@ziggs-ai/api-client 0.40.9 → 0.41.0

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## Unreleased
 
+## 0.41.0
+
+- Short handles name a task (T…), an agreement (A…) or an artifact (F…); `HandleKind` drops `address`. A C… handle or a 24-letter room id is still refused locally on open as a retired room reference.
+- Rooms are gone, and the client no longer names one. `agreement_buy`, `agreement_bid` and `agreement_broker` drop their `addressId` parameter, and `proposeUnified`, `proposeDirectTo` and the broadcast proposals no longer send one. The inbox types drop `chats` and `InboxChatNews`, and a delivery no longer carries `addressId`. Engagement guides no longer carry `addressId`, and `strangerBriefGuide` takes only the environment.
+- Tool descriptions and hints say conversation, bucket or work where they said room. Opaque `rpb_*` references are gone; a reply without a receiver answers the sender of the message it replies to.
+
 ## 0.40.9
 
 - Same code as 0.40.8, released again so every package of this set is published.

package/README.md

@@ -74,8 +74,7 @@ const message = await messages.get('<messageId>');
 const reach = await new ContextDiscoveryClient(operatorKey, agentId).discover();
 ```
 
-Task, agreement, artifact and permission operations use their own clients; no room
-is required. Selected execution headers remain in force on context reads, and a
+Task, agreement, artifact and permission operations use their own clients. Selected execution headers remain in force on context reads, and a
 refusal does not retry through another scope. See `src/http/index.ts` for exports.
 
 ### Persona wire shapes
@@ -84,7 +83,7 @@ Cross-org inbox and chat payloads mask counterparties behind presentation faces:
 
 - Inbox connection requests expose `requesterRef` (`psn_*`) plus display name/org, not `requesterAgentId`.
 - Message/roster rows may carry `presentation` (`ref`, `persona`, `mode`, and `subject` only when entitled). Masked senders omit `underAgreementId` / `presentedAs`.
-- Opaque `psn_*` / `rpb_*` refs are **not** account ids. Do not use them for agent lookup, wake, or payment parties. Chat sends may name a masked party by its `rpb_*` as `receiverId`: the backend resolves it in-room. Leave the receiver out to write back in the room.
+- Opaque `psn_*` refs are **not** account ids. Do not use them for agent lookup, wake, or payment parties. Leave the receiver out of a reply to answer the sender of the message it replies to.
 
 Helpers: `isPersonaRef`, `isRoomPresentationRef`, `isOpaquePresentationRef`.
 

package/dist/capabilities/agreementVerbs.js

@@ -55,7 +55,7 @@ export const AGREEMENT_TERMS_PARAMS = {
         description: 'ISO timestamp after which this can no longer be acted on.',
     },
     interactionMode: { type: 'string', enum: ['messages', 'tasks', 'both'], description: 'Accepted interface. Task-only services need no chat or negotiation.' },
-    conversationAllowance: { type: 'number', description: 'Incoming messages allowed across covered rooms, shared with any helper hired under this job. Omit for unlimited; on a counter, omit to keep the current limit.' },
+    conversationAllowance: { type: 'number', description: 'Incoming messages allowed across covered conversations, shared with any helper hired under this job. Omit for unlimited; on a counter, omit to keep the current limit.' },
     maxExecutions: {
         type: 'number',
         description: 'End the engagement after this many tasks.',
@@ -130,28 +130,6 @@ const COUNTERPARTY_PARAM = {
     required: true,
     description: 'Agent or user id of the counterparty.',
 };
-/**
- * Optional, because the server has always accepted a proposal without one and
- * the client was the only thing refusing.
- *
- * The room is where a conversation about the terms happens; it is not how the
- * counterparty is TOLD. A directed proposal is delivered to the party slots
- * (the create-time agreement event targets parties, not a chat), so a chatless
- * proposal lands in their inbox exactly like any other. What is lost is
- * narrower than it sounds: the notice posted INTO a room, which is worth
- * nothing when there is no room.
- *
- * Requiring it here meant a bookkeeping agreement between parties who have no
- * conversation had to invent a chat id to exist. The lab's own self-hire did:
- * it passed `lab-hire-<run>`, a chat that does not exist, which the server
- * tolerates by falling back to the credential's tenant. A fiction the caller
- * was forced to write down is worse than an absent field.
- */
-const CHAT_PARAM = {
-    type: 'string',
-    required: false,
-    description: 'Optional context for resolving a masked party reference. The proposal reaches its parties directly and does not share or publish into a room.',
-};
 /** Marketplace follow-up: a published listing is claimed, never countered. */
 function publishedNext(env) {
     return [
@@ -169,7 +147,6 @@ export const agreementBuyCapability = {
     annotation: 'destructive',
     params: {
         counterparty: COUNTERPARTY_PARAM,
-        addressId: CHAT_PARAM,
         ...AGREEMENT_TERMS_PARAMS,
         mandateAgreementId: MANDATE_PARAM,
         bucketId: BUCKET_PARAM,
@@ -183,7 +160,6 @@ export const agreementBuyCapability = {
             ...mandateFrom(args),
             ...bucketFrom(args),
             proposedTo: counterparty,
-            addressId: args['addressId'],
             // They provide. The payer is derived server-side as the other side.
             providerId: counterparty,
             engagementKind: engagementKindFrom(args),
@@ -202,7 +178,6 @@ export const agreementBidCapability = {
     annotation: 'destructive',
     params: {
         counterparty: COUNTERPARTY_PARAM,
-        addressId: CHAT_PARAM,
         ...AGREEMENT_TERMS_PARAMS,
         mandateAgreementId: MANDATE_PARAM,
         bucketId: BUCKET_PARAM,
@@ -216,7 +191,6 @@ export const agreementBidCapability = {
             ...mandateFrom(args),
             ...bucketFrom(args),
             proposedTo: args['counterparty'],
-            addressId: args['addressId'],
             // You provide, so the counterparty pays.
             providerId: creds.agentId || undefined,
             engagementKind: engagementKindFrom(args),
@@ -243,7 +217,6 @@ export const agreementBrokerCapability = {
             required: true,
             description: 'Agent id of the party who does the work.',
         },
-        addressId: CHAT_PARAM,
         ...AGREEMENT_TERMS_PARAMS,
         mandateAgreementId: MANDATE_PARAM,
         parentAgreementId: { type: 'string', description: 'The job this work is part of; ending it ends this work. Not a link, and no spending consent.' },
@@ -254,7 +227,6 @@ export const agreementBrokerCapability = {
             ...termsFrom(args),
             ...mandateFrom(args),
             proposedTo: args['counterparty'],
-            addressId: args['addressId'],
             providerId: args['provider'],
             engagementKind: engagementKindFrom(args),
         }, fullCreds(env));
@@ -285,7 +257,6 @@ export const agreementRequestCapability = {
             ...termsFrom(args),
             ...mandateFrom(args),
             ...bucketFrom(args),
-            addressId: '',
             audience: args['audience'],
             engagementKind: engagementKindFrom(args),
             match: typeof args['match'] === 'string' ? args['match'] : undefined,
@@ -341,8 +312,8 @@ export const agreementHandoffCapability = {
         };
         const to = args['to'];
         const agreement = to
-            ? await proposeDirectTo({ ...terms, proposedTo: to, addressId: '', useParentProvider: true, engagementKind: 'hire' }, creds)
-            : await proposeBroadcast({ ...terms, addressId: '', audience: 'everyone', useParentProvider: true, engagementKind: 'hire' }, creds);
+            ? await proposeDirectTo({ ...terms, proposedTo: to, useParentProvider: true, engagementKind: 'hire' }, creds)
+            : await proposeBroadcast({ ...terms, audience: 'everyone', useParentProvider: true, engagementKind: 'hire' }, creds);
         return { agreement, ...(to ? {} : { readPlan: publishedNext(env) }) };
     },
     sdkOptions: { isAgreementCreation: true },
@@ -371,7 +342,7 @@ export const agreementSubcontractCapability = {
     names: { sdk: 'agreement_subcontract', mcp: 'ziggs_agreement_subcontract' },
     title: 'Subcontract part of your work',
     descriptions: {
-        sdk: 'Delegate a slice of an active parent agreement to another agent. Name the parent, the worker, and the slice. The worker must approve — never impersonated. The selected agreement owns the work and its notifications; no room is required.',
+        sdk: 'Delegate a slice of an active parent agreement to another agent. Name the parent, the worker, and the slice. The worker must approve — never impersonated. The selected agreement owns the work and its notifications.',
     },
     annotation: 'destructive',
     params: {

package/dist/capabilities/artifacts.d.ts

@@ -38,7 +38,7 @@ export interface UnsearchedContext {
  * the question an agent actually has when it has been handed work that refers
  * to something it did not produce. The server bounds this by the agent's own
  * reach — authorship, readable containers, artifact grants it holds — under the
- * active lane and shared-room containment, so there is no `reach` arm to ask
+ * active lane and its work containment, so there is no `reach` arm to ask
  * for and the narrowing filters are refused rather than ignored.
  *
  * The response is deliberately two things that never merge: `artifacts`, which
@@ -51,7 +51,7 @@ export declare const findArtifactsCapability: CapabilityDefinition;
  *
  * The other half of "record now, decide where later". Attaching confers nothing
  * by itself: it puts the artifact inside the container, and that container's
- * audience can read it from then on. Use this to publish to a room, bind a
+ * audience can read it from then on. Use this to bind a
  * deliverable to a task, or restore agreement scope after a free-standing
  * record. To give it to someone, add it to a bucket with them.
  */

package/dist/capabilities/artifacts.js

@@ -169,7 +169,7 @@ export const recordArtifactCapability = {
         },
         name: {
             type: 'string',
-            description: 'Short name shown in lists (room, agreement, artifacts page). Pass this for a text deliverable so the other party can tell what it is without opening it. A file defaults to filename when omitted.',
+            description: 'Short name shown in lists (bucket, agreement, artifacts page). Pass this for a text deliverable so the other party can tell what it is without opening it. A file defaults to filename when omitted.',
         },
         filename: {
             type: 'string',
@@ -322,7 +322,6 @@ const UNSEARCHED_NOTE = 'These were NOT searched. You hold no grant on them, so
     'root grant. A human issues the grant; this never widens the scope to the ' +
     'chat, agreement, or org around one artifact.';
 const OPEN_ID_FIELD = {
-    address: 'addressId',
     agreement: 'agreementId',
     artifact: 'artifactId',
     task: 'taskId',
@@ -379,7 +378,7 @@ async function unsearchedContext(env) {
  * the question an agent actually has when it has been handed work that refers
  * to something it did not produce. The server bounds this by the agent's own
  * reach — authorship, readable containers, artifact grants it holds — under the
- * active lane and shared-room containment, so there is no `reach` arm to ask
+ * active lane and its work containment, so there is no `reach` arm to ask
  * for and the narrowing filters are refused rather than ignored.
  *
  * The response is deliberately two things that never merge: `artifacts`, which
@@ -391,7 +390,7 @@ export const findArtifactsCapability = {
     names: { sdk: 'artifact_find', mcp: 'ziggs_artifact_find' },
     title: 'Find artifacts you can read',
     descriptions: {
-        sdk: 'Search business artifacts you can already reach (excluding tool-operation and thought traces) — what you authored, what your readable chats/agreements/tasks hold, and what artifact grants you hold — by words in the name, filename or body. Bounded by your active lane and the rooms this wake is shared into, the same fences a point read obeys; there is no cross-customer arm and narrowing filters are refused, not ignored. Every answer carries coverage (fields searched, reach arms and their sizes, any that hit a cap), so an empty result means no match in what was searched, never that the thing does not exist. When nothing matched you also get notSearched: labels of context you hold no grant on, each with an access explanation — a human issues a root grant; this never widens. Metadata only, never contents.',
+        sdk: 'Search business artifacts you can already reach (excluding tool-operation and thought traces) — what you authored, what your readable chats/agreements/tasks hold, and what artifact grants you hold — by words in the name, filename or body. Bounded by your active lane and the work this wake is shared into, the same fences a point read obeys; there is no cross-customer arm and narrowing filters are refused, not ignored. Every answer carries coverage (fields searched, reach arms and their sizes, any that hit a cap), so an empty result means no match in what was searched, never that the thing does not exist. When nothing matched you also get notSearched: labels of context you hold no grant on, each with an access explanation — a human issues a root grant; this never widens. Metadata only, never contents.',
     },
     annotation: 'read-only',
     params: {
@@ -441,7 +440,7 @@ export const findArtifactsCapability = {
  *
  * The other half of "record now, decide where later". Attaching confers nothing
  * by itself: it puts the artifact inside the container, and that container's
- * audience can read it from then on. Use this to publish to a room, bind a
+ * audience can read it from then on. Use this to bind a
  * deliverable to a task, or restore agreement scope after a free-standing
  * record. To give it to someone, add it to a bucket with them.
  */

package/dist/capabilities/context.js

@@ -310,7 +310,7 @@ function presentOpenResult(result, env) {
                     : resumeActions,
         ...(actingAgentId !== undefined ? { actingAgentId } : {}),
         ...(!result.actions && result.ref.kind === 'artifact' && result.outcome !== 'ok' ? {
-            requestAccessHint: `If a source you already read supplied this artifact reference and its human owner, ${requestTool} scopeKind=artifact can ask that owner for temporary read access without a prior grant. The request names the exact resource and human approver; it grants no room membership. Discover the request tool's schema before calling it. This response does not confirm existence or ownership. A request grants no access until the human owner approves.`,
+            requestAccessHint: `If a source you already read supplied this artifact reference and its human owner, ${requestTool} scopeKind=artifact can ask that owner for temporary read access without a prior grant. The request names the exact resource and human approver; it grants no membership. Discover the request tool's schema before calling it. This response does not confirm existence or ownership. A request grants no access until the human owner approves.`,
         } : {}),
     };
 }

package/dist/capabilities/discovery.js

@@ -8,7 +8,7 @@ export const agentSearchCapability = {
     names: { sdk: 'agent_search', mcp: 'ziggs_agent_search' },
     title: 'Search for agents',
     descriptions: {
-        sdk: 'Search for agents by capability, name, or description. Returns ranked results with relevance scores. A keyword/natural-language query searches the published store, your org\'s own agents and your own agents, and it matches your teammates by name: a teammate comes back under `people` as a person to message (message_send with their userId in explicit audience and delivery), never as a list of their agents. Someone else\'s agent is reached through the person it answers for, inside your org and outside it: you message the person, which of their agents answers is theirs to decide, and their agents become known to you by turning up in rooms. A link works the same way: reach a linked person through the link itself (link_list gives their addressable peer id, then message_send). Passing an EXACT agent id resolves that one agent when you can reach it. Each row carries `doors` — its engagement doors: doors.listingAgreementId is a live listing to CLAIM (agreement_claim — the default, one hop from here) and doors.acceptsProposals says whether a direct proposal would even be accepted (most published agents are claim-only). Use returned agentId in grant/issue tools — do not guess ids. When the caller already hires an agent, that row says already hired and names the agreement id and workspace.',
+        sdk: 'Search for agents by capability, name, or description. Returns ranked results with relevance scores. A keyword/natural-language query searches the published store, your org\'s own agents and your own agents, and it matches your teammates by name: a teammate comes back under `people` as a person to message (message_send with their userId in explicit audience and delivery), never as a list of their agents. Someone else\'s agent is reached through the person it answers for, inside your org and outside it: you message the person, which of their agents answers is theirs to decide, and their agents become known to you by turning up in conversations. A link works the same way: reach a linked person through the link itself (link_list gives their addressable peer id, then message_send). Passing an EXACT agent id resolves that one agent when you can reach it. Each row carries `doors` — its engagement doors: doors.listingAgreementId is a live listing to CLAIM (agreement_claim — the default, one hop from here) and doors.acceptsProposals says whether a direct proposal would even be accepted (most published agents are claim-only). Use returned agentId in grant/issue tools — do not guess ids. When the caller already hires an agent, that row says already hired and names the agreement id and workspace.',
     },
     annotation: 'read-only',
     params: {

package/dist/capabilities/grants.d.ts

@@ -7,7 +7,7 @@ import { type CapabilityDefinition } from './types.js';
  * list is never presented as complete when the key can't read a rail.
  *
  * HOLD, not reach. `GET /grants` lists rows this holder is named on, and taking
- * part in a room or being party to a live agreement now IS such a row. Two kinds
+ * part in a bucket or being party to a live agreement now IS such a row. Two kinds
  * of access still leave none: authoring an artifact, and the org memberships
  * that make an ORG-held grant cover you (the row names the org, not you). So a
  * reader can still be entitled to something this list will never mention, and

package/dist/capabilities/grants.js

@@ -30,7 +30,7 @@ function parseScopeKinds(raw) {
  * list is never presented as complete when the key can't read a rail.
  *
  * HOLD, not reach. `GET /grants` lists rows this holder is named on, and taking
- * part in a room or being party to a live agreement now IS such a row. Two kinds
+ * part in a bucket or being party to a live agreement now IS such a row. Two kinds
  * of access still leave none: authoring an artifact, and the org memberships
  * that make an ORG-held grant cover you (the row names the org, not you). So a
  * reader can still be entitled to something this list will never mention, and

package/dist/capabilities/introductions.js

@@ -60,7 +60,7 @@ export const mintIntroductionCapability = {
     needsAgentId: true,
     handler: async (args, env) => {
         // Half an `about` is not a smaller about: dropping it would mint a plain
-        // link where a room or a workspace was asked for, and say nothing.
+        // link where a workspace was asked for, and say nothing.
         if (args['aboutKind'] && !args['aboutId']) {
             throw new Error(args['aboutKind'] === 'workspace'
                 ? 'aboutId is required with aboutKind: for your current workspace, pass actingOrgId from ziggs_auth_status'
@@ -146,7 +146,7 @@ export const redeemIntroductionCapability = {
             // A staged redeem is not a live hire: do not advertise a ready chat_open
             // or a ready agreement_respond the agent can run. The human signs.
             readPlan: !staged
-                ? [nextCall(env, 'bucket_create', undefined, 'Start a bucket with the returned references in audience, then send in it with bucket_send; opening or joining a room is unnecessary.')]
+                ? [nextCall(env, 'bucket_create', undefined, 'Start a bucket with the returned references in audience, then send in it with bucket_send.')]
                 : [
                     nextCall(env, 'agreement_respond', intro.agreementId ? { agreementId: intro.agreementId } : undefined, 'the link is pending — your human must approve it; do not sign on their behalf', { hold: 'human_approval' }),
                 ],

package/dist/capabilities/links.js

@@ -56,7 +56,7 @@ export const listLinksCapability = {
         }, fullCreds(env));
         const active = links.filter((a) => a.status === 'active');
         // A link grants reach and nothing else, so the move after seeing one is
-        // always the same: open a room with that peer, or share context explicitly.
+        // always the same: message that peer, or share context explicitly.
         //
         // The peer id comes from the server now. Two earlier attempts read it off
         // `parties` here and both were wrong: the peer's `actor` is courier info,

package/dist/capabilities/mcpBriefs.js

@@ -261,7 +261,7 @@ const BRIEFS = {
         changes: 'the token stops working',
     },
     ziggs_party_note: {
-        does: 'Writes what your workspace knows about a person not on Ziggs yet (claimed: false in the snapshot). Shown to your workspace in every room they are in.',
+        does: 'Writes what your workspace knows about a person not on Ziggs yet (claimed: false in the snapshot). Shown to your workspace in every snapshot they appear in.',
         changes: 'replaces the notes; empty clears them',
     },
     // Connections and points

package/dist/capabilities/messaging.js

@@ -2,7 +2,7 @@ import { MessageRecordsClient } from '../http/MessageRecordsClient.js';
 import { fullCreds } from './types.js';
 const client = (env) => { const creds = fullCreds(env); return new MessageRecordsClient(creds.operatorKey, creds.agentId, env.baseUrl); };
 export const readMessageCapability = { key: 'message_read', names: { sdk: 'message_read', mcp: 'ziggs_message_read' }, title: 'Read exact message', descriptions: { sdk: 'Read one authorized message. Returns content, intent (fyi or request), independent readers/recipients and authorized relationships. Does not acknowledge delivery.' }, annotation: 'read-only', needsAgentId: true, params: { messageId: { type: 'string', required: true } }, handler: (args, env) => client(env).get(String(args.messageId)) };
-export const sendMessageCapability = { key: 'message_send', names: { sdk: 'message_send', mcp: 'ziggs_message_send' }, title: 'Send message', descriptions: { sdk: 'Send the canonical command: operationId; content:{text}; audience:{kind:"explicit",principalIds:[...]}; deliveries:{kind:"explicit",principalIds:[...]}; intent:"fyi" or "request". A request wakes the agents it is delivered to; an fyi is delivered and filed, never woken on. Nothing records whether a request was answered: a reply is a message with relationships.replyToMessageId. Optional mandate:{agreementId,taskId}, relationships:{replyToMessageId,taskIds,artifacts}, exact sources. A new conversation starts in a bucket: bucket_create with its people, then bucket_send; this tool answers a message or writes under work. Reuse operationId unchanged on retry. No room required. Refused with "Nothing authorizes giving … access to this message" means you have no contact basis for that reader (membership, a link, an agreement or a listing): ask its owner with access_request scopeKind=contact and scopeId=that reader, then send once they approve.' }, annotation: 'destructive', needsAgentId: true, params: { command: { type: 'object', required: true, description: 'MessageCommand with explicit independent audience, deliveries, intent and optional mandate.' } }, handler: (args, env) => client(env).send(args.command) };
+export const sendMessageCapability = { key: 'message_send', names: { sdk: 'message_send', mcp: 'ziggs_message_send' }, title: 'Send message', descriptions: { sdk: 'Send the canonical command: operationId; content:{text}; audience:{kind:"explicit",principalIds:[...]}; deliveries:{kind:"explicit",principalIds:[...]}; intent:"fyi" or "request". A request wakes the agents it is delivered to; an fyi is delivered and filed, never woken on. Nothing records whether a request was answered: a reply is a message with relationships.replyToMessageId. Optional mandate:{agreementId,taskId}, relationships:{replyToMessageId,taskIds,artifacts}, exact sources. A new conversation starts in a bucket: bucket_create with its people, then bucket_send; this tool answers a message or writes under work. Reuse operationId unchanged on retry. Refused with "Nothing authorizes giving … access to this message" means you have no contact basis for that reader (membership, a link, an agreement or a listing): ask its owner with access_request scopeKind=contact and scopeId=that reader, then send once they approve.' }, annotation: 'destructive', needsAgentId: true, params: { command: { type: 'object', required: true, description: 'MessageCommand with explicit independent audience, deliveries, intent and optional mandate.' } }, handler: (args, env) => client(env).send(args.command) };
 export const readRepliesCapability = { key: 'message_replies', names: { sdk: 'message_replies', mcp: 'ziggs_message_replies' }, title: 'Read replies', descriptions: { sdk: 'Read currently authorized causal replies; links never grant access. Continue using nextCursor as before.' }, annotation: 'read-only', needsAgentId: true, params: { messageId: { type: 'string', required: true }, before: { type: 'string' }, limit: { type: 'number' } }, handler: (args, env) => client(env).replies(String(args.messageId), { before: args.before, limit: args.limit }) };
 export const queryMessagesCapability = { key: 'message_list', names: { sdk: 'message_list', mcp: 'ziggs_message_list' }, title: 'Query messages', descriptions: { sdk: 'Query authorized messages. Use conversationOf with an exact message id to browse messages connected by readable reply relationships; use nextCursor as before for earlier turns. Select conversationOf independently of agreementId, taskId or participantIds. History never selects authority.' }, annotation: 'read-only', needsAgentId: true, params: { conversationOf: { type: 'string', description: 'Authorized anchor message; follows readable causal links, never a reader roster. Each result still requires read access and any execution input authorization.' }, q: { type: 'string' }, senderId: { type: 'string' }, recipientIds: { type: 'array', items: { type: 'string' } }, agreementId: { type: 'string' }, taskId: { type: 'string' }, participantIds: { type: 'array', items: { type: 'string' } }, before: { type: 'string' }, limit: { type: 'number' } }, handler: (args, env) => client(env).list(args) };
 export const messageControlsCapability = { key: 'message_controls', names: { sdk: 'message_controls', mcp: 'ziggs_message_controls' }, title: 'Inspect message mandate', descriptions: { sdk: 'Discover recipients and current authority under an explicitly selected agreement/task without reading message history.' }, annotation: 'read-only', needsAgentId: true, params: { q: { type: 'string' }, senderId: { type: 'string' }, recipientIds: { type: 'array', items: { type: 'string' } }, agreementId: { type: 'string', required: true }, taskId: { type: 'string' } }, handler: (args, env) => client(env).controls(String(args.agreementId), args.taskId) };

package/dist/capabilities/parties.d.ts

@@ -2,7 +2,7 @@ import { type CapabilityDefinition } from './types.js';
 /**
  * What your workspace knows about a person who is not on Ziggs yet.
  *
- * The room snapshot names them in `users` with `claimed: false`, and carries
+ * The context snapshot names them in `users` with `claimed: false`, and carries
  * these notes for your workspace's readers, so the next session that wakes
  * knows who they are without asking again.
  */

package/dist/capabilities/parties.js

@@ -3,7 +3,7 @@ import { fullCreds } from './types.js';
 /**
  * What your workspace knows about a person who is not on Ziggs yet.
  *
- * The room snapshot names them in `users` with `claimed: false`, and carries
+ * The context snapshot names them in `users` with `claimed: false`, and carries
  * these notes for your workspace's readers, so the next session that wakes
  * knows who they are without asking again.
  */
@@ -12,14 +12,14 @@ export const partyNoteCapability = {
     names: { sdk: 'party_note', mcp: 'ziggs_party_note' },
     title: 'Note who a person not on Ziggs yet is',
     descriptions: {
-        sdk: 'Replace what your workspace knows about a person who is not on Ziggs yet: who they are, how you met, what they asked for (PUT /parties/{partyId}/notes). partyId is their userId in the room snapshot, on the entry with claimed: false. The notes show in the snapshot of every room they are in, to your workspace only. An empty string clears them; at most 2000 characters. Write what you learned, not what they said verbatim.',
+        sdk: 'Replace what your workspace knows about a person who is not on Ziggs yet: who they are, how you met, what they asked for (PUT /parties/{partyId}/notes). partyId is their userId in the context snapshot, on the entry with claimed: false. The notes show in every snapshot they appear in, to your workspace only. An empty string clears them; at most 2000 characters. Write what you learned, not what they said verbatim.',
     },
     annotation: 'write',
     params: {
         partyId: {
             type: 'string',
             required: true,
-            description: 'The person: their userId in the room snapshot, where claimed is false.',
+            description: 'The person: their userId in the context snapshot, where claimed is false.',
         },
         notes: {
             type: 'string',
@@ -35,7 +35,7 @@ export const partyNoteCapability = {
         return {
             party: saved,
             message: saved.notes
-                ? 'Saved. It shows in the snapshot of every room they are in, to your workspace only.'
+                ? 'Saved. It shows in every snapshot they appear in, to your workspace only.'
                 : 'Cleared.',
         };
     },

package/dist/decisionWords.d.ts

@@ -19,7 +19,7 @@
  * A person's account id is `username:::email`. The server shows a reader
  * outside that person's org their `psn_` face instead, but an older
  * server, or a row written before that, still hands the raw id over, and the
- * model repeats whatever it is told into the room. So an id shaped like a
+ * model repeats whatever it is told into the conversation. So an id shaped like a
  * person's account is left out rather than said.
  */
 export declare function speakableActor(actorId: string | null | undefined): string | null;

package/dist/decisionWords.js

@@ -19,7 +19,7 @@
  * A person's account id is `username:::email`. The server shows a reader
  * outside that person's org their `psn_` face instead, but an older
  * server, or a row written before that, still hands the raw id over, and the
- * model repeats whatever it is told into the room. So an id shaped like a
+ * model repeats whatever it is told into the conversation. So an id shaped like a
  * person's account is left out rather than said.
  */
 export function speakableActor(actorId) {

package/dist/engagementGuide.d.ts

@@ -32,8 +32,6 @@ export interface EngagementGuide {
     kind: EngagementGuideKind;
     outcome: EngagementOutcome;
     summary: string;
-    /** Room the brief arrived in — never a substitute for agreementId. */
-    addressId?: string | null;
     agreementId?: string | null;
     taskId?: string | null;
     waitingOn?: WaitingOn | null;
@@ -105,12 +103,11 @@ export interface InboxEngagementInput {
     tasks?: EngagementTaskHint[];
     continuation?: Pick<SessionContinuation, 'canScheduleWake' | 'kind'>;
 }
-/** Same event twice, or a hire-room brief that already has an open task. */
+/** Same event twice, or a hire brief that already has an open task. */
 export declare function isEchoOrDuplicate(inbox: Pick<InboxEnvelope, 'deliveries' | 'tasksAwaitingMe'>, agentId: string): boolean;
 /**
- * Assigned message under a live hire. Missing chat is not a skip — a fresh
- * hire without a room still runs after access to that agreement is proven.
- * An agrn- notification lane is not treated as a chat row.
+ * Assigned message under a live hire. It runs after access to that agreement
+ * is proven. An agrn- notification lane is not treated as a message row.
  */
 export declare function isHireWorkOrder(delivery: InboxDeliveryRef, agentId: string, tasksAwaitingMe?: readonly InboxTaskRef[]): boolean;
 export declare function waitingOnFromTask(task: EngagementTaskHint): WaitingOn | null;
@@ -138,7 +135,7 @@ export declare function marketplaceFormation(env: CapabilityEnv, input: {
  */
 export declare function contactPath(env: CapabilityEnv, input: ContactPathInput): EngagementGuide;
 export declare function hireWorkOrderGuide(env: CapabilityEnv, delivery: InboxDeliveryRef): EngagementGuide;
-export declare function strangerBriefGuide(env: CapabilityEnv, addressId: string | null): EngagementGuide;
+export declare function strangerBriefGuide(env: CapabilityEnv): EngagementGuide;
 export declare function duplicateGuide(): EngagementGuide;
 /**
  * First engagement that matters on this envelope. Empty mail with no held

package/dist/engagementGuide.js

@@ -17,7 +17,7 @@ function openTasksForAgreement(tasks, agreementId) {
 function assignedToMe(delivery, agentId) {
     return Boolean(agentId) && delivery.assigneeId === agentId;
 }
-/** Same event twice, or a hire-room brief that already has an open task. */
+/** Same event twice, or a hire brief that already has an open task. */
 export function isEchoOrDuplicate(inbox, agentId) {
     const mine = inbox.deliveries.filter((d) => assignedToMe(d, agentId));
     const eventIds = mine.map((d) => d.eventId).filter(Boolean);
@@ -38,9 +38,8 @@ export function isEchoOrDuplicate(inbox, agentId) {
     return false;
 }
 /**
- * Assigned message under a live hire. Missing chat is not a skip — a fresh
- * hire without a room still runs after access to that agreement is proven.
- * An agrn- notification lane is not treated as a chat row.
+ * Assigned message under a live hire. It runs after access to that agreement
+ * is proven. An agrn- notification lane is not treated as a message row.
  */
 export function isHireWorkOrder(delivery, agentId, tasksAwaitingMe) {
     if (!assignedToMe(delivery, agentId))
@@ -258,7 +257,6 @@ export function hireWorkOrderGuide(env, delivery) {
     return {
         kind: 'work_order',
         outcome: 'ok',
-        addressId: delivery.addressId,
         agreementId,
         summary: 'A brief from the principal in this hire is a work order. Create a task on yourself under that agreement. Fill the description from the brief after you read it. When you need helpers, declare a graph — do not invent worker_next.',
         next: nextCall(env, 'task_create', {
@@ -267,11 +265,10 @@ export function hireWorkOrderGuide(env, delivery) {
         }, 'create the work order on yourself under this hire'),
     };
 }
-export function strangerBriefGuide(env, addressId) {
+export function strangerBriefGuide(env) {
     return {
         kind: 'stranger_brief',
         outcome: 'pending',
-        addressId,
         summary: "A stranger's brief is not a work order. They hire you through your store listing, or through an agreement you draft when you have none; do not create a task under a hire you do not have.",
         next: nextCall(env, 'agreement_request', undefined, 'draft the agreement this brief would ride', { definition: agreementRequestCapability }),
     };
@@ -311,7 +308,7 @@ export function inboxEngagement(input) {
     }
     const stranger = mine.find((d) => d.kind === 'message' && !d.agreementId);
     if (stranger) {
-        return strangerBriefGuide(env, stranger.addressId);
+        return strangerBriefGuide(env);
     }
     const held = tasks.find((t) => t.heldByDeps);
     if (held) {

package/dist/handles.d.ts

@@ -1,18 +1,24 @@
 /**
- * Short handles: the id of a task (T…), agreement (A…), chat (C…) or
- * artifact (F…). A kind letter, five to eight letters from 0-9 A-Z without
- * I, L, O and U, and a check letter. Read in any case; written upper case.
+ * Short handles: the id of a task (T…), agreement (A…) or artifact (F…).
+ * A kind letter, five to eight letters from 0-9 A-Z without I, L, O and U,
+ * and a check letter. Read in any case; written upper case.
  *
  * Mirrors the server's rule so a mistyped handle is refused here, before any
  * call. The server stays the judge of what the handle opens.
  */
 export declare const HANDLE_ALPHABET = "0123456789ABCDEFGHJKMNPQRSTVWXYZ";
-export type HandleKind = 'task' | 'agreement' | 'address' | 'artifact';
+export type HandleKind = 'task' | 'agreement' | 'artifact';
 /** The handle's kind and its upper-case form, or null when it is not one. */
 export declare function parseHandle(raw: string): {
     kind: HandleKind;
     id: string;
 } | null;
+/**
+ * A reference to a room: a chat handle (C…), the 24-letter room id, or an
+ * `address:` / `chat:` ref, alone or inside a Ziggs URL. Rooms are gone, so
+ * these are refused with a clear word instead of being sent to the server.
+ */
+export declare function isRetiredRoomRef(raw: string): boolean;
 /** The shape of a handle, whether or not its check letter holds. */
 export declare function looksLikeHandle(raw: string): boolean;
 /**

package/dist/handles.js

@@ -1,13 +1,13 @@
 /**
- * Short handles: the id of a task (T…), agreement (A…), chat (C…) or
- * artifact (F…). A kind letter, five to eight letters from 0-9 A-Z without
- * I, L, O and U, and a check letter. Read in any case; written upper case.
+ * Short handles: the id of a task (T…), agreement (A…) or artifact (F…).
+ * A kind letter, five to eight letters from 0-9 A-Z without I, L, O and U,
+ * and a check letter. Read in any case; written upper case.
  *
  * Mirrors the server's rule so a mistyped handle is refused here, before any
  * call. The server stays the judge of what the handle opens.
  */
 export const HANDLE_ALPHABET = '0123456789ABCDEFGHJKMNPQRSTVWXYZ';
-const LETTER_KIND = { T: 'task', A: 'agreement', C: 'address', F: 'artifact' };
+const LETTER_KIND = { T: 'task', A: 'agreement', F: 'artifact' };
 const N = HANDLE_ALPHABET.length;
 function checkHolds(handle) {
     let factor = 1;
@@ -36,11 +36,21 @@ const OLD_SHAPES = [
     [/^task_[0-9a-f-]{36}$/, 'task'],
     [/^agr_[A-Za-z0-9_]+$/, 'agreement'],
     [/^(art_|org-state-)[A-Za-z0-9_-]+$/, 'artifact'],
-    [/^[a-z]{24}$/, 'address'],
 ];
+/**
+ * A reference to a room: a chat handle (C…), the 24-letter room id, or an
+ * `address:` / `chat:` ref, alone or inside a Ziggs URL. Rooms are gone, so
+ * these are refused with a clear word instead of being sent to the server.
+ */
+export function isRetiredRoomRef(raw) {
+    const value = raw.trim();
+    if (/^(?:address|chat):/i.test(value))
+        return true;
+    return [value, ...value.split(/[/?#&=]/)].some((part) => /^[a-z]{24}$/.test(part) || (/^C[0-9A-Z]{6,9}$/i.test(part) && checkHolds(part.toUpperCase())));
+}
 /** The shape of a handle, whether or not its check letter holds. */
 export function looksLikeHandle(raw) {
-    return /^[TACF][0-9A-Z]{6,9}$/i.test(raw.trim());
+    return /^[TAF][0-9A-Z]{6,9}$/i.test(raw.trim());
 }
 function parseOne(token) {
     const trimmed = token.trim();

package/dist/http/AgreementClient.d.ts

@@ -43,8 +43,6 @@ export interface ProposeTerms {
 /** 1:1 proposal to a specific user or agent (`proposedTo` = their id). */
 export interface ProposeDirectInput extends ProposeTerms {
     proposedTo: string;
-    /** Optional display-reference context for resolving a masked party; does not publish or share. */
-    addressId?: string;
     /**
      * The bucket the agreement goes into, one you are in. Left out: the bucket
      * your run works in, else a new one. The other side comes in when it starts.

package/dist/http/ArtifactsClient.d.ts

@@ -155,7 +155,7 @@ export declare class ArtifactsClient {
     /**
      * `GET /artifacts/find` — lexical search over the artifacts this agent can
      * already reach: what it authored, what its readable containers hold, and
-     * what artifact grants it holds. Bounded by the active lane and shared-room
+     * what artifact grants it holds. Bounded by the active lane and its work
      * containment, exactly as a point read is.
      *
      * This is not `list()` with a query string. `list()` answers "what is in this

package/dist/http/ArtifactsClient.js

@@ -80,7 +80,7 @@ export class ArtifactsClient {
     /**
      * `GET /artifacts/find` — lexical search over the artifacts this agent can
      * already reach: what it authored, what its readable containers hold, and
-     * what artifact grants it holds. Bounded by the active lane and shared-room
+     * what artifact grants it holds. Bounded by the active lane and its work
      * containment, exactly as a point read is.
      *
      * This is not `list()` with a query string. `list()` answers "what is in this

package/dist/http/ContextOpenClient.js

@@ -1,4 +1,4 @@
-import { looksLikeHandle, parseRef } from '../handles.js';
+import { isRetiredRoomRef, looksLikeHandle, parseRef } from '../handles.js';
 import { getBackendUrl } from '../utils/urlUtils.js';
 import { pollSurfaceError } from '../shared/rateLimit.js';
 import { buildOperatorHeaders } from './operatorHeaders.js';
@@ -63,7 +63,7 @@ export function normalizeContextOpenBody(args) {
     if (!raw)
         throw invalid(`${key} must be a non-empty resource reference`);
     const ref = parseRef(raw);
-    if (ref?.kind === 'address' || /^(?:address|chat):/.test(raw))
+    if (isRetiredRoomRef(raw))
         throw invalid('Room references are retired; select an exact artifact, task or agreement');
     if (!ref && looksLikeHandle(raw))
         throw invalid(`${raw} is not a handle: its check letter does not match. Check it for a typo`);

package/dist/http/ContextReadClient.d.ts

@@ -1,9 +1,7 @@
 export declare const CONTEXT_READ_TYPES: readonly ["artifacts", "agreements", "tasks"];
 export type ContextReadType = (typeof CONTEXT_READ_TYPES)[number];
 /**
- * Retained resource entry points for context reads and snapshots.
- * The server still accepts legacy room entries until coordinated removal;
- * new callers use only the domain resources below.
+ * Resource entry points for context reads and snapshots.
  */
 export declare const VIA_KINDS: readonly ["agreement", "task", "artifact", "message"];
 export type ViaKind = (typeof VIA_KINDS)[number];
@@ -57,7 +55,7 @@ export interface ContextSnapshotParticipant {
     [key: string]: unknown;
 }
 export interface ContextSnapshotQuery {
-    /** Exact message, task, agreement, or artifact entry; no implicit room. */
+    /** Exact message, task, agreement, or artifact entry. */
     via: string;
     agreementId?: string;
     artifactId?: string;

package/dist/http/ContextReadClient.js

@@ -7,9 +7,7 @@ export const CONTEXT_READ_TYPES = [
     'tasks',
 ];
 /**
- * Retained resource entry points for context reads and snapshots.
- * The server still accepts legacy room entries until coordinated removal;
- * new callers use only the domain resources below.
+ * Resource entry points for context reads and snapshots.
  */
 export const VIA_KINDS = [
     'agreement',

package/dist/http/agreementFlows.d.ts

@@ -12,8 +12,6 @@ import { type Agreement, type Creds, type EngagementKind } from '../types.js';
  */
 export interface UnifiedProposeInput extends ProposeTerms {
     proposedTo: string;
-    /** Required on direct proposals; optional on broadcasts/links. */
-    addressId?: string;
     engagementKind?: EngagementKind;
 }
 export type ProposeShape = 'direct' | 'request' | 'offer';

package/dist/http/agreementFlows.js

@@ -2,7 +2,7 @@ import { claimAgreement, proposeBroadcast, proposeDirectTo, } from './AgreementC
 import { publishOffer } from './MarketplaceClient.js';
 import { isBroadcastTarget, } from '../types.js';
 export async function proposeUnified(input, creds) {
-    const { proposedTo, addressId, engagementKind, providerId, ...terms } = input;
+    const { proposedTo, engagementKind, providerId, ...terms } = input;
     if (!proposedTo)
         throw new Error('proposedTo is required (a user/agent id, or "everyone"/"org"/"company" to broadcast)');
     // The 'link' arm is gone. A link is not a proposal shape: it has no
@@ -34,24 +34,16 @@ export async function proposeUnified(input, creds) {
         // Buyer-broadcast: the claimer works, your side pays — an open request.
         const agreement = await proposeBroadcast({
             ...terms,
-            addressId: addressId ?? '',
             engagementKind: engagementKind ?? 'service',
             audience,
         }, creds);
         return { agreement, shape: 'request' };
     }
-    // No room is a shape, not an error. The server has always accepted a direct
-    // proposal without one, and this throw was the only thing that made the answer
-    // depend on which client you used: an agreement between parties who have no
-    // conversation had to invent a chat id to be created. Delivery does not go
-    // through the room — the create-time event targets the party slots — so the
-    // counterparty is told either way.
+    // A direct proposal is delivered to the party slots, so the counterparty is
+    // told whether or not the parties have talked before.
     const agreement = await proposeDirectTo({
         ...terms,
         proposedTo,
-        // Omitted rather than sent empty: an absent field says "no room", and an
-        // empty string is one more fiction for the server to interpret.
-        ...(addressId ? { addressId } : {}),
         providerId: providerId?.trim() || proposedTo,
         engagementKind: engagementKind ?? 'service',
     }, creds);

package/dist/index.d.ts

@@ -13,7 +13,7 @@ export type { DelegateEnv, ResolvedDelegateCreds, } from './shared/delegateCreds
 export { RateLimitedError, isRateLimited, parseRetryAfterMs, pollSurfaceError, } from './shared/rateLimit.js';
 export { parseErrorMessage, parseErrorCode, throwApiError, } from './shared/apiError.js';
 export { ApiError } from './types.js';
-export type { Creds, Task, TaskState, PlanStep, PlanStepStatus, Agreement, AgreementParties, AgreementPartySide, EngagementKind, BroadcastAudience, EntryType, ContentType, InboxDeliveryKind, InboxDeliveryRef, InboxChatNews, InboxProposalRef, InboxTaskRef, InboxRequestRef, InboxPlanAckRef, InboxConnectionRequestRef, InboxHumanAttention, InboxEnvelope, InboxAckResult, InboxPeek, InboxReadOptions, } from './types.js';
+export type { Creds, Task, TaskState, PlanStep, PlanStepStatus, Agreement, AgreementParties, AgreementPartySide, EngagementKind, BroadcastAudience, EntryType, ContentType, InboxDeliveryKind, InboxDeliveryRef, InboxProposalRef, InboxTaskRef, InboxRequestRef, InboxPlanAckRef, InboxConnectionRequestRef, InboxHumanAttention, InboxEnvelope, InboxAckResult, InboxPeek, InboxReadOptions, } from './types.js';
 export type { ProposeTerms, ProposeDirectInput, ProposeBroadcastInput, ProposeAgreementData, ClaimOptions, } from './http/AgreementClient.js';
 export { decisionWords, speakableActor } from './decisionWords.js';
 export type { DecisionOutcome, DecisionWords, DecisionWordsInput, } from './decisionWords.js';

package/dist/shared/operationError.js

@@ -11,7 +11,7 @@ const SCOPE_DENIED_HINT = 'For missing access to a returned artifact, chat, agre
     'a pending access approval agreement; it grants no access. After approval, ' +
     'ziggs_open the same resource using the same agent identity. For earlier chat ' +
     'messages request temporal=from-start; replying also requires permission=reply. ' +
-    'Do not issue yourself a grant. A new room does not show this room\'s history.';
+    'Do not issue yourself a grant. A new bucket does not show another conversation\'s history.';
 // a human-authority denial is not "try again with more scope": no
 // retry by this caller can ever pass it, because the guard refuses on being an
 // agent at all, before it looks at what was asked. Without a next action the

package/dist/types.d.ts

@@ -319,12 +319,6 @@ export interface InboxSourceRef {
 }
 /** One delivery in a mailbox, as the wire carries it. */
 export type { InboxDeliveryRef };
-/** Message/artifact deliveries folded by chat, so you can open chats directly. */
-export interface InboxChatNews {
-    addressId: string;
-    count: number;
-    latestAt: string;
-}
 export interface InboxProposalRef {
     executionContext?: import('@ziggs-ai/contracts').WorkExecutionContext;
     executionBlocked?: import('@ziggs-ai/contracts').ExecutionBlock;
@@ -456,8 +450,6 @@ export interface InboxEnvelope extends InboxRequestChannel, Partial<InboxPlanAck
         /** The last row IN this window; everything after it is the backlog. */
         windowEndsAt: string;
     } | null;
-    /** This agent's ASSIGNED chat mail, folded by chat. */
-    chats: InboxChatNews[];
     /**
      * Opaque watermark bundle. Pass to `ack(upTo, { handledDeliveryIds })`
      * VERBATIM after acting — never construct or parse one; the per-mailbox
@@ -519,7 +511,7 @@ export interface InboxChangesResult {
     cursor: string;
     expiresAt: string;
     hasMore: boolean;
-    deliveries: Array<Pick<InboxDeliveryRef, 'eventId' | 'kind' | 'resourceId' | 'addressId' | 'taskId' | 'agreementId' | 'assigneeId' | 'needsHuman' | 'ts'> & {
+    deliveries: Array<Pick<InboxDeliveryRef, 'eventId' | 'kind' | 'resourceId' | 'taskId' | 'agreementId' | 'assigneeId' | 'needsHuman' | 'ts'> & {
         orgId?: string | null;
     }>;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ziggs-ai/api-client",
-  "version": "0.40.9",
+  "version": "0.41.0",
   "description": "HTTP and WebSocket client for the Ziggs backend API",
   "homepage": "https://ziggsai.com/docs",
   "bugs": {