muse-crew 0.6.6 → 0.6.7

package/workflows/docs.js

@@ -22,11 +22,19 @@ const startStepIndex = inputs.start_step_index || 0;
 // resolution back via updatetask in the self-claim below.
 const RESOLVED_WORKFLOW = inputs.resolved_workflow || null;
 const WORKFLOW_WAS_NULL = inputs.workflow_was_null === true;
-const CLAIM_WORKFLOW_PERSIST = (WORKFLOW_WAS_NULL && RESOLVED_WORKFLOW) ? ", \"workflow\": \"" + RESOLVED_WORKFLOW + "\"" : "";
 
 // Config from args — backward-compatible fallbacks for manual launches
-const DASHBOARD_SLUG = inputs.dashboardSlug || "orchestra-dashboard";
 const crewHome = inputs.crewHome || "~/workspace/.jarvis";
+// Crew API: the workflow calls the crew-owned CLI, not the dashboard.
+// The CLI implements the API.md contract against $CREW_HOME/crew-state.db.
+const CREW_API = crewHome + "/current/lib/crew-api.js";
+// Build a shell command invoking the CLI. Args are JSON-encoded and
+// single-quote-wrapped for safe shell passing. The agent runs this and
+// returns the stdout verbatim (the CLI emits JSON on stdout).
+function crewCmd(command, args) {
+  var json = JSON.stringify(args || {}).replace(/'/g, "'\\''");
+  return "node " + CREW_API + " --crew-home " + crewHome + " " + command + " --json '" + json + "'";
+}
 const ORCH_PATH = crewHome + "/.orchestration";
 
 // Project repo — the docs workflow edits the project's own docs, not the crew home.
@@ -35,7 +43,12 @@ const projectConfig = inputs.project_config || {};
 // Project this run was dispatched for — the mid-run project-change guard
 // compares the task's live project against this on every phase boundary.
 const LAUNCH_PROJECT_ID = inputs.project_id || "";
-const REPO_PATH = projectConfig.repo_path || crewHome;
+// Fail closed on a missing repo_path — docs go in the project repo,
+// never in the crew home.
+const REPO_PATH = projectConfig.repo_path || "";
+if (!REPO_PATH) {
+  throw new Error("Project '" + (inputs.project_id || "unknown") + "' has no repo_path configured — set it via updateproject before dispatching tasks.");
+}
 
 if (!taskId) {
   throw new Error("task_id is required in args");
@@ -69,10 +82,9 @@ async function parkTask(reason) {
   try {
     await agent(
       "Park this task for human attention.\n" +
-      "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"parktask\", args: " +
-      JSON.stringify({ task_id: taskId, message: parkMessage }) + ".\n" +
+      "Run in shell and return the stdout verbatim:\n" + crewCmd("park-task", { task_id: taskId, message: parkMessage }) + "\n" +
       "The parked state is the human-attention signal — the dispatcher skips parked tasks.",
-      { key: "park-task", label: "Parking task for human attention", schema: { type: "object" } }
+      { key: "park-task", label: "Parking task for human attention" }
     );
   } catch (parkErr) {
     log("PARK FAILED for task " + taskId + ": " + (parkErr && parkErr.message ? parkErr.message : parkErr) + " — park did not land, reporting failed so the next tick retries");
@@ -97,8 +109,8 @@ while (i < STEPS.length) {
   // the step with the new project's config.
   if (LAUNCH_PROJECT_ID) {
     const projectCheck = await agent(
-      "Read this task's current project from the dashboard.\n" +
-      "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"getstate\", args: { \"events_limit\": 1 }.\n" +
+      "Read this task's current project from the crew API.\n" +
+      "Run in shell and return the stdout verbatim:\n" + crewCmd("get-state", { events_limit: 1 }) + "\n" +
       "Find the task with id \"" + taskId + "\" in the returned tasks array.\n" +
       "Return exactly { \"project\": \"<the task's project field, or empty string if absent>\" } and nothing else.",
       {
@@ -113,12 +125,12 @@ while (i < STEPS.length) {
       log(abortMessage);
       await agent(
         "Abort the stale run.\n" +
-        "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"upsertagentsession\", args:\n" +
-        "{ \"task_id\": \"" + taskId + "\", \"identity\": \"" + step.identity + "\", \"step\": \"" + REWORK_STEP + "\", \"status\": \"failed\", " +
-        "\"notes\": " + JSON.stringify(abortMessage) + " }.\n" +
-        "Then call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"logevent\", args:\n" +
-        "{ \"task_id\": \"" + taskId + "\", \"type\": \"failed\", \"identity\": \"" + step.identity + "\", \"message\": " + JSON.stringify(abortMessage) + " }.",
-        { key: "abort-project-change", label: "Aborting stale run (project changed)", schema: { type: "object" } }
+        "Run in shell and return the stdout verbatim:\n" + crewCmd("record-phase", {
+          task_id: taskId,
+          session: { task_id: taskId, identity: step.identity, step: REWORK_STEP, status: "failed", notes: abortMessage },
+          event: { task_id: taskId, type: "failed", identity: step.identity, message: abortMessage }
+        }) + "\n",
+        { key: "abort-project-change", label: "Aborting stale run (project changed)" }
       );
       return { status: "failed", task_id: taskId, reason: abortMessage };
     }
@@ -131,11 +143,12 @@ while (i < STEPS.length) {
   // claimed:false and this run stands down as a duplicate.
   let activeSessionId;
   if (isFirstClaim) {
+    var firstClaimUpdateArgs = { id: taskId, state: "in_progress" };
+    if (WORKFLOW_WAS_NULL && RESOLVED_WORKFLOW) firstClaimUpdateArgs.workflow = RESOLVED_WORKFLOW;
     const claimResult = await agent(
       "Claim this task for the " + step.name + " step.\n" +
-      "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"updatetask\", args: { \"id\": \"" + taskId + "\", \"state\": \"in_progress\"" + CLAIM_WORKFLOW_PERSIST + " }.\n" +
-      "Then call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"claimtask\", args:\n" +
-      "{ \"task_id\": \"" + taskId + "\", \"identity\": \"" + step.identity + "\", \"step\": \"" + step.name + "\", \"notes\": \"" + step.name + " step started\" }.\n" +
+      "Run in shell and return the stdout verbatim:\n" + crewCmd("update-task", firstClaimUpdateArgs) + "\n" +
+      "Then run in shell and return the stdout verbatim:\n" + crewCmd("claim-task", { task_id: taskId, identity: step.identity, step: step.name, notes: step.name + " step started" }) + "\n" +
       "Do not interpret the claim response. It already contains an explicit \"claimed\" field — copy it verbatim.\n" +
       "Return { claimed: <verbatim>, session_id: \"<...>\" }. If claimed is false there is no session_id; return { claimed: false, session_id: \"\" }.",
       {
@@ -156,8 +169,7 @@ while (i < STEPS.length) {
   } else {
     const claimResult = await agent(
       "Claim a session for this task step.\n" +
-      "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"claimtask\", args:\n" +
-      "{ \"task_id\": \"" + taskId + "\", \"identity\": \"" + step.identity + "\", \"step\": \"" + step.name + "\", \"notes\": \"" + step.name + " step started" + (reworkCount > 0 ? " (rework #" + reworkCount + ")" : "") + "\" }.\n" +
+      "Run in shell and return the stdout verbatim:\n" + crewCmd("claim-task", { task_id: taskId, identity: step.identity, step: step.name, notes: step.name + " step started" + (reworkCount > 0 ? " (rework #" + reworkCount + ")" : "") }) + "\n" +
       "Return the session_id from the response.",
       {
         key: "claim-" + step.name + (reworkCount > 0 ? "-r" + reworkCount : ""),
@@ -274,20 +286,58 @@ while (i < STEPS.length) {
   function workRetryKey(stepName, reworkSuffix, attempt) {
     return "work-" + stepName + reworkSuffix + "-t" + attempt;
   }
+  // TOOL_CHECK_PREAMBLE - every work agent runs this first. The artifact tool
+  // namespace is deferred for workflow children: present but invisible until
+  // the child loads it via tool_search.load_tool_namespace (bug 3472bf36 root
+  // cause, verified 2026-09-11 by direct probe: 5/5 children self-loaded it;
+  // the "non-deterministic platform flake" was children never being told to
+  // load it). The two signal lines are the ONLY machine-read tool-availability
+  // evidence - the workflow never guesses from English prose.
+  // Byte-identical across standard/bugfix/chore/docs - pinned by
+  // tests/artifact-tools.test.js.
+  var TOOL_CHECK_PREAMBLE =
+    "TOOL CHECK (do this first, before any other work):\n" +
+    "1. Call tool_search.load_tool_namespace with paths [\"artifact\"].\n" +
+    "2. Write exactly one line: artifact_tools: ok - or artifact_tools: missing if the call failed or the tool does not exist.\n" +
+    "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) or "empty" (agent() returned without throwing but produced
-    // nothing usable). The trailer tells the retry what to expect, not just to
-    // try again.
+    // reason: "discarded" (the runtime threw the output away - it could not be
+    // machine-read), "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).
+    // The trailer tells the retry what to expect, not just to try again.
     var why = reason === "empty"
       ? "your previous attempt returned no usable output"
+      : reason === "no-tools"
+      ? "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";
     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) — " +
+      "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.";
   }
-  function describeWorkAgentFailure(stepName, identity, attempts) {
+  // parseToolSignals - bug 3472bf36. The work agent's TOOL CHECK emits two
+  // exact signal lines: artifact_tools: ok|missing and
+  // shell_transport: ok|unavailable. The workflow reads ONLY these lines.
+  // It never guesses tool availability from English prose: the old regex
+  // misclassified ordinary inability-prose (e.g. a Map agent describing what
+  // it could not inspect) and burned all three attempts on useless retries.
+  // Byte-identical across standard/bugfix/chore/docs - pinned by
+  // tests/artifact-tools.test.js.
+  function parseToolSignals(text) {
+    var t = String(text || "");
+    var out = { artifactTools: "unknown", shellTransport: "unknown" };
+    var am = t.match(/^artifact_tools:\s*(ok|missing)\s*$/m);
+    if (am) out.artifactTools = am[1];
+    var sm = t.match(/^shell_transport:\s*(ok|unavailable)\s*$/m);
+    if (sm) out.shellTransport = sm[1];
+    return out;
+  }
+function describeWorkAgentFailure(stepName, identity, attempts) {
     // Honest classification of a work-agent call that yielded no usable
     // report, with the per-attempt evidence preserved in the session notes.
     // Two distinct cases:
@@ -359,10 +409,11 @@ while (i < STEPS.length) {
   // string. 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 + "\nDashboard slug: " + DASHBOARD_SLUG + "\n\n" +
-    "## Instructions\n\n" + instructions + "\n\nCONSTRAINT: Do NOT call logevent or upsertagentsession — the workflow handles all phase tracking after your step completes.\n\nStay in character. All file work under " + REPO_PATH + "/.\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.";
   var workKeyBase = "work-" + step.name + (reworkCount > 0 ? "-r" + reworkCount : "");
@@ -370,7 +421,8 @@ while (i < STEPS.length) {
   var workAttempts = [];
   for (var workAttempt = 0; workAttempt <= 2; workAttempt++) {
     var workKey = workAttempt === 0 ? workKeyBase : workRetryKey(step.name, (reworkCount > 0 ? "-r" + reworkCount : ""), workAttempt);
-    var retryReason = workAttempt === 0 ? null : (workAttempts[workAttempt - 1].threw ? "discarded" : "empty");
+    var prevAttempt = workAttempt === 0 ? null : workAttempts[workAttempt - 1];
+    var retryReason = workAttempt === 0 ? null : (prevAttempt.threw ? "discarded" : (prevAttempt.outcome === "missing-artifact-tools" ? "no-tools" : (prevAttempt.outcome === "unavailable-shell-transport" ? "no-transport" : "empty")));
     try {
       workerResult = await agent(
         workPromptBase + (workAttempt === 0 ? "" : buildTransportRetryTrailer(step.name, REPO_PATH, taskId, workAttempt, retryReason)),
@@ -385,6 +437,22 @@ while (i < STEPS.length) {
       // always fails (regression shipped in ecef136 when Date.now() was
       // removed from this loop).
       if (typeof workerResult === "string" && workerResult.trim()) {
+        // Bug 3472bf36: a worker whose TOOL CHECK reports artifact_tools: missing
+        // (or shell_transport: unavailable) gets a fresh launch - the load is
+        // per-launch - instead of a useless report.
+        var toolSignals = parseToolSignals(workerResult);
+        if (toolSignals.artifactTools === "missing") {
+          workAttempts.push({ threw: false, error: "", outcome: "missing-artifact-tools" });
+          log(step.name + " work agent attempt " + (workAttempt + 1) + " of 3 reported artifact_tools: missing — retrying with a fresh launch");
+          workerResult = null;
+          continue;
+        }
+        if (toolSignals.shellTransport === "unavailable") {
+          workAttempts.push({ threw: false, error: "", outcome: "unavailable-shell-transport" });
+          log(step.name + " work agent attempt " + (workAttempt + 1) + " of 3 reported shell_transport: unavailable — retrying with a fresh launch");
+          workerResult = null;
+          continue;
+        }
         if (workAttempt > 0) log(step.name + " work agent transport retry " + workAttempt + " returned a machine-readable report");
         break;
       }
@@ -407,11 +475,12 @@ while (i < STEPS.length) {
     log(step.name + " " + workFailure.notes + " — marking failed for retry");
     await agent(
       "Record work failure.\n" +
-      "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"upsertagentsession\", args:\n" +
-      "{ \"id\": \"" + activeSessionId + "\", \"task_id\": \"" + taskId + "\", \"identity\": \"" + step.identity + "\", \"step\": \"" + step.name + "\", \"status\": \"failed\", \"notes\": \"" + workFailure.notes + "\" }.\n" +
-      "Then call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"logevent\", args:\n" +
-      "{ \"task_id\": \"" + taskId + "\", \"type\": \"failed\", \"message\": \"" + workFailure.eventMessage + "\" }.",
-      { key: "record-block-" + step.name, label: "Recording work failure", schema: { type: "object" } }
+      "Run in shell and return the stdout verbatim:\n" + crewCmd("record-phase", {
+        task_id: taskId,
+        session: { id: activeSessionId, task_id: taskId, identity: step.identity, step: step.name, status: "failed", notes: workFailure.notes },
+        event: { task_id: taskId, type: "failed", message: workFailure.eventMessage }
+      }) + "\n",
+      { key: "record-block-" + step.name, label: "Recording work failure" }
     );
     return {
       __hatchWorkflowControl: "blocked",
@@ -441,11 +510,12 @@ while (i < STEPS.length) {
       log(step.name + " verdict re-ask exhausted — marking failed for retry");
       await agent(
         "Record verdict failure.\n" +
-        "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"upsertagentsession\", args:\n" +
-        "{ \"id\": \"" + activeSessionId + "\", \"task_id\": \"" + taskId + "\", \"identity\": \"" + step.identity + "\", \"step\": \"" + step.name + "\", \"status\": \"failed\", \"notes\": \"Worker report had no single unambiguous VERDICT: PASS/FAIL line (bounded re-ask exhausted)\" }.\n" +
-        "Then call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"logevent\", args:\n" +
-        "{ \"task_id\": \"" + taskId + "\", \"type\": \"failed\", \"message\": \"" + step.name + " verdict line missing or ambiguous, re-ask exhausted — phase failed, dispatcher will retry\" }.",
-        { key: "record-block-" + step.name, label: "Recording verdict failure", schema: { type: "object" } }
+        "Run in shell and return the stdout verbatim:\n" + crewCmd("record-phase", {
+          task_id: taskId,
+          session: { id: activeSessionId, task_id: taskId, identity: step.identity, step: step.name, status: "failed", notes: "Worker report had no single unambiguous VERDICT: PASS/FAIL line (bounded re-ask exhausted)" },
+          event: { task_id: taskId, type: "failed", message: step.name + " verdict line missing or ambiguous, re-ask exhausted — phase failed, dispatcher will retry" }
+        }) + "\n",
+        { key: "record-block-" + step.name, label: "Recording verdict failure" }
       );
       return {
         __hatchWorkflowControl: "blocked",
@@ -475,14 +545,14 @@ while (i < STEPS.length) {
 
   await agent(
     "Update session and log event.\n" +
-    "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"upsertagentsession\", args:\n" +
-    "{ \"id\": \"" + activeSessionId + "\", \"task_id\": \"" + taskId + "\", \"identity\": \"" + step.identity + "\", \"step\": \"" + step.name + "\", \"status\": \"" + status + "\", \"notes\": " + JSON.stringify(summary) + " }.\n" +
-    "Then call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"logevent\", args:\n" +
-    "{ \"task_id\": \"" + taskId + "\", \"type\": \"" + status + "\", \"identity\": \"" + step.identity + "\", \"message\": \"" + step.name + " " + status + " by " + step.identity + "\" }.",
+    "Run in shell and return the stdout verbatim:\n" + crewCmd("record-phase", {
+      task_id: taskId,
+      session: { id: activeSessionId, task_id: taskId, identity: step.identity, step: step.name, status: status, notes: summary },
+      event: { task_id: taskId, type: status, identity: step.identity, message: step.name + " " + status + " by " + step.identity }
+    }) + "\n",
     {
       key: "record-" + step.name + (reworkCount > 0 ? "-r" + reworkCount : ""),
-      label: "Recording " + step.name + " result",
-      schema: { type: "object" }
+      label: "Recording " + step.name + " result"
     }
   );
 
@@ -504,10 +574,9 @@ while (i < STEPS.length) {
 
 await agent(
   "Mark this task as done.\n" +
-  "Call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"updatetask\", args: { \"id\": \"" + taskId + "\", \"state\": \"done\" }.\n" +
-  "Then call artifact_invoke_action on slug \"" + DASHBOARD_SLUG + "\", action \"logevent\", args:\n" +
-  "{ \"task_id\": \"" + taskId + "\", \"type\": \"completed\", \"message\": \"All docs workflow steps complete.\" }.",
-  { key: "task-done", label: "Completing task: " + taskTitle, schema: { type: "object" } }
+  "Run in shell and return the stdout verbatim:\n" + crewCmd("update-task", { id: taskId, state: "done" }) + "\n" +
+  "Then run in shell and return the stdout verbatim:\n" + crewCmd("log-event", { task_id: taskId, type: "completed", message: "All docs workflow steps complete." }),
+  { key: "task-done", label: "Completing task: " + taskTitle }
 );
 
 log("Docs workflow complete for task " + taskId);