muse-crew 0.14.8 → 0.14.9

package/docs/decisions/workflow-core.md

@@ -52,17 +52,24 @@ Applies to: standard, bugfix, chore.
 <a id="closeout-envelope"></a>
 ## Closeout transport envelope
 
-Invariant: the work agent returns the runtime's native transport envelope with no schema; the verdict is extracted mechanically by extractVerdict — never by an agent.
+Invariant: the schema-less work-agent courier returns prose; the workflow
+consumes the return as a plain string. Nothing parses a JSON envelope —
+the prompt names none (blocker 43, 2026-09-22: the old "Return your work as
+JSON in exactly this shape: {...}" invitation drew brace-shaped prose into
+a channel where the runtime's JSON-candidate scan threw, and three
+consecutive Integrate failures parked room #28's J1 at the retry cap; the
+second REVIEW pass then reworded the retry trailer's "discarded" why off
+"machine-read as JSON" to the true event — {...}-shaped fragments the scan
+could not parse — since the old framing was an implicit JSON invitation).
+The verdict is extracted mechanically by extractVerdict — never by an agent.
 
 Applies to: standard, bugfix, chore.
 
 ```
-  // Work agent returns the runtime's native envelope {"status": "ok",
-  // "result": "<prose>"} with no schema. The runtime requires JSON output;
-  // the envelope is its own documented shape, so there is nothing for the
-  // agent to improvise. The workflow receives the prose report as a plain
-  // string. The verdict is still extracted deterministically from the report
-  // text by extractVerdict below — never by an agent.
+  // Work agent returns prose on the schema-less courier; the workflow
+  // consumes the return as a plain string. The verdict is still extracted
+  // deterministically from the report text by extractVerdict — never by
+  // an agent.
 ```
 
 <a id="extract-verdict-contract"></a>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "muse-crew",
-  "version": "0.14.8",
+  "version": "0.14.9",
   "description": "Opinionated orchestration for Muse — workflows, identities, and tooling for autonomous software development.",
   "license": "UNLICENSED",
   "private": false,

package/workflows/bugfix.js

@@ -131,7 +131,7 @@ if (!taskId) {
   throw new Error("task_id is required in args");
 }
 
-// See docs/decisions/workflow-core.md#closeout-envelope: the work agent returns the runtime's native envelope; verdict extracted mechanically.
+// See docs/decisions/workflow-core.md#closeout-envelope: the schema-less work courier returns prose, consumed as a plain string; verdict extracted mechanically.
 const VERDICT_STEPS = ["Build", "Review", "QA", "Reproduce", "Integrate", "Publish"];
 function extractVerdict(workerText) {
  // The verdict is the LAST VERDICT: PASS/FAIL in the report (contract: end
@@ -221,7 +221,6 @@ function buildVerdictReaskPrompt(stepName, workerText) {
     "Decide from the report's own content whether the " + stepName + " step clearly describes successful completion: " +
     "if it does, the verdict is PASS; otherwise — failure, error, unfinished work, or unclear — the verdict is FAIL. " +
     "Do NOT copy any VERDICT line from the report — decide from the content.\n\n" +
-    "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your verdict line here\"}. " +
     "The result must be exactly one line and nothing else: VERDICT: PASS or VERDICT: FAIL.";
 }
 async function reaskVerdict(stepName, reworkSuffix, workerText) {
@@ -294,8 +293,8 @@ var TOOL_CHECK_PREAMBLE =
   "3. Run the shell command: echo tool-probe-ok - then write exactly one line: shell_transport: ok - or shell_transport: unavailable if you cannot run shell commands.\n" +
   "Then do the assignment below.\n\n";
 function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason) {
-  // reason: "discarded" (the runtime threw the output away - it could not be
-  // machine-read), "empty" (agent() returned without throwing but produced
+  // reason: "discarded" (the runtime threw the output away — the JSON-candidate
+  // scan found {...}-shaped fragments it could not parse), "empty" (agent() returned without throwing but produced
   // nothing usable), "no-tools" (the worker's TOOL CHECK reported
   // artifact_tools: missing), or "no-transport" (the worker's TOOL CHECK
   // reported shell_transport: unavailable).
@@ -306,11 +305,11 @@ function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason)
     ? "your previous attempt's TOOL CHECK reported artifact_tools: missing (this is a fresh launch, so run the TOOL CHECK's load step again before the work)"
     : reason === "no-transport"
     ? "your previous attempt's TOOL CHECK reported shell_transport: unavailable (this is a fresh launch, so run the TOOL CHECK's shell probe again before the work)"
-    : "your previous attempt's output could not be machine-read as JSON and was discarded";
+    : "your previous attempt's output was discarded by the transport because it contained {...}-shaped fragments the transport could not parse; write plain prose with no JSON-shaped fragments";
   return "\n\nTRANSPORT RETRY (attempt " + attempt + " of 2): " + why + ". " +
     "First check existing state (worktree/branch at " + repoPath + "/.worktrees/" + taskId + ", the task branch, dashboard sessions for this task) - " +
     "if the " + stepName + " work is already complete, report on what was done rather than duplicating side effects. " +
-    "Then return your report as JSON in exactly the shape specified above.";
+    "Then return your report in exactly the shape specified above.";
 }
 // parseClassifyFerry — room #26 blocker 42 (2026-09-21). Shape-constrained
 // ferry parser for the classify-branch agent() call. The schema buys SHAPE,
@@ -2172,12 +2171,12 @@ while (i < STEPS.length) {
       "The returned events are filtered to this task. They contain notes and decisions from prior phases.\n\n";
   }
 
-  // Work agent returns the runtime's native envelope {"status": "ok",
-  // "result": "<prose>"} with no schema. The runtime requires JSON output;
-  // the envelope is its own documented shape, so there is nothing for the
-  // agent to improvise. The workflow receives the prose report as a plain
-  // string. The verdict is still extracted deterministically from the report
-  // text by extractVerdict below — never by an agent.
+  // Work agent returns prose on the schema-less courier; the workflow
+  // consumes the return as a plain string (blocker 43, 2026-09-22: the
+  // prompt names no JSON envelope — nothing parses one, and the runtime's
+  // JSON-candidate scan discards brace-shaped prose, which is what the
+  // transport retry guards). The verdict is still extracted deterministically
+  // from the report text by extractVerdict below — never by an agent.
   var workPromptBase =
     TOOL_CHECK_PREAMBLE +
     "Read the identity file at " + ORCH_PATH + "/identities/" + step.identity + ".md using the read tool, and embody that character fully.\n\n" +
@@ -2190,8 +2189,7 @@ while (i < STEPS.length) {
     "\n## Instructions\n\n" + eventPreamble + instructions + "\n\n" +
     "CONSTRAINT: Do NOT call logevent or upsertagentsession — the workflow handles all phase tracking after your step completes.\n\n" +
     "Stay in character. Do the work thoroughly.\n\n" +
-    "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your report here\"}. " +
-    "The result is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
+    "Your report is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
   var workKeyBase = "work-" + step.name + (totalReworkCount > 0 ? "-r" + totalReworkCount : "");
   // See docs/decisions/qa-reproduce.md#wrong-layer-guard: reproduce must run at the correct layer.
   var reproEvidenceDir = crewHome + "/task-evidence/" + taskId + "/repro";

package/workflows/chore.js

@@ -197,7 +197,7 @@ if (!taskId) {
   throw new Error("task_id is required in args");
 }
 
-// See docs/decisions/workflow-core.md#closeout-envelope: the work agent returns the runtime's native envelope; verdict extracted mechanically.
+// See docs/decisions/workflow-core.md#closeout-envelope: the schema-less work courier returns prose, consumed as a plain string; verdict extracted mechanically.
 const VERDICT_STEPS = ["Build", "Review", "Integrate", "Publish"];
 function extractVerdict(workerText) {
   // The verdict is the LAST VERDICT: PASS/FAIL in the report (contract: end
@@ -231,7 +231,6 @@ function buildVerdictReaskPrompt(stepName, workerText) {
     "Decide from the report's own content whether the " + stepName + " step clearly describes successful completion: " +
     "if it does, the verdict is PASS; otherwise — failure, error, unfinished work, or unclear — the verdict is FAIL. " +
     "Do NOT copy any VERDICT line from the report — decide from the content.\n\n" +
-    "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your verdict line here\"}. " +
     "The result must be exactly one line and nothing else: VERDICT: PASS or VERDICT: FAIL.";
 }
 async function reaskVerdict(stepName, reworkSuffix, workerText) {
@@ -304,8 +303,8 @@ var TOOL_CHECK_PREAMBLE =
   "3. Run the shell command: echo tool-probe-ok - then write exactly one line: shell_transport: ok - or shell_transport: unavailable if you cannot run shell commands.\n" +
   "Then do the assignment below.\n\n";
 function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason) {
-  // reason: "discarded" (the runtime threw the output away - it could not be
-  // machine-read), "empty" (agent() returned without throwing but produced
+  // reason: "discarded" (the runtime threw the output away — the JSON-candidate
+  // scan found {...}-shaped fragments it could not parse), "empty" (agent() returned without throwing but produced
   // nothing usable), "no-tools" (the worker's TOOL CHECK reported
   // artifact_tools: missing), or "no-transport" (the worker's TOOL CHECK
   // reported shell_transport: unavailable).
@@ -316,11 +315,11 @@ function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason)
     ? "your previous attempt's TOOL CHECK reported artifact_tools: missing (this is a fresh launch, so run the TOOL CHECK's load step again before the work)"
     : reason === "no-transport"
     ? "your previous attempt's TOOL CHECK reported shell_transport: unavailable (this is a fresh launch, so run the TOOL CHECK's shell probe again before the work)"
-    : "your previous attempt's output could not be machine-read as JSON and was discarded";
+    : "your previous attempt's output was discarded by the transport because it contained {...}-shaped fragments the transport could not parse; write plain prose with no JSON-shaped fragments";
   return "\n\nTRANSPORT RETRY (attempt " + attempt + " of 2): " + why + ". " +
     "First check existing state (worktree/branch at " + repoPath + "/.worktrees/" + taskId + ", the task branch, dashboard sessions for this task) - " +
     "if the " + stepName + " work is already complete, report on what was done rather than duplicating side effects. " +
-    "Then return your report as JSON in exactly the shape specified above.";
+    "Then return your report in exactly the shape specified above.";
 }
 // parseClassifyFerry — room #26 blocker 42 (2026-09-21). Shape-constrained
 // ferry parser for the classify-branch agent() call. The schema buys SHAPE,
@@ -1900,12 +1899,12 @@ while (i < STEPS.length) {
       "The returned events are filtered to this task. They contain notes and decisions from prior phases.\n\n";
   }
 
-  // Work agent returns the runtime's native envelope {"status": "ok",
-  // "result": "<prose>"} with no schema. The runtime requires JSON output;
-  // the envelope is its own documented shape, so there is nothing for the
-  // agent to improvise. The workflow receives the prose report as a plain
-  // string. The verdict is still extracted deterministically from the report
-  // text by extractVerdict below — never by an agent.
+  // Work agent returns prose on the schema-less courier; the workflow
+  // consumes the return as a plain string (blocker 43, 2026-09-22: the
+  // prompt names no JSON envelope — nothing parses one, and the runtime's
+  // JSON-candidate scan discards brace-shaped prose, which is what the
+  // transport retry guards). The verdict is still extracted deterministically
+  // from the report text by extractVerdict below — never by an agent.
   var workPromptBase =
     TOOL_CHECK_PREAMBLE +
     "Read the identity file at " + ORCH_PATH + "/identities/" + step.identity + ".md using the read tool, and embody that character fully.\n\n" +
@@ -1913,8 +1912,7 @@ while (i < STEPS.length) {
     "Task: " + taskTitle + "\nTask ID: " + taskId + "\nDescription: " + taskDescription + "\nStep: " + step.name + "\n" +
     (step.name !== "Review" ? "Crew API: " + CREW_API + "\n" : "") +
     "\n## Instructions\n\n" + eventPreamble + instructions + "\n\nCONSTRAINT: Do NOT call logevent or upsertagentsession — the workflow handles all phase tracking after your step completes.\n\nStay in character. Do the work thoroughly.\n\n" +
-    "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your report here\"}. " +
-    "The result is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
+    "Your report is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
   var workKeyBase = "work-" + step.name + (reworkCount > 0 ? "-r" + reworkCount : "");
   var workerResult = null;
   var workAttempts = [];

package/workflows/docs.js

@@ -269,7 +269,6 @@ while (i < STEPS.length) {
       "Decide from the report's own content whether the " + stepName + " step clearly describes successful completion: " +
       "if it does, the verdict is PASS; otherwise — failure, error, unfinished work, or unclear — the verdict is FAIL. " +
       "Do NOT copy any VERDICT line from the report — decide from the content.\n\n" +
-      "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your verdict line here\"}. " +
       "The result must be exactly one line and nothing else: VERDICT: PASS or VERDICT: FAIL.";
   }
   async function reaskVerdict(stepName, reworkSuffix, workerText) {
@@ -329,8 +328,8 @@ while (i < STEPS.length) {
     "3. Run the shell command: echo tool-probe-ok - then write exactly one line: shell_transport: ok - or shell_transport: unavailable if you cannot run shell commands.\n" +
     "Then do the assignment below.\n\n";
   function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason) {
-    // reason: "discarded" (the runtime threw the output away - it could not be
-    // machine-read), "empty" (agent() returned without throwing but produced
+    // reason: "discarded" (the runtime threw the output away — the JSON-candidate
+    // scan found {...}-shaped fragments it could not parse), "empty" (agent() returned without throwing but produced
     // nothing usable), "no-tools" (the worker's TOOL CHECK reported
     // artifact_tools: missing), or "no-transport" (the worker's TOOL CHECK
     // reported shell_transport: unavailable).
@@ -341,11 +340,11 @@ while (i < STEPS.length) {
       ? "your previous attempt's TOOL CHECK reported artifact_tools: missing (this is a fresh launch, so run the TOOL CHECK's load step again before the work)"
       : reason === "no-transport"
       ? "your previous attempt's TOOL CHECK reported shell_transport: unavailable (this is a fresh launch, so run the TOOL CHECK's shell probe again before the work)"
-      : "your previous attempt's output could not be machine-read as JSON and was discarded";
+      : "your previous attempt's output was discarded by the transport because it contained {...}-shaped fragments the transport could not parse; write plain prose with no JSON-shaped fragments";
     return "\n\nTRANSPORT RETRY (attempt " + attempt + " of 2): " + why + ". " +
       "First check existing state (worktree/branch at " + repoPath + "/.worktrees/" + taskId + ", the task branch, dashboard sessions for this task) - " +
       "if the " + stepName + " work is already complete, report on what was done rather than duplicating side effects. " +
-      "Then return your report as JSON in exactly the shape specified above.";
+      "Then return your report in exactly the shape specified above.";
   }
   // parseToolSignals - bug 3472bf36. The work agent's TOOL CHECK emits two
   // exact signal lines: artifact_tools: ok|missing and
@@ -432,20 +431,19 @@ function describeWorkAgentFailure(stepName, identity, attempts) {
       "\nWrite your review as plain prose — findings, then decision. End your report with exactly one line: VERDICT: PASS if it passes, VERDICT: FAIL if it fails.";
   }
 
-  // Work agent returns the runtime's native envelope {"status": "ok",
-  // "result": "<prose>"} with no schema. The runtime requires JSON output;
-  // the envelope is its own documented shape, so there is nothing for the
-  // agent to improvise. The workflow receives the prose report as a plain
-  // string. The verdict is still extracted deterministically from the report
-  // text by extractVerdict below — never by an agent.
+  // Work agent returns prose on the schema-less courier; the workflow
+  // consumes the return as a plain string (blocker 43, 2026-09-22: the
+  // prompt names no JSON envelope — nothing parses one, and the runtime's
+  // JSON-candidate scan discards brace-shaped prose, which is what the
+  // transport retry guards). The verdict is still extracted deterministically
+  // from the report text by extractVerdict below — never by an agent.
   var workPromptBase =
     TOOL_CHECK_PREAMBLE +
     "Read the identity file at " + ORCH_PATH + "/identities/" + step.identity + ".md using the read tool, and embody that character fully.\n\n" +
     "## Your Assignment\n\n" +
     "Task: " + taskTitle + "\nTask ID: " + taskId + "\nDescription: " + taskDescription + "\nStep: " + step.name + "\n\n" +
     "## Instructions\n\n" + instructions + "\n\nCONSTRAINT: Do NOT call the crew API directly — the workflow handles all phase tracking after your step completes.\n\nStay in character. All file work under " + REPO_PATH + "/.\n\n" +
-    "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your report here\"}. " +
-    "The result is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
+    "Your report is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
   var workKeyBase = "work-" + step.name + (reworkCount > 0 ? "-r" + reworkCount : "");
   var workerResult = null;
   var workAttempts = [];

package/workflows/standard.js

@@ -139,7 +139,7 @@ if (!taskId) {
   throw new Error("task_id is required in args");
 }
 
-// See docs/decisions/workflow-core.md#closeout-envelope: the work agent returns the runtime's native envelope; verdict extracted mechanically.
+// See docs/decisions/workflow-core.md#closeout-envelope: the schema-less work courier returns prose, consumed as a plain string; verdict extracted mechanically.
 const VERDICT_STEPS = ["Build", "Review", "QA", "Reproduce", "Integrate", "Publish"];
 function extractVerdict(workerText) {
   // The verdict is the LAST VERDICT: PASS/FAIL in the report (contract: end
@@ -229,7 +229,6 @@ function buildVerdictReaskPrompt(stepName, workerText) {
     "Decide from the report's own content whether the " + stepName + " step clearly describes successful completion: " +
     "if it does, the verdict is PASS; otherwise — failure, error, unfinished work, or unclear — the verdict is FAIL. " +
     "Do NOT copy any VERDICT line from the report — decide from the content.\n\n" +
-    "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your verdict line here\"}. " +
     "The result must be exactly one line and nothing else: VERDICT: PASS or VERDICT: FAIL.";
 }
 async function reaskVerdict(stepName, reworkSuffix, workerText) {
@@ -302,8 +301,8 @@ var TOOL_CHECK_PREAMBLE =
   "3. Run the shell command: echo tool-probe-ok - then write exactly one line: shell_transport: ok - or shell_transport: unavailable if you cannot run shell commands.\n" +
   "Then do the assignment below.\n\n";
 function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason) {
-  // reason: "discarded" (the runtime threw the output away - it could not be
-  // machine-read), "empty" (agent() returned without throwing but produced
+  // reason: "discarded" (the runtime threw the output away — the JSON-candidate
+  // scan found {...}-shaped fragments it could not parse), "empty" (agent() returned without throwing but produced
   // nothing usable), "no-tools" (the worker's TOOL CHECK reported
   // artifact_tools: missing), or "no-transport" (the worker's TOOL CHECK
   // reported shell_transport: unavailable).
@@ -314,11 +313,11 @@ function buildTransportRetryTrailer(stepName, repoPath, taskId, attempt, reason)
     ? "your previous attempt's TOOL CHECK reported artifact_tools: missing (this is a fresh launch, so run the TOOL CHECK's load step again before the work)"
     : reason === "no-transport"
     ? "your previous attempt's TOOL CHECK reported shell_transport: unavailable (this is a fresh launch, so run the TOOL CHECK's shell probe again before the work)"
-    : "your previous attempt's output could not be machine-read as JSON and was discarded";
+    : "your previous attempt's output was discarded by the transport because it contained {...}-shaped fragments the transport could not parse; write plain prose with no JSON-shaped fragments";
   return "\n\nTRANSPORT RETRY (attempt " + attempt + " of 2): " + why + ". " +
     "First check existing state (worktree/branch at " + repoPath + "/.worktrees/" + taskId + ", the task branch, dashboard sessions for this task) - " +
     "if the " + stepName + " work is already complete, report on what was done rather than duplicating side effects. " +
-    "Then return your report as JSON in exactly the shape specified above.";
+    "Then return your report in exactly the shape specified above.";
 }
 // parseClassifyFerry — room #26 blocker 42 (2026-09-21). Shape-constrained
 // ferry parser for the classify-branch agent() call. The schema buys SHAPE,
@@ -2061,12 +2060,12 @@ while (i < STEPS.length) {
       "The returned events are filtered to this task. They contain notes and decisions from prior phases.\n\n";
   }
 
-  // Work agent returns the runtime's native envelope {"status": "ok",
-  // "result": "<prose>"} with no schema. The runtime requires JSON output;
-  // the envelope is its own documented shape, so there is nothing for the
-  // agent to improvise. The workflow receives the prose report as a plain
-  // string. The verdict is still extracted deterministically from the report
-  // text by extractVerdict below — never by an agent.
+  // Work agent returns prose on the schema-less courier; the workflow
+  // consumes the return as a plain string (blocker 43, 2026-09-22: the
+  // prompt names no JSON envelope — nothing parses one, and the runtime's
+  // JSON-candidate scan discards brace-shaped prose, which is what the
+  // transport retry guards). The verdict is still extracted deterministically
+  // from the report text by extractVerdict below — never by an agent.
   var workPromptBase =
       TOOL_CHECK_PREAMBLE +
       "Read the identity file at " + ORCH_PATH + "/identities/" + step.identity + ".md using the read tool, and embody that character fully.\n\n" +
@@ -2079,8 +2078,7 @@ while (i < STEPS.length) {
       "\n## Instructions\n\n" + eventPreamble + instructions + "\n\n" +
       "CONSTRAINT: Do NOT call logevent or upsertagentsession — the workflow handles all phase tracking after your step completes.\n\n" +
       "Stay in character. Do the work thoroughly.\n\n" +
-      "Return your work as JSON in exactly this shape: {\"status\": \"ok\", \"result\": \"your report here\"}. " +
-      "The result is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
+      "Your report is plain prose describing what you did and found. For verdict steps, end the report with exactly one line: VERDICT: PASS or VERDICT: FAIL.";
   var workKeyBase = "work-" + step.name + (totalReworkCount > 0 ? "-r" + totalReworkCount : "");
   var workerResult = null;
   var workAttempts = [];