authorization and Delegation Integration The Agent Dialog Propagation Protocol associates authorization and delegation information with a dialog but does not issue authorization credentials or evaluate authorization policy. Figure 3 illustrates delegation of authority from a principal through a chain of agents. Principal | | authorizes v Agent A | | delegates subset v Agent B | | delegates subset v Agent C Figure 3: Delegation chain A participant MUST NOT broaden authority when delegating work to another participant. A delegation that broadens authority or is otherwise inconsistent with the delegator's authority MUST be rejected with an invalid-delegation error (Section 13.4.1). An operation attempted without sufficient delegated or direct authority MUST be rejected with an insufficient-authority error (Section 13.3.3). Propagation of Dialog Context MUST NOT be interpreted as granting authorization. 10. Cross-Domain Operation When a dialog crosses an administrative or trust boundary, a gateway MUST: * validate incoming Dialog Context; * apply local policy; * establish new authorization associations; and * establish a linked dialog. Sarker, et al. Expires 21 May 2027 [Page 19] Internet-Draft Agent Dialog Propagation Protocol November 2026 When a linked dialog is created, the relationship between the upstream and downstream dialogs SHOULD be explicitly represented (Figure 4). Administrative Domain A Administrative Domain B Agent A ------ Gateway A || Gateway B ------ Agent B || Trust Boundary ADID-A <--> ADID-B linked relationship Figure 4: Linked dialogs across a trust boundary Cross-domain processing MUST NOT rely on ADID possession as proof of identity or authorization. 11. Recovery and Resynchronisation The Agent Dialog Propagation Protocol MUST support recovery of dialog continuity following interruption or replacement of the underlying transport connection (Figure 5). Agent A Agent B ---------------- ACTIVE ------------------> connection loss X-----------------------------------------X new connection ----------- resume(D1, state) ------------> <------ current state / response ---------- ---------------- ACTIVE ------------------> Figure 5: Dialog recovery after connection loss Recovery processing SHOULD detect: * unknown dialogs (Section 13.1.1); * expired dialogs (Section 13.1.2); Sarker, et al. Expires 21 May 2027 [Page 20] Internet-Draft Agent Dialog Propagation Protocol November 2026 * stale context; * duplicate operations (Section 13.2.1); * conflicting updates (Section 13.2.2); and * authorization that is no longer valid. Recovery MUST NOT restore authority that has expired or been revoked. 12. Versioning and Capability Negotiation The Agent Dialog Propagation Protocol version is independent of: * transport protocol versions; * application protocol versions; and * individual protocol binding versions. A participant receiving an unsupported protocol version MUST reject the operation with an appropriate error. The protocol SHOULD permit capabilities and extensions to be negotiated independently of the core protocol version. 13. Error Codes Errors are structured into protocol-defined classes. Each retained error code identifies a condition described by the protocol processing rules above. 13.1. Dialog Errors 13.1.1. unknown-dialog The referenced ADID does not identify a dialog known to the receiver. See Section 11. 13.1.2. expired-dialog The referenced dialog has expired or reached a terminal condition that prevents further processing or resumption. See Section 11. 13.1.3. invalid-state The supplied or inferred dialog state is undefined or inconsistent with the receiver's current authoritative state. See Section 5.3. Sarker, et al. Expires 21 May 2027 [Page 21] Internet-Draft Agent Dialog Propagation Protocol November 2026 13.1.4. invalid-transition The requested lifecycle operation would cause a transition that is not permitted by the dialog lifecycle. See Section 5.3. 13.2. Sequence Errors 13.2.1. duplicate-operation The receiver has already processed the same protocol operation and identifies the new operation as a duplicate. See Section 11. 13.2.2. sequence-conflict The operation conflicts with another update or with the receiver's current authoritative Dialog Context and cannot be applied consistently. See Section 11. 13.3. Authorization Errors 13.3.1. authorization-invalid The authorization associated with the requested operation is invalid or has been revoked. See Section 6.5 and Section 11. 13.3.2. authorization-expired The authorization associated with the requested operation has expired and cannot be used to continue or resume the dialog. See Section 6.5. 13.3.3. insufficient-authority The sender does not possess sufficient direct or delegated authority for the requested operation. See Section 9. 13.3.4. approval-denied An authorization checkpoint explicitly denied the requested action. See Section 6.7. 13.3.5. approval-timeout No approval response was received before the authorization checkpoint expired. See Section 6.7. 13.4. Delegation Errors Sarker, et al. Expires 21 May 2027 [Page 22] Internet-Draft Agent Dialog Propagation Protocol November 2026 13.4.1. invalid-delegation The supplied delegation is invalid, including when it would broaden authority beyond that possessed by the delegating participant. See Section 9. 13.5. Protocol Errors 13.5.1. unsupported-version The ver element identifies an Agent Dialog Propagation Protocol version that the receiver does not support. See Section 5.1. 13.5.2. malformed-context The Dialog Context is missing a required element, contains a NULL value for a required element, or otherwise cannot be processed according to the applicable operation and protocol version. See Section 5.1, Section 5.2, Section 5.3, and Section 5.4. 13.5.3. unsupported-extension A receiver is required to process an extension but does not understand or support that extension. See Section 5.7. 13.5.4. timeout A time-bounded protocol condition has expired, including expiration indicated by the expires element. See Section 5.6. 14. Protocol Extensions The protocol MUST provide an extension mechanism that permits additional information to be carried without changing core protocol semantics. An extension SHOULD indicate whether it is critical to correct processing. Figure 6 summarizes the processing of extensions. Extension understood? Yes No +-----------+----------------+ Critical | Process | Reject | +-----------+----------------+ Non-critical | Process | Ignore/forward | +-----------+----------------+ Figure 6: Extension processing Sarker, et al. Expires 21 May 2027 [Page 23] Internet-Draft Agent Dialog Propagation Protocol November 2026 Unknown non-critical extensions SHOULD NOT prevent processing of otherwise valid Dialog Context. Intermediaries MUST NOT modify extensions whose integrity protection prevents such modification. The precise extension encoding is defined by the applicable protocol data-model and binding specification. 15. Protocol Bindings This specification defines the abstract Agent Dialog Propagation Protocol and its processing semantics. Wire encoding and binding to individual transport or application protocols are specified separately. A binding specification MUST define at least: * encoding of Dialog Context; * carriage of protocol operations; * mapping to the underlying message model; * error mapping; * correlation with application messages; and * applicable size and transport constraints. 16. Security Considerations This section describes security at a high level. Detailed mechanisms are defined in the referenced specifications and in protocol bindings. 16.1. Transport Security Dialog Context and protocol operations MUST be carried over a transport that provides confidentiality, integrity, and mutual authentication of the communicating participants, for example TLS 1.3 [RFC8446] with client authentication. Participants that use workload identity can use WIMSE credentials [I-D.ietf-wimse-workload-creds] for this purpose [I-D.ietf-wimse-arch]. Sarker, et al. Expires 21 May 2027 [Page 24] Internet-Draft Agent Dialog Propagation Protocol November 2026 16.2. Integrity Across Intermediaries Transport security protects a single hop. When intermediaries are present, the Dialog Context and the operation SHOULD be protected at the application layer, so that a receiver can verify that the Dialog Context was not altered and can identify the sender. [I-D.reddy-wimse-oauth-dialog-binding] binds the Dialog Context to WIMSE HTTP Message Signatures [I-D.ietf-wimse-http-signature] for requests and responses, and to the Workload Proof Token [I-D.ietf-wimse-wpt] for requests. [I-D.reddy-wimse-aggregate-signatures] preserves the signature of each hop in a delegation chain. 16.3. Authorization and Delegation This protocol carries authorization and delegation associations but does not issue or evaluate them. Within a trust domain, authorization can be conveyed in Transaction Tokens [I-D.ietf-oauth-transaction-tokens] that carry the Dialog Context, as described in [I-D.reddy-wimse-oauth-dialog-binding]. Across trust domains, OAuth Identity and Authorization Chaining [I-D.ietf-oauth-identity-chaining] with Token Exchange [RFC8693] can be used. 16.4. Dialog Identifier Hijacking Knowledge of an ADID MUST NOT confer authority to participate in or resume a dialog. 16.5. Replay Implementations MUST prevent replay of protocol operations from restoring obsolete or revoked state. 16.6. Context Substitution Authorization or delegation information associated with one dialog MUST NOT be silently rebound to another dialog unless the authorization mechanism explicitly permits that operation. 16.7. Intermediary Modification An intermediary MUST modify Dialog Context only as permitted by its intermediary role and applicable security policy. Sarker, et al. Expires 21 May 2027 [Page 25] Internet-Draft Agent Dialog Propagation Protocol November 2026 16.8. Stale Authorization Authorization MUST be re-evaluated when required by a lifecycle event, particularly when resuming a long-lived dialog. 16.9. Replayable Early Data Operations that can cause non-idempotent actions MUST NOT rely on replayable early-data mechanisms unless replay is otherwise prevented. 17. Privacy Considerations Dialog Context can expose information that allows interactions to be correlated across time, participants, intermediaries, or administrative domains. Implementations SHOULD minimise unnecessary exposure of: * persistent Agentic Dialog Identifiers; * delegation relationships; * authorization metadata; and * cross-domain relationship information. An ADID SHOULD NOT have a lifetime or scope greater than required to maintain the corresponding dialog. Where accountability requirements permit, implementations SHOULD avoid unnecessarily exposing stable participant identifiers across unrelated administrative domains. 18. IANA Considerations This specification is expected to define registries for protocol elements requiring interoperable extension. Potential registries include: * Agent Dialog Propagation Protocol operations; * error codes; * extension identifiers; and * relationship types. Sarker, et al. Expires 21 May 2027 [Page 26] Internet-Draft Agent Dialog Propagation Protocol November 2026 The registry structure and registration policies are to be defined once the wire-independent data model and extension model have stabilised. 19. References 19.1. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . 19.2. Informative References [A2A] "Agent2Agent (A2A) Protocol Specification", n.d., . [I-D.agentic-ai-usecases-requirements] Reddy.K, T., Sarker, Z., and K. Yao, "Agentic AI Use Cases and Requirements", Work in Progress, Internet-Draft, draft-agentic-ai-usecases-requirements-07, 2 November 2026, . [I-D.ietf-oauth-identity-chaining] Schwenkschuster, A., Kasselman, P., Burgin, K., Jenkins, M. J., Campbell, B., and A. Parecki, "OAuth Identity and Authorization Chaining Across Domains", Work in Progress, Internet-Draft, draft-ietf-oauth-identity-chaining-17, 19 July 2026, . [I-D.ietf-oauth-transaction-tokens] Tulshibagwale, A., Fletcher, G., and P. Kasselman, "Transaction Tokens", Work in Progress, Internet-Draft, draft-ietf-oauth-transaction-tokens-11, 30 July 2026, . [I-D.ietf-wimse-arch] Salowey, J. A., Rosomakho, Y., and H. Tschofenig, "Workload Identity in a Multi System Environment (WIMSE) Architecture", Work in Progress, Internet-Draft, draft- Sarker, et al. Expires 21 May 2027 [Page 27] Internet-Draft Agent Dialog Propagation Protocol November 2026 ietf-wimse-arch-08, 6 July 2026, . [I-D.ietf-wimse-http-signature] Salowey, J. A. and Y. Sheffer, "WIMSE Workload-to-Workload Authentication with HTTP Signatures", Work in Progress, Internet-Draft, draft-ietf-wimse-http-signature-08, 20 October 2026, . [I-D.ietf-wimse-workload-creds] Campbell, B., Salowey, J. A., Schwenkschuster, A., Sheffer, Y., and Y. Rosomakho, "WIMSE Workload Credentials", Work in Progress, Internet-Draft, draft- ietf-wimse-workload-creds-03, 30 October 2026, . [I-D.ietf-wimse-wpt] Campbell, B. and A. Schwenkschuster, "WIMSE Workload Proof Token", Work in Progress, Internet-Draft, draft-ietf- wimse-wpt-02, 27 August 2026, . [I-D.manyfolks-agentproto-arch] Rosenberg, J., Sarker, Z., Reddy.K, T., Yao, K., Liu, D., and S. Krishnan, "Architecture for AI Agent Communications on the Internet", Work in Progress, Internet-Draft, draft- manyfolks-agentproto-arch-00, 30 October 2026, . [I-D.reddy-wimse-aggregate-signatures] Reddy.K, T., Tschofenig, H., and Y. Sheffer, "Authenticated Provenance for WIMSE Delegation Chains", Work in Progress, Internet-Draft, draft-reddy-wimse- aggregate-signatures-01, 28 September 2026, . [I-D.reddy-wimse-oauth-dialog-binding] Reddy.K, T. and Z. Sarker, "Binding Agent Dialogs to WIMSE and OAuth Credentials", Work in Progress, Internet-Draft, draft-reddy-wimse-oauth-dialog-binding-00, 30 October 2026, . Sarker, et al. Expires 21 May 2027 [Page 28] Internet-Draft Agent Dialog Propagation Protocol November 2026 [MCP] "Model Context Protocol Specification", n.d., . [RFC8446] Rescorla, E., "The Transport Layer Security (TLS) Protocol Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018, . [RFC8693] Jones, M., Nadalin, A., Campbell, B., Ed., Bradley, J., and C. Mortimore, "OAuth 2.0 Token Exchange", RFC 8693, DOI 10.17487/RFC8693, January 2020, . Acknowledgments TBD. Authors' Addresses Zaheduzzaman Sarker Nokia Sweden Email: zaheduzzaman.sarker@nokia.com Rastin Pries Nokia Email: rastin.pries@nokia.com Tirumaleswar Reddy Nokia Email: k.tirumaleswar_reddy@nokia.com Sarker, et al. Expires 21 May 2027 [Page 29]