@cabane/companion 0.6.66 → 0.6.67

package/dist/cli.js

@@ -2287,7 +2287,6 @@ var ACTIVE_RUN_OUTBOX_SEQ = 0;
 function activeRunOutboxKey(conversationId, agentId) {
   return `activerun:${conversationId}:${agentId}`;
 }
-var ACTIVE_CONVERSATION_HEADER = "x-cabane-active-conversation";
 var CabaneApi = class {
   constructor(opts) {
     this.opts = opts;
@@ -2425,45 +2424,6 @@ var CabaneApi = class {
   getConversation(workspaceId, conversationId) {
     return this.request("GET", `/api/workspaces/${workspaceId}/conversations/${conversationId}`);
   }
-  // CT339: spawn a sub-agent conversation — the public create-conversation
-  // endpoint (`POST /conversations/messages`) with an addressed agent ask,
-  // the machinery the `sub_agent` turn-control tool is sugar over. Two things make
-  // it distinct from an ordinary `request` call, so it does its own `fetch`:
-  //   - a PER-CALL bearer — the turn's OBO token when the API minted one, else the
-  //     companion PAT — so the spawn carries the same authority as the agent's other
-  //     cabane calls this turn (mirrors build-options' `turnToken ?? agentPat`);
-  //   - the `x-cabane-active-conversation` header naming the caller's turn, which
-  //     the server verifies against the live run to resolve the caller pair for
-  //     the explicit child's birth subscription (the same origin that stamps
-  //     provenance).
-  // Single-shot (no outbox/retry): a `sub_agent` spawn is a live, user-visible
-  // action, and its result is returned to the agent immediately. Returns
-  // `{ status, body }` un-thrown so the caller maps validation errors
-  // to a legible tool result rather than an exception.
-  async createSubAgentConversation(workspaceId, activeConversationId, bearer, body) {
-    const res = await fetch(`${this.base}/api/workspaces/${workspaceId}/conversations/messages`, {
-      method: "POST",
-      headers: {
-        Authorization: `Bearer ${bearer}`,
-        "content-type": "application/json",
-        accept: "application/json",
-        // CT238/CT339: the per-turn active-conversation header — server-verified
-        // against the caller's live run before it's honored (never client-trusted).
-        [ACTIVE_CONVERSATION_HEADER]: activeConversationId
-      },
-      body: JSON.stringify(body)
-    });
-    const raw = await res.text();
-    let parsed = void 0;
-    if (raw.length > 0) {
-      try {
-        parsed = JSON.parse(raw);
-      } catch {
-        parsed = raw;
-      }
-    }
-    return { status: res.status, body: parsed };
-  }
   // SJ383: companion-only participant ops. All three authenticate with the
   // workspace's agent-bound PAT (passed as `token` on this client) — never
   // the user PAT.
@@ -3407,7 +3367,7 @@ var turnRequestSchema = z8.object({
     // EXTERNAL adapters (Codex / opencode) mount it by URL under the key
     // `cabane_companion` — using the same `bearer` (the turn token) and the same
     // active-conversation header they send to the `cabane` server — so their
-    // agents get `ask`/`wake_me`/`send`/`sub_agent`/`skip_turn`, the
+    // agents get `ask`/`reply_to`/`wake_me`/`send`/`skip_turn`, the
     // verbs they can't get from the companion's in-process SDK server. Optional:
     // claude-code ignores it (it mounts the in-process instance instead), and
     // every existing `cabane`-block fixture keeps parsing unchanged; the
@@ -3770,7 +3730,7 @@ var CLAUDE_CODE_ADDENDUM = [
   "`mcp__cabane__finalize_upload`, and `mcp__cabane__mint_render_token`; the turn tools are",
   "`mcp__cabane_companion__ask`, `mcp__cabane_companion__reply_to`,",
   "`mcp__cabane_companion__wake_me`,",
-  "`mcp__cabane_companion__send`, `mcp__cabane_companion__sub_agent`, and",
+  "`mcp__cabane_companion__send`, and",
   "`mcp__cabane_companion__skip_turn`. Where this prompt names a tool by its bare verb, call",
   "the prefixed form from your tool list. Your harness's own tools (shell, local file",
   "read/edit) keep their ordinary names and act on the local machine only."
@@ -3882,7 +3842,7 @@ function decideResume(stored, currentCwd) {
 
 // packages/agent-runtime/src/claude-code/options.ts
 var CABANE_MCP_SERVER = "cabane";
-var ACTIVE_CONVERSATION_HEADER2 = "x-cabane-active-conversation";
+var ACTIVE_CONVERSATION_HEADER = "x-cabane-active-conversation";
 async function preToolUseHook(input) {
   const toolInput = "tool_input" in input && input.tool_input && typeof input.tool_input === "object" ? input.tool_input : {};
   const toolName = "tool_name" in input ? input.tool_name : void 0;
@@ -3919,7 +3879,7 @@ function buildClaudeCodeOptions(req, augment) {
       url: req.cabane.mcpUrl,
       headers: {
         Authorization: `Bearer ${req.cabane.bearer}`,
-        [ACTIVE_CONVERSATION_HEADER2]: req.cabane.activeConversationId
+        [ACTIVE_CONVERSATION_HEADER]: req.cabane.activeConversationId
       },
       // Claude Code 2.1.x defers every MCP tool behind tool-search by default;
       // for cabane — the always-needed core toolset — force it into the turn-1
@@ -4741,7 +4701,7 @@ function selectAdapter(registry, runtime) {
 // packages/agent-runtime/src/opencode/addendum.ts
 var OPENCODE_ADDENDUM = [
   "Every tool here goes by its plain name: the workspace tool is `sdk`; the turn tools are",
-  "`ask`, `reply_to`, `wake_me`, `send`, `sub_agent`, `skip_turn`; the ancillary tools are",
+  "`ask`, `reply_to`, `wake_me`, `send`, `skip_turn`; the ancillary tools are",
   "`list_workspaces`, `read_binary`, `upload`, `begin_upload`, `finalize_upload`,",
   "`mint_render_token`; and the host tools are plain verbs too (`bash`, `read`, `edit`). Mind",
   "the collision: a bare `read` or `edit` is the HOST tool, acting on the local machine \u2014 a",
@@ -5008,7 +4968,7 @@ function parseOpencodeModel(model) {
 // packages/agent-runtime/src/opencode/run-spec.ts
 var CABANE_MCP_SERVER2 = "cabane";
 var TURN_CONTROL_MCP_SERVER = "cabane_companion";
-var ACTIVE_CONVERSATION_HEADER3 = "x-cabane-active-conversation";
+var ACTIVE_CONVERSATION_HEADER2 = "x-cabane-active-conversation";
 function buildRunSpec(req, resumeSessionId) {
   const { policy, config } = req;
   const directory = req.local.cwd ?? "";
@@ -5065,7 +5025,7 @@ function buildMcp(req) {
     url: req.cabane.mcpUrl,
     headers: {
       Authorization: `Bearer ${req.cabane.bearer}`,
-      [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
+      [ACTIVE_CONVERSATION_HEADER2]: req.cabane.activeConversationId
     },
     enabled: true
   };
@@ -5075,7 +5035,7 @@ function buildMcp(req) {
       url: req.cabane.turnControlUrl,
       headers: {
         Authorization: `Bearer ${req.cabane.bearer}`,
-        [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
+        [ACTIVE_CONVERSATION_HEADER2]: req.cabane.activeConversationId
       },
       enabled: true
     };
@@ -5718,7 +5678,7 @@ var CODEX_ADDENDUM = [
   "the deferred-tool inventory and invoke that exact qualified name; never conclude the SDK is",
   "absent without attempting discovery and invocation, and if a call fails, report the recorded",
   "error. The turn tools are `mcp__cabane_companion__ask` / `reply_to` / `wake_me` /",
-  "`send` / `sub_agent` / `skip_turn`, and the ancillary tools",
+  "`send` / `skip_turn`, and the ancillary tools",
   "`mcp__cabane__list_workspaces` /",
   "`read_binary` / `upload` / `begin_upload` / `finalize_upload` / `mint_render_token` \u2014 all",
   "possibly deferred too. There is no Cabane `read`/`write`/`search` tool \u2014 those are",
@@ -6082,7 +6042,7 @@ function parseCodexModel(model) {
 // packages/agent-runtime/src/codex/run-spec.ts
 var CABANE_MCP_SERVER3 = "cabane";
 var TURN_CONTROL_MCP_SERVER2 = "cabane_companion";
-var ACTIVE_CONVERSATION_HEADER4 = "x-cabane-active-conversation";
+var ACTIVE_CONVERSATION_HEADER3 = "x-cabane-active-conversation";
 var ENV_ENVELOPE_KEYS = [
   "CABANE_PLAYGROUND",
   "CABANE_PLAYGROUND_BIN",
@@ -6166,7 +6126,7 @@ function buildConfig(req, baseInstructionsFile) {
     url: req.cabane.mcpUrl,
     http_headers: {
       Authorization: `Bearer ${req.cabane.bearer}`,
-      [ACTIVE_CONVERSATION_HEADER4]: req.cabane.activeConversationId
+      [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
     },
     default_tools_approval_mode: "approve"
   };
@@ -6175,7 +6135,7 @@ function buildConfig(req, baseInstructionsFile) {
       url: req.cabane.turnControlUrl,
       http_headers: {
         Authorization: `Bearer ${req.cabane.bearer}`,
-        [ACTIVE_CONVERSATION_HEADER4]: req.cabane.activeConversationId
+        [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
       },
       default_tools_approval_mode: "approve"
     };
@@ -7229,8 +7189,6 @@ var ASK_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${ASK_TOOL}`;
 var REPLY_TO_TOOL = "reply_to";
 var REPLY_TO_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${REPLY_TO_TOOL}`;
 var MAX_ASK_ITEMS = 20;
-var SUB_AGENT_TOOL = "sub_agent";
-var SUB_AGENT_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${SUB_AGENT_TOOL}`;
 var WAKE_ME_TOOL = "wake_me";
 var WAKE_ME_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${WAKE_ME_TOOL}`;
 var CANCEL_WAKE_TOOL = "cancel_wake";
@@ -7269,8 +7227,8 @@ function resolveDeclaredReplyField(input) {
   );
   const askRaised = Boolean(input.askState.targetUserId);
   const wakeArmed = input.wakeState.afterSeconds !== null || input.wakeState.at !== null;
-  if (outwardSend || askRaised || wakeArmed || input.spawnedSubAgent) return {};
-  return { answersMessageId: input.owedReplyMessageId };
+  if (outwardSend || askRaised || wakeArmed) return {};
+  return { answersMessageId: input.owedReplyMessageId, answersAutoDeclared: true };
 }
 function wakeCommitField(state) {
   if (state.cancelled) return { wake: { cancel: true } };
@@ -7284,7 +7242,7 @@ function wakeCommitField(state) {
     }
   };
 }
-function createTurnControlMcpServer(sendState, skipState, askState, subAgentCreate, wakeState, replyState, controlOrder) {
+function createTurnControlMcpServer(sendState, skipState, askState, wakeState, replyState, controlOrder) {
   return createSdkMcpServer({
     name: COMPANION_LOCAL_MCP_SERVER,
     version: "0.0.0",
@@ -7420,42 +7378,6 @@ function createTurnControlMcpServer(sendState, skipState, askState, subAgentCrea
           { annotations: { readOnlyHint: true, openWorldHint: false }, alwaysLoad: true }
         )
       ] : [],
-      ...subAgentCreate ? [
-        tool(
-          SUB_AGENT_TOOL,
-          "Ask into a child conversation \u2014 a bounded question or piece of work whose ANSWER returns to you automatically. This creates the child and dispatches its agent now, returning the child's `conversationId` immediately; the answer lands LATER as a message in THIS conversation, and you're woken once every child you asked this turn has answered. So DON'T wait: after asking, finish whatever else this turn can do and end your turn (never poll the child with reads in a loop \u2014 the wake is automatic and it carries the answer). Parallel fan-out = call this N times in one turn (they run concurrently; ONE wake carrying all the answers); series = one call per turn. `prompt` is the ask itself \u2014 write it as a request that expects an answer back, not a role you are assigning or a job description: self-contained (the child starts fresh, with only this prompt), and say what shape the answer should take. `agentId` (optional) asks a PEER instead of yourself \u2014 same mechanics, a different mind (use for capability/context you lack); default (self) is a fresh context window of you. `title` (optional) names the child thread (answers link it, so a legible title helps). A single child buys no wall-clock time (you idle either way) \u2014 it pays when the callee has capability/context you lack, or to isolate a big read from your own session; the real win is fan-out. Don't ask a child for a lookup you can do in-turn with your own tools. The answer returns to YOU to act on \u2014 reach for it when you're the consumer of the output, not as a way to notify a human: if a person just wants to read the result, dispatch a plain (no-return) conversation and link it instead.",
-          {
-            prompt: z13.string().min(1).max(65536).describe(
-              "The ask \u2014 a self-contained request that expects an answer back (the child starts fresh: only this prompt + the thread it lands in). Say what shape the answer should take."
-            ),
-            agentId: z13.string().uuid().optional().describe(
-              "Optional peer to run the sub-agent as (a workspace agent id, from your turn context's roster); omit to spawn yourself with a fresh context window."
-            ),
-            title: z13.string().max(200).optional().describe("Optional title for the child thread (result chips link it).")
-          },
-          async (args) => {
-            const result = await subAgentCreate(args);
-            if (!result.ok) {
-              return {
-                content: [{ type: "text", text: result.error }],
-                isError: true
-              };
-            }
-            return {
-              content: [
-                {
-                  type: "text",
-                  text: JSON.stringify({
-                    conversationId: result.conversationId,
-                    note: "Asked. Don't wait for it \u2014 finish what else this turn can do, then end your turn; the answer posts back here and you're woken once every child you asked this turn has answered."
-                  })
-                }
-              ]
-            };
-          },
-          { annotations: { readOnlyHint: false, openWorldHint: true }, alwaysLoad: true }
-        )
-      ] : [],
       ...wakeState ? [
         tool(
           WAKE_ME_TOOL,
@@ -7995,7 +7917,10 @@ var TurnCommitter = class {
   }
   // The declared reply: explicit `reply_to` first, else the runtime's own
   // declaration for the turn that just answers — see
-  // `resolveDeclaredReplyField` for the whole rule and its reasons.
+  // `resolveDeclaredReplyField` for the whole rule and its reasons. CT1224: an
+  // auto-declaration is flagged as such on the wire, because the server drops it
+  // when the turn also opened a conversation (a fan-out is mid-arc). An explicit
+  // `reply_to` carries no flag and always stands.
   answersField(kind) {
     return resolveDeclaredReplyField({
       kind,
@@ -8004,8 +7929,7 @@ var TurnCommitter = class {
       replyState: this.deps.replyState,
       sendState: this.deps.sendState,
       askState: this.deps.askState,
-      wakeState: this.deps.wakeState,
-      spawnedSubAgent: this.deps.spawnedSubAgent()
+      wakeState: this.deps.wakeState
     });
   }
   // Carry both halves of the addressed send. The server writes `dispatchBody`
@@ -8126,17 +8050,6 @@ function describeErrorBody(body) {
   if (text.length === 0) return void 0;
   return text.length > ERROR_BODY_LOG_CAP ? `${text.slice(0, ERROR_BODY_LOG_CAP)}\u2026` : text;
 }
-function describeSubAgentError(status2, body) {
-  const code = body && typeof body === "object" && "error" in body ? String(body.error) : void 0;
-  switch (code) {
-    case "dispatch_agent_not_found":
-      return "sub_agent: no live agent in this workspace matches that `agentId`. Check the roster in your turn context, or omit `agentId` to spawn yourself.";
-    case "dispatch_ask_requires_dispatch":
-      return `sub_agent: the spawn was rejected (${code}). This is a turn-context problem, not something to retry blindly \u2014 report it rather than looping.`;
-    default:
-      return `sub_agent: the spawn failed (${code ?? `HTTP ${status2}`}).`;
-  }
-}
 var Dispatcher = class {
   constructor(opts) {
     this.opts = opts;
@@ -8499,36 +8412,10 @@ ${reason}`,
     const wakeState = createWakeState();
     const replyState = createReplyState();
     const controlOrder = createTurnControlOrder();
-    let spawnedSubAgent = false;
-    const subAgentCreate = async (args) => {
-      const dispatchTarget = args.agentId ?? payload.agentId;
-      const projectId = turnContext.conversation.projectId;
-      const { status: status2, body } = await this.opts.api.createSubAgentConversation(
-        workspaceId,
-        payload.conversationId,
-        turnContext.turnToken ?? this.opts.credential,
-        {
-          ...projectId ? { projectId } : {},
-          ...args.title ? { title: args.title } : {},
-          body: args.prompt,
-          dispatch: dispatchTarget,
-          dispatchAsk: true,
-          parentConversationId: payload.conversationId
-        }
-      );
-      if (status2 >= 400) return { ok: false, error: describeSubAgentError(status2, body) };
-      const conversationId = body?.conversation?.id;
-      if (!conversationId) {
-        return { ok: false, error: "sub_agent: the spawn returned no conversation id." };
-      }
-      spawnedSubAgent = true;
-      return { ok: true, conversationId };
-    };
     const turnControlServer = createTurnControlMcpServer(
       sendState,
       skipState,
       askState,
-      subAgentCreate,
       wakeState,
       replyState,
       controlOrder
@@ -8664,8 +8551,7 @@ ${reason}`,
       replyState,
       // The ledger-derived owed reply, for the runtime's own declaration when
       // the agent doesn't call `reply_to` (resolveDeclaredReplyField).
-      owedReplyMessageId: turnContext.owedReplyMessageId ?? null,
-      spawnedSubAgent: () => spawnedSubAgent
+      owedReplyMessageId: turnContext.owedReplyMessageId ?? null
     });
     const usesHttpTurnControl = turnRuntime === "codex" || turnRuntime === "opencode";
     let turnControlIntentFetched = false;

package/dist/runtime.js

@@ -1707,7 +1707,6 @@ var ACTIVE_RUN_OUTBOX_SEQ = 0;
 function activeRunOutboxKey(conversationId, agentId) {
   return `activerun:${conversationId}:${agentId}`;
 }
-var ACTIVE_CONVERSATION_HEADER = "x-cabane-active-conversation";
 var CabaneApi = class {
   constructor(opts) {
     this.opts = opts;
@@ -1845,45 +1844,6 @@ var CabaneApi = class {
   getConversation(workspaceId, conversationId) {
     return this.request("GET", `/api/workspaces/${workspaceId}/conversations/${conversationId}`);
   }
-  // CT339: spawn a sub-agent conversation — the public create-conversation
-  // endpoint (`POST /conversations/messages`) with an addressed agent ask,
-  // the machinery the `sub_agent` turn-control tool is sugar over. Two things make
-  // it distinct from an ordinary `request` call, so it does its own `fetch`:
-  //   - a PER-CALL bearer — the turn's OBO token when the API minted one, else the
-  //     companion PAT — so the spawn carries the same authority as the agent's other
-  //     cabane calls this turn (mirrors build-options' `turnToken ?? agentPat`);
-  //   - the `x-cabane-active-conversation` header naming the caller's turn, which
-  //     the server verifies against the live run to resolve the caller pair for
-  //     the explicit child's birth subscription (the same origin that stamps
-  //     provenance).
-  // Single-shot (no outbox/retry): a `sub_agent` spawn is a live, user-visible
-  // action, and its result is returned to the agent immediately. Returns
-  // `{ status, body }` un-thrown so the caller maps validation errors
-  // to a legible tool result rather than an exception.
-  async createSubAgentConversation(workspaceId, activeConversationId, bearer, body) {
-    const res = await fetch(`${this.base}/api/workspaces/${workspaceId}/conversations/messages`, {
-      method: "POST",
-      headers: {
-        Authorization: `Bearer ${bearer}`,
-        "content-type": "application/json",
-        accept: "application/json",
-        // CT238/CT339: the per-turn active-conversation header — server-verified
-        // against the caller's live run before it's honored (never client-trusted).
-        [ACTIVE_CONVERSATION_HEADER]: activeConversationId
-      },
-      body: JSON.stringify(body)
-    });
-    const raw = await res.text();
-    let parsed = void 0;
-    if (raw.length > 0) {
-      try {
-        parsed = JSON.parse(raw);
-      } catch {
-        parsed = raw;
-      }
-    }
-    return { status: res.status, body: parsed };
-  }
   // SJ383: companion-only participant ops. All three authenticate with the
   // workspace's agent-bound PAT (passed as `token` on this client) — never
   // the user PAT.
@@ -2906,7 +2866,7 @@ var turnRequestSchema = z8.object({
     // EXTERNAL adapters (Codex / opencode) mount it by URL under the key
     // `cabane_companion` — using the same `bearer` (the turn token) and the same
     // active-conversation header they send to the `cabane` server — so their
-    // agents get `ask`/`wake_me`/`send`/`sub_agent`/`skip_turn`, the
+    // agents get `ask`/`reply_to`/`wake_me`/`send`/`skip_turn`, the
     // verbs they can't get from the companion's in-process SDK server. Optional:
     // claude-code ignores it (it mounts the in-process instance instead), and
     // every existing `cabane`-block fixture keeps parsing unchanged; the
@@ -3269,7 +3229,7 @@ var CLAUDE_CODE_ADDENDUM = [
   "`mcp__cabane__finalize_upload`, and `mcp__cabane__mint_render_token`; the turn tools are",
   "`mcp__cabane_companion__ask`, `mcp__cabane_companion__reply_to`,",
   "`mcp__cabane_companion__wake_me`,",
-  "`mcp__cabane_companion__send`, `mcp__cabane_companion__sub_agent`, and",
+  "`mcp__cabane_companion__send`, and",
   "`mcp__cabane_companion__skip_turn`. Where this prompt names a tool by its bare verb, call",
   "the prefixed form from your tool list. Your harness's own tools (shell, local file",
   "read/edit) keep their ordinary names and act on the local machine only."
@@ -3381,7 +3341,7 @@ function decideResume(stored, currentCwd) {
 
 // packages/agent-runtime/src/claude-code/options.ts
 var CABANE_MCP_SERVER = "cabane";
-var ACTIVE_CONVERSATION_HEADER2 = "x-cabane-active-conversation";
+var ACTIVE_CONVERSATION_HEADER = "x-cabane-active-conversation";
 async function preToolUseHook(input) {
   const toolInput = "tool_input" in input && input.tool_input && typeof input.tool_input === "object" ? input.tool_input : {};
   const toolName = "tool_name" in input ? input.tool_name : void 0;
@@ -3418,7 +3378,7 @@ function buildClaudeCodeOptions(req, augment) {
       url: req.cabane.mcpUrl,
       headers: {
         Authorization: `Bearer ${req.cabane.bearer}`,
-        [ACTIVE_CONVERSATION_HEADER2]: req.cabane.activeConversationId
+        [ACTIVE_CONVERSATION_HEADER]: req.cabane.activeConversationId
       },
       // Claude Code 2.1.x defers every MCP tool behind tool-search by default;
       // for cabane — the always-needed core toolset — force it into the turn-1
@@ -4240,7 +4200,7 @@ function selectAdapter(registry, runtime) {
 // packages/agent-runtime/src/opencode/addendum.ts
 var OPENCODE_ADDENDUM = [
   "Every tool here goes by its plain name: the workspace tool is `sdk`; the turn tools are",
-  "`ask`, `reply_to`, `wake_me`, `send`, `sub_agent`, `skip_turn`; the ancillary tools are",
+  "`ask`, `reply_to`, `wake_me`, `send`, `skip_turn`; the ancillary tools are",
   "`list_workspaces`, `read_binary`, `upload`, `begin_upload`, `finalize_upload`,",
   "`mint_render_token`; and the host tools are plain verbs too (`bash`, `read`, `edit`). Mind",
   "the collision: a bare `read` or `edit` is the HOST tool, acting on the local machine \u2014 a",
@@ -4507,7 +4467,7 @@ function parseOpencodeModel(model) {
 // packages/agent-runtime/src/opencode/run-spec.ts
 var CABANE_MCP_SERVER2 = "cabane";
 var TURN_CONTROL_MCP_SERVER = "cabane_companion";
-var ACTIVE_CONVERSATION_HEADER3 = "x-cabane-active-conversation";
+var ACTIVE_CONVERSATION_HEADER2 = "x-cabane-active-conversation";
 function buildRunSpec(req, resumeSessionId) {
   const { policy, config } = req;
   const directory = req.local.cwd ?? "";
@@ -4564,7 +4524,7 @@ function buildMcp(req) {
     url: req.cabane.mcpUrl,
     headers: {
       Authorization: `Bearer ${req.cabane.bearer}`,
-      [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
+      [ACTIVE_CONVERSATION_HEADER2]: req.cabane.activeConversationId
     },
     enabled: true
   };
@@ -4574,7 +4534,7 @@ function buildMcp(req) {
       url: req.cabane.turnControlUrl,
       headers: {
         Authorization: `Bearer ${req.cabane.bearer}`,
-        [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
+        [ACTIVE_CONVERSATION_HEADER2]: req.cabane.activeConversationId
       },
       enabled: true
     };
@@ -5217,7 +5177,7 @@ var CODEX_ADDENDUM = [
   "the deferred-tool inventory and invoke that exact qualified name; never conclude the SDK is",
   "absent without attempting discovery and invocation, and if a call fails, report the recorded",
   "error. The turn tools are `mcp__cabane_companion__ask` / `reply_to` / `wake_me` /",
-  "`send` / `sub_agent` / `skip_turn`, and the ancillary tools",
+  "`send` / `skip_turn`, and the ancillary tools",
   "`mcp__cabane__list_workspaces` /",
   "`read_binary` / `upload` / `begin_upload` / `finalize_upload` / `mint_render_token` \u2014 all",
   "possibly deferred too. There is no Cabane `read`/`write`/`search` tool \u2014 those are",
@@ -5581,7 +5541,7 @@ function parseCodexModel(model) {
 // packages/agent-runtime/src/codex/run-spec.ts
 var CABANE_MCP_SERVER3 = "cabane";
 var TURN_CONTROL_MCP_SERVER2 = "cabane_companion";
-var ACTIVE_CONVERSATION_HEADER4 = "x-cabane-active-conversation";
+var ACTIVE_CONVERSATION_HEADER3 = "x-cabane-active-conversation";
 var ENV_ENVELOPE_KEYS = [
   "CABANE_PLAYGROUND",
   "CABANE_PLAYGROUND_BIN",
@@ -5665,7 +5625,7 @@ function buildConfig(req, baseInstructionsFile) {
     url: req.cabane.mcpUrl,
     http_headers: {
       Authorization: `Bearer ${req.cabane.bearer}`,
-      [ACTIVE_CONVERSATION_HEADER4]: req.cabane.activeConversationId
+      [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
     },
     default_tools_approval_mode: "approve"
   };
@@ -5674,7 +5634,7 @@ function buildConfig(req, baseInstructionsFile) {
       url: req.cabane.turnControlUrl,
       http_headers: {
         Authorization: `Bearer ${req.cabane.bearer}`,
-        [ACTIVE_CONVERSATION_HEADER4]: req.cabane.activeConversationId
+        [ACTIVE_CONVERSATION_HEADER3]: req.cabane.activeConversationId
       },
       default_tools_approval_mode: "approve"
     };
@@ -6728,8 +6688,6 @@ var ASK_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${ASK_TOOL}`;
 var REPLY_TO_TOOL = "reply_to";
 var REPLY_TO_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${REPLY_TO_TOOL}`;
 var MAX_ASK_ITEMS = 20;
-var SUB_AGENT_TOOL = "sub_agent";
-var SUB_AGENT_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${SUB_AGENT_TOOL}`;
 var WAKE_ME_TOOL = "wake_me";
 var WAKE_ME_TOOL_NAME = `mcp__${COMPANION_LOCAL_MCP_SERVER}__${WAKE_ME_TOOL}`;
 var CANCEL_WAKE_TOOL = "cancel_wake";
@@ -6768,8 +6726,8 @@ function resolveDeclaredReplyField(input) {
   );
   const askRaised = Boolean(input.askState.targetUserId);
   const wakeArmed = input.wakeState.afterSeconds !== null || input.wakeState.at !== null;
-  if (outwardSend || askRaised || wakeArmed || input.spawnedSubAgent) return {};
-  return { answersMessageId: input.owedReplyMessageId };
+  if (outwardSend || askRaised || wakeArmed) return {};
+  return { answersMessageId: input.owedReplyMessageId, answersAutoDeclared: true };
 }
 function wakeCommitField(state) {
   if (state.cancelled) return { wake: { cancel: true } };
@@ -6783,7 +6741,7 @@ function wakeCommitField(state) {
     }
   };
 }
-function createTurnControlMcpServer(sendState, skipState, askState, subAgentCreate, wakeState, replyState, controlOrder) {
+function createTurnControlMcpServer(sendState, skipState, askState, wakeState, replyState, controlOrder) {
   return createSdkMcpServer({
     name: COMPANION_LOCAL_MCP_SERVER,
     version: "0.0.0",
@@ -6919,42 +6877,6 @@ function createTurnControlMcpServer(sendState, skipState, askState, subAgentCrea
           { annotations: { readOnlyHint: true, openWorldHint: false }, alwaysLoad: true }
         )
       ] : [],
-      ...subAgentCreate ? [
-        tool(
-          SUB_AGENT_TOOL,
-          "Ask into a child conversation \u2014 a bounded question or piece of work whose ANSWER returns to you automatically. This creates the child and dispatches its agent now, returning the child's `conversationId` immediately; the answer lands LATER as a message in THIS conversation, and you're woken once every child you asked this turn has answered. So DON'T wait: after asking, finish whatever else this turn can do and end your turn (never poll the child with reads in a loop \u2014 the wake is automatic and it carries the answer). Parallel fan-out = call this N times in one turn (they run concurrently; ONE wake carrying all the answers); series = one call per turn. `prompt` is the ask itself \u2014 write it as a request that expects an answer back, not a role you are assigning or a job description: self-contained (the child starts fresh, with only this prompt), and say what shape the answer should take. `agentId` (optional) asks a PEER instead of yourself \u2014 same mechanics, a different mind (use for capability/context you lack); default (self) is a fresh context window of you. `title` (optional) names the child thread (answers link it, so a legible title helps). A single child buys no wall-clock time (you idle either way) \u2014 it pays when the callee has capability/context you lack, or to isolate a big read from your own session; the real win is fan-out. Don't ask a child for a lookup you can do in-turn with your own tools. The answer returns to YOU to act on \u2014 reach for it when you're the consumer of the output, not as a way to notify a human: if a person just wants to read the result, dispatch a plain (no-return) conversation and link it instead.",
-          {
-            prompt: z13.string().min(1).max(65536).describe(
-              "The ask \u2014 a self-contained request that expects an answer back (the child starts fresh: only this prompt + the thread it lands in). Say what shape the answer should take."
-            ),
-            agentId: z13.string().uuid().optional().describe(
-              "Optional peer to run the sub-agent as (a workspace agent id, from your turn context's roster); omit to spawn yourself with a fresh context window."
-            ),
-            title: z13.string().max(200).optional().describe("Optional title for the child thread (result chips link it).")
-          },
-          async (args) => {
-            const result = await subAgentCreate(args);
-            if (!result.ok) {
-              return {
-                content: [{ type: "text", text: result.error }],
-                isError: true
-              };
-            }
-            return {
-              content: [
-                {
-                  type: "text",
-                  text: JSON.stringify({
-                    conversationId: result.conversationId,
-                    note: "Asked. Don't wait for it \u2014 finish what else this turn can do, then end your turn; the answer posts back here and you're woken once every child you asked this turn has answered."
-                  })
-                }
-              ]
-            };
-          },
-          { annotations: { readOnlyHint: false, openWorldHint: true }, alwaysLoad: true }
-        )
-      ] : [],
       ...wakeState ? [
         tool(
           WAKE_ME_TOOL,
@@ -7494,7 +7416,10 @@ var TurnCommitter = class {
   }
   // The declared reply: explicit `reply_to` first, else the runtime's own
   // declaration for the turn that just answers — see
-  // `resolveDeclaredReplyField` for the whole rule and its reasons.
+  // `resolveDeclaredReplyField` for the whole rule and its reasons. CT1224: an
+  // auto-declaration is flagged as such on the wire, because the server drops it
+  // when the turn also opened a conversation (a fan-out is mid-arc). An explicit
+  // `reply_to` carries no flag and always stands.
   answersField(kind) {
     return resolveDeclaredReplyField({
       kind,
@@ -7503,8 +7428,7 @@ var TurnCommitter = class {
       replyState: this.deps.replyState,
       sendState: this.deps.sendState,
       askState: this.deps.askState,
-      wakeState: this.deps.wakeState,
-      spawnedSubAgent: this.deps.spawnedSubAgent()
+      wakeState: this.deps.wakeState
     });
   }
   // Carry both halves of the addressed send. The server writes `dispatchBody`
@@ -7625,17 +7549,6 @@ function describeErrorBody(body) {
   if (text.length === 0) return void 0;
   return text.length > ERROR_BODY_LOG_CAP ? `${text.slice(0, ERROR_BODY_LOG_CAP)}\u2026` : text;
 }
-function describeSubAgentError(status, body) {
-  const code = body && typeof body === "object" && "error" in body ? String(body.error) : void 0;
-  switch (code) {
-    case "dispatch_agent_not_found":
-      return "sub_agent: no live agent in this workspace matches that `agentId`. Check the roster in your turn context, or omit `agentId` to spawn yourself.";
-    case "dispatch_ask_requires_dispatch":
-      return `sub_agent: the spawn was rejected (${code}). This is a turn-context problem, not something to retry blindly \u2014 report it rather than looping.`;
-    default:
-      return `sub_agent: the spawn failed (${code ?? `HTTP ${status}`}).`;
-  }
-}
 var Dispatcher = class {
   constructor(opts) {
     this.opts = opts;
@@ -7998,36 +7911,10 @@ ${reason}`,
     const wakeState = createWakeState();
     const replyState = createReplyState();
     const controlOrder = createTurnControlOrder();
-    let spawnedSubAgent = false;
-    const subAgentCreate = async (args) => {
-      const dispatchTarget = args.agentId ?? payload.agentId;
-      const projectId = turnContext.conversation.projectId;
-      const { status, body } = await this.opts.api.createSubAgentConversation(
-        workspaceId,
-        payload.conversationId,
-        turnContext.turnToken ?? this.opts.credential,
-        {
-          ...projectId ? { projectId } : {},
-          ...args.title ? { title: args.title } : {},
-          body: args.prompt,
-          dispatch: dispatchTarget,
-          dispatchAsk: true,
-          parentConversationId: payload.conversationId
-        }
-      );
-      if (status >= 400) return { ok: false, error: describeSubAgentError(status, body) };
-      const conversationId = body?.conversation?.id;
-      if (!conversationId) {
-        return { ok: false, error: "sub_agent: the spawn returned no conversation id." };
-      }
-      spawnedSubAgent = true;
-      return { ok: true, conversationId };
-    };
     const turnControlServer = createTurnControlMcpServer(
       sendState,
       skipState,
       askState,
-      subAgentCreate,
       wakeState,
       replyState,
       controlOrder
@@ -8163,8 +8050,7 @@ ${reason}`,
       replyState,
       // The ledger-derived owed reply, for the runtime's own declaration when
       // the agent doesn't call `reply_to` (resolveDeclaredReplyField).
-      owedReplyMessageId: turnContext.owedReplyMessageId ?? null,
-      spawnedSubAgent: () => spawnedSubAgent
+      owedReplyMessageId: turnContext.owedReplyMessageId ?? null
     });
     const usesHttpTurnControl = turnRuntime === "codex" || turnRuntime === "opencode";
     let turnControlIntentFetched = false;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cabane/companion",
-  "version": "0.6.66",
+  "version": "0.6.67",
   "type": "module",
   "description": "The Cabane Companion (headless): connect a coding agent on your machine to your Cabane workspace as a responder — drive work against your own codebase, files, and MCP servers without putting any of it in Cabane.",
   "license": "UNLICENSED",