ai-whisper 0.12.1 → 0.13.0

package/README.md

@@ -100,7 +100,7 @@ whisper collab mount claude
 whisper collab mount codex
 ```
 
-The first `mount` creates the collab and starts the broker daemon for the workspace; the second binds the other agent. From either session, start a workflow against a spec or goal file — `spec-driven-development` for a spec, `ralph-loop` for an open-ended goal, plus `complex-bug-fixing` and `deliberation` (see [Workflows](docs/workflows.md)). Watch it run with:
+The first `mount` creates the collab and starts the broker daemon for the workspace; the second binds the other agent. From either session, start a workflow against a spec or goal file — `spec-driven-development` for a spec, `ralph-loop` for an open-ended goal, plus `quick-task`, `complex-bug-fixing`, and `deliberation` (see [Workflows](docs/workflows.md)). Watch it run with:
 
 ```bash
 whisper collab dashboard
@@ -136,7 +136,7 @@ It's cosmetic and entirely optional. Turn it off per mount with `--no-duo`, or p
 
 ## Learn more
 
-- [Workflows](docs/workflows.md) — how to use the four workflows well: choosing between `spec-driven-development`, `ralph-loop`, `complex-bug-fixing`, and `deliberation`, and authoring the spec, goal, bug report, or seed that drives the run.
+- [Workflows](docs/workflows.md) — how to use the five workflows well: choosing between `spec-driven-development`, `quick-task`, `ralph-loop`, `complex-bug-fixing`, and `deliberation`, and authoring the spec, task brief, goal, bug report, or seed that drives the run.
 - [Concepts](docs/concepts.md) — the mental model: baton handoff, real mounted sessions, supervised autonomy, workflow-first execution.
 - [Relay & handoff flows](docs/relay-handoff-flows.md) — the complete handoff state machine, capture-status table, hotkey reference, per-step verdicts, and troubleshooting.
 - [Evaluator configuration](docs/evaluator-configuration.md) — required credentials and options for the LLM evaluator that gates workflows.

package/dist/bin/broker-daemon.js

@@ -2808,6 +2808,47 @@ var DELIBERATION = {
     }
   ]
 };
+var QUICK_TASK_KICKOFF = `You are the implementer in an autonomous quick-task workflow. No human is in the loop \u2014 never ask for confirmation, permission, or clarification; do the work yourself.
+
+Read the task brief at {specPath}. It was written after a human approved the approach \u2014 treat it as the ratified contract: implement the "Approved approach" section as written; do NOT redesign it.
+
+1. The files under "Scope" are the declared blast radius. Implement the task, add or adjust the tests the brief's "Acceptance checks" imply, run the project's verification command and ensure it passes.
+2. Commit your changes (do NOT commit the task brief; .ai-whisper/ is gitignored).
+
+Scope guard: if the work is materially bigger than the brief \u2014 you need to touch non-test files beyond the "Scope" list, the approved approach does not survive contact with the code, or a hidden dependency/migration appears \u2014 do NOT push through. Hand back that you CANNOT PROCEED, naming what exploded; the run halts to the human, who re-scopes or escalates to spec-driven-development.
+
+Hand back the commit SHAs and the verification output, plus a 1-2 sentence summary; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.`;
+var QUICK_TASK_REVIEW = `The implementer claims the task in the brief at {specPath} is complete \u2014 the changes are commits {commitRange}. The upper bound is a LIVE \`HEAD\`: resolve it against the current repository at review time and INCLUDE any commits added during this review round (e.g. fixes for your prior findings); do not pin the review to an earlier tip. Verify against the brief's "Acceptance checks" and confirm the change stays inside the brief's "Scope": touching non-test files beyond the declared Scope list is a blocking finding \u2014 the implementer should have halted instead. Run the project's verification/tests yourself. This is an autonomous workflow with no human in the loop.
+
+` + CODE_REVIEW_SKILL_GUIDANCE + WORKFLOW_REVIEW_PROTOCOL;
+var QUICK_TASK_FIX = "Apply the reviewer's findings to commits {commitRange} now (amend or add commits as needed). This is an autonomous workflow \u2014 no human will respond. Do the work yourself and hand back the updated commit SHAs and verification output; never ask for confirmation, permission, or clarification. End your handback with a 1-2 sentence summary of what you changed; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.";
+var QUICK_TASK = {
+  type: "quick-task",
+  displayName: "Quick Task",
+  description: "Human-approved task brief \u2192 implement \u2192 acceptance review, hard-gated to small pre-scoped tasks",
+  defaultImplementer: "claude",
+  defaultReviewer: "codex",
+  phases: [
+    {
+      name: "implement-and-review",
+      implementerRole: "implementer",
+      reviewerRole: "reviewer",
+      maxRounds: 5,
+      initialHandoffStep: "implement",
+      kickoffTemplate: QUICK_TASK_KICKOFF,
+      stepTemplates: {
+        implement: QUICK_TASK_KICKOFF,
+        review: QUICK_TASK_REVIEW,
+        fix: QUICK_TASK_FIX
+      },
+      reviewMode: "acceptance-review",
+      evaluatorPromptKey: "review-loop",
+      artifactOut: { kind: "commit-range" },
+      anchorCommitBaseOnEntry: true,
+      renderFixTemplateOnFindings: true
+    }
+  ]
+};
 function withOperatorControl(def) {
   return {
     ...def,
@@ -2823,7 +2864,8 @@ var REGISTRY = {
   [SPEC_DRIVEN_DEVELOPMENT.type]: withOperatorControl(SPEC_DRIVEN_DEVELOPMENT),
   [RALPH_LOOP.type]: withOperatorControl(RALPH_LOOP),
   [COMPLEX_BUG_FIXING.type]: withOperatorControl(COMPLEX_BUG_FIXING),
-  [DELIBERATION.type]: withOperatorControl(DELIBERATION)
+  [DELIBERATION.type]: withOperatorControl(DELIBERATION),
+  [QUICK_TASK.type]: withOperatorControl(QUICK_TASK)
 };
 function ralphRunDir(workspaceRoot, workflowId) {
   return join(workspaceRoot, ".ai-whisper", "ralph", workflowId);

package/dist/bin/companion-agent.js

@@ -2831,6 +2831,47 @@ var DELIBERATION = {
     }
   ]
 };
+var QUICK_TASK_KICKOFF = `You are the implementer in an autonomous quick-task workflow. No human is in the loop \u2014 never ask for confirmation, permission, or clarification; do the work yourself.
+
+Read the task brief at {specPath}. It was written after a human approved the approach \u2014 treat it as the ratified contract: implement the "Approved approach" section as written; do NOT redesign it.
+
+1. The files under "Scope" are the declared blast radius. Implement the task, add or adjust the tests the brief's "Acceptance checks" imply, run the project's verification command and ensure it passes.
+2. Commit your changes (do NOT commit the task brief; .ai-whisper/ is gitignored).
+
+Scope guard: if the work is materially bigger than the brief \u2014 you need to touch non-test files beyond the "Scope" list, the approved approach does not survive contact with the code, or a hidden dependency/migration appears \u2014 do NOT push through. Hand back that you CANNOT PROCEED, naming what exploded; the run halts to the human, who re-scopes or escalates to spec-driven-development.
+
+Hand back the commit SHAs and the verification output, plus a 1-2 sentence summary; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.`;
+var QUICK_TASK_REVIEW = `The implementer claims the task in the brief at {specPath} is complete \u2014 the changes are commits {commitRange}. The upper bound is a LIVE \`HEAD\`: resolve it against the current repository at review time and INCLUDE any commits added during this review round (e.g. fixes for your prior findings); do not pin the review to an earlier tip. Verify against the brief's "Acceptance checks" and confirm the change stays inside the brief's "Scope": touching non-test files beyond the declared Scope list is a blocking finding \u2014 the implementer should have halted instead. Run the project's verification/tests yourself. This is an autonomous workflow with no human in the loop.
+
+` + CODE_REVIEW_SKILL_GUIDANCE + WORKFLOW_REVIEW_PROTOCOL;
+var QUICK_TASK_FIX = "Apply the reviewer's findings to commits {commitRange} now (amend or add commits as needed). This is an autonomous workflow \u2014 no human will respond. Do the work yourself and hand back the updated commit SHAs and verification output; never ask for confirmation, permission, or clarification. End your handback with a 1-2 sentence summary of what you changed; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.";
+var QUICK_TASK = {
+  type: "quick-task",
+  displayName: "Quick Task",
+  description: "Human-approved task brief \u2192 implement \u2192 acceptance review, hard-gated to small pre-scoped tasks",
+  defaultImplementer: "claude",
+  defaultReviewer: "codex",
+  phases: [
+    {
+      name: "implement-and-review",
+      implementerRole: "implementer",
+      reviewerRole: "reviewer",
+      maxRounds: 5,
+      initialHandoffStep: "implement",
+      kickoffTemplate: QUICK_TASK_KICKOFF,
+      stepTemplates: {
+        implement: QUICK_TASK_KICKOFF,
+        review: QUICK_TASK_REVIEW,
+        fix: QUICK_TASK_FIX
+      },
+      reviewMode: "acceptance-review",
+      evaluatorPromptKey: "review-loop",
+      artifactOut: { kind: "commit-range" },
+      anchorCommitBaseOnEntry: true,
+      renderFixTemplateOnFindings: true
+    }
+  ]
+};
 function withOperatorControl(def) {
   return {
     ...def,
@@ -2846,7 +2887,8 @@ var REGISTRY = {
   [SPEC_DRIVEN_DEVELOPMENT.type]: withOperatorControl(SPEC_DRIVEN_DEVELOPMENT),
   [RALPH_LOOP.type]: withOperatorControl(RALPH_LOOP),
   [COMPLEX_BUG_FIXING.type]: withOperatorControl(COMPLEX_BUG_FIXING),
-  [DELIBERATION.type]: withOperatorControl(DELIBERATION)
+  [DELIBERATION.type]: withOperatorControl(DELIBERATION),
+  [QUICK_TASK.type]: withOperatorControl(QUICK_TASK)
 };
 function ralphRunDir(workspaceRoot, workflowId) {
   return join(workspaceRoot, ".ai-whisper", "ralph", workflowId);

package/dist/bin/relay-monitor.js

@@ -2805,6 +2805,47 @@ var DELIBERATION = {
     }
   ]
 };
+var QUICK_TASK_KICKOFF = `You are the implementer in an autonomous quick-task workflow. No human is in the loop \u2014 never ask for confirmation, permission, or clarification; do the work yourself.
+
+Read the task brief at {specPath}. It was written after a human approved the approach \u2014 treat it as the ratified contract: implement the "Approved approach" section as written; do NOT redesign it.
+
+1. The files under "Scope" are the declared blast radius. Implement the task, add or adjust the tests the brief's "Acceptance checks" imply, run the project's verification command and ensure it passes.
+2. Commit your changes (do NOT commit the task brief; .ai-whisper/ is gitignored).
+
+Scope guard: if the work is materially bigger than the brief \u2014 you need to touch non-test files beyond the "Scope" list, the approved approach does not survive contact with the code, or a hidden dependency/migration appears \u2014 do NOT push through. Hand back that you CANNOT PROCEED, naming what exploded; the run halts to the human, who re-scopes or escalates to spec-driven-development.
+
+Hand back the commit SHAs and the verification output, plus a 1-2 sentence summary; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.`;
+var QUICK_TASK_REVIEW = `The implementer claims the task in the brief at {specPath} is complete \u2014 the changes are commits {commitRange}. The upper bound is a LIVE \`HEAD\`: resolve it against the current repository at review time and INCLUDE any commits added during this review round (e.g. fixes for your prior findings); do not pin the review to an earlier tip. Verify against the brief's "Acceptance checks" and confirm the change stays inside the brief's "Scope": touching non-test files beyond the declared Scope list is a blocking finding \u2014 the implementer should have halted instead. Run the project's verification/tests yourself. This is an autonomous workflow with no human in the loop.
+
+` + CODE_REVIEW_SKILL_GUIDANCE + WORKFLOW_REVIEW_PROTOCOL;
+var QUICK_TASK_FIX = "Apply the reviewer's findings to commits {commitRange} now (amend or add commits as needed). This is an autonomous workflow \u2014 no human will respond. Do the work yourself and hand back the updated commit SHAs and verification output; never ask for confirmation, permission, or clarification. End your handback with a 1-2 sentence summary of what you changed; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.";
+var QUICK_TASK = {
+  type: "quick-task",
+  displayName: "Quick Task",
+  description: "Human-approved task brief \u2192 implement \u2192 acceptance review, hard-gated to small pre-scoped tasks",
+  defaultImplementer: "claude",
+  defaultReviewer: "codex",
+  phases: [
+    {
+      name: "implement-and-review",
+      implementerRole: "implementer",
+      reviewerRole: "reviewer",
+      maxRounds: 5,
+      initialHandoffStep: "implement",
+      kickoffTemplate: QUICK_TASK_KICKOFF,
+      stepTemplates: {
+        implement: QUICK_TASK_KICKOFF,
+        review: QUICK_TASK_REVIEW,
+        fix: QUICK_TASK_FIX
+      },
+      reviewMode: "acceptance-review",
+      evaluatorPromptKey: "review-loop",
+      artifactOut: { kind: "commit-range" },
+      anchorCommitBaseOnEntry: true,
+      renderFixTemplateOnFindings: true
+    }
+  ]
+};
 function withOperatorControl(def) {
   return {
     ...def,
@@ -2820,7 +2861,8 @@ var REGISTRY = {
   [SPEC_DRIVEN_DEVELOPMENT.type]: withOperatorControl(SPEC_DRIVEN_DEVELOPMENT),
   [RALPH_LOOP.type]: withOperatorControl(RALPH_LOOP),
   [COMPLEX_BUG_FIXING.type]: withOperatorControl(COMPLEX_BUG_FIXING),
-  [DELIBERATION.type]: withOperatorControl(DELIBERATION)
+  [DELIBERATION.type]: withOperatorControl(DELIBERATION),
+  [QUICK_TASK.type]: withOperatorControl(QUICK_TASK)
 };
 function ralphRunDir(workspaceRoot, workflowId) {
   return join(workspaceRoot, ".ai-whisper", "ralph", workflowId);

package/dist/bin/whisper.js

@@ -2890,7 +2890,7 @@ function derivePlanPath(specPath, dateIso) {
   const date = dateIso.slice(0, 10);
   return `docs/superpowers/plans/${date}-${slug}.md`;
 }
-var RALPH_ITEM_DELIVERED_MARKER, RALPH_GOAL_COMPLETE_MARKER, WORKFLOW_REVIEW_PROTOCOL, WORKFLOW_OPERATOR_CONTROL, WORKFLOW_DIAGNOSIS_PROTOCOL, WORKFLOW_DELIBERATION_PROTOCOL, CODE_REVIEW_SKILL_GUIDANCE, PLAN_EXECUTION_SKILL_GUIDANCE, DELIBERATION_CRAFT_SKILL_GUIDANCE, SDD_SPEC_REVIEW, SDD_CODE_REVIEW, SDD_PLAN_EXECUTION, SPEC_DRIVEN_DEVELOPMENT, RALPH_KICKOFF, RALPH_FIX, RALPH_ITEM_REVIEW, RALPH_ACCEPTANCE_REVIEW, RALPH_LOOP, BUGFIX_DIAGNOSIS_KICKOFF, BUGFIX_DIAGNOSIS_REVIEW, BUGFIX_DIAGNOSIS_FIX, BUGFIX_FIX_KICKOFF, BUGFIX_FIX_REVIEW, BUGFIX_FIX_FIX, BUGFIX_POSTMORTEM_KICKOFF, BUGFIX_POSTMORTEM_REVIEW, BUGFIX_POSTMORTEM_FIX, COMPLEX_BUG_FIXING, DELIB_REVIEW, DELIB_SYNTHESIS_REVIEW, DELIB_FIX, DELIB_OBJECTIVES, DELIB_APPROACHES, DELIB_TRADEOFFS, DELIB_SYNTHESIS_FIX, DELIB_SYNTHESIS, DELIBERATION, REGISTRY;
+var RALPH_ITEM_DELIVERED_MARKER, RALPH_GOAL_COMPLETE_MARKER, WORKFLOW_REVIEW_PROTOCOL, WORKFLOW_OPERATOR_CONTROL, WORKFLOW_DIAGNOSIS_PROTOCOL, WORKFLOW_DELIBERATION_PROTOCOL, CODE_REVIEW_SKILL_GUIDANCE, PLAN_EXECUTION_SKILL_GUIDANCE, DELIBERATION_CRAFT_SKILL_GUIDANCE, SDD_SPEC_REVIEW, SDD_CODE_REVIEW, SDD_PLAN_EXECUTION, SPEC_DRIVEN_DEVELOPMENT, RALPH_KICKOFF, RALPH_FIX, RALPH_ITEM_REVIEW, RALPH_ACCEPTANCE_REVIEW, RALPH_LOOP, BUGFIX_DIAGNOSIS_KICKOFF, BUGFIX_DIAGNOSIS_REVIEW, BUGFIX_DIAGNOSIS_FIX, BUGFIX_FIX_KICKOFF, BUGFIX_FIX_REVIEW, BUGFIX_FIX_FIX, BUGFIX_POSTMORTEM_KICKOFF, BUGFIX_POSTMORTEM_REVIEW, BUGFIX_POSTMORTEM_FIX, COMPLEX_BUG_FIXING, DELIB_REVIEW, DELIB_SYNTHESIS_REVIEW, DELIB_FIX, DELIB_OBJECTIVES, DELIB_APPROACHES, DELIB_TRADEOFFS, DELIB_SYNTHESIS_FIX, DELIB_SYNTHESIS, DELIBERATION, QUICK_TASK_KICKOFF, QUICK_TASK_REVIEW, QUICK_TASK_FIX, QUICK_TASK, REGISTRY;
 var init_workflow_registry = __esm({
   "../broker/dist/runtime/workflow-registry.js"() {
     "use strict";
@@ -3344,11 +3344,53 @@ ${DELIBERATION_CRAFT_SKILL_GUIDANCE}End your handback with a 1-2 sentence summar
         }
       ]
     };
+    QUICK_TASK_KICKOFF = `You are the implementer in an autonomous quick-task workflow. No human is in the loop \u2014 never ask for confirmation, permission, or clarification; do the work yourself.
+
+Read the task brief at {specPath}. It was written after a human approved the approach \u2014 treat it as the ratified contract: implement the "Approved approach" section as written; do NOT redesign it.
+
+1. The files under "Scope" are the declared blast radius. Implement the task, add or adjust the tests the brief's "Acceptance checks" imply, run the project's verification command and ensure it passes.
+2. Commit your changes (do NOT commit the task brief; .ai-whisper/ is gitignored).
+
+Scope guard: if the work is materially bigger than the brief \u2014 you need to touch non-test files beyond the "Scope" list, the approved approach does not survive contact with the code, or a hidden dependency/migration appears \u2014 do NOT push through. Hand back that you CANNOT PROCEED, naming what exploded; the run halts to the human, who re-scopes or escalates to spec-driven-development.
+
+Hand back the commit SHAs and the verification output, plus a 1-2 sentence summary; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.`;
+    QUICK_TASK_REVIEW = `The implementer claims the task in the brief at {specPath} is complete \u2014 the changes are commits {commitRange}. The upper bound is a LIVE \`HEAD\`: resolve it against the current repository at review time and INCLUDE any commits added during this review round (e.g. fixes for your prior findings); do not pin the review to an earlier tip. Verify against the brief's "Acceptance checks" and confirm the change stays inside the brief's "Scope": touching non-test files beyond the declared Scope list is a blocking finding \u2014 the implementer should have halted instead. Run the project's verification/tests yourself. This is an autonomous workflow with no human in the loop.
+
+` + CODE_REVIEW_SKILL_GUIDANCE + WORKFLOW_REVIEW_PROTOCOL;
+    QUICK_TASK_FIX = "Apply the reviewer's findings to commits {commitRange} now (amend or add commits as needed). This is an autonomous workflow \u2014 no human will respond. Do the work yourself and hand back the updated commit SHAs and verification output; never ask for confirmation, permission, or clarification. End your handback with a 1-2 sentence summary of what you changed; your reply must be at least two sentences, well over 100 characters \u2014 never hand back only a single word.";
+    QUICK_TASK = {
+      type: "quick-task",
+      displayName: "Quick Task",
+      description: "Human-approved task brief \u2192 implement \u2192 acceptance review, hard-gated to small pre-scoped tasks",
+      defaultImplementer: "claude",
+      defaultReviewer: "codex",
+      phases: [
+        {
+          name: "implement-and-review",
+          implementerRole: "implementer",
+          reviewerRole: "reviewer",
+          maxRounds: 5,
+          initialHandoffStep: "implement",
+          kickoffTemplate: QUICK_TASK_KICKOFF,
+          stepTemplates: {
+            implement: QUICK_TASK_KICKOFF,
+            review: QUICK_TASK_REVIEW,
+            fix: QUICK_TASK_FIX
+          },
+          reviewMode: "acceptance-review",
+          evaluatorPromptKey: "review-loop",
+          artifactOut: { kind: "commit-range" },
+          anchorCommitBaseOnEntry: true,
+          renderFixTemplateOnFindings: true
+        }
+      ]
+    };
     REGISTRY = {
       [SPEC_DRIVEN_DEVELOPMENT.type]: withOperatorControl(SPEC_DRIVEN_DEVELOPMENT),
       [RALPH_LOOP.type]: withOperatorControl(RALPH_LOOP),
       [COMPLEX_BUG_FIXING.type]: withOperatorControl(COMPLEX_BUG_FIXING),
-      [DELIBERATION.type]: withOperatorControl(DELIBERATION)
+      [DELIBERATION.type]: withOperatorControl(DELIBERATION),
+      [QUICK_TASK.type]: withOperatorControl(QUICK_TASK)
     };
   }
 });
@@ -6755,6 +6797,97 @@ var init_workflow_event_bridge = __esm({
   }
 });
 
+// ../broker/dist/runtime/task-brief.js
+function isTestFilePath(path4) {
+  if (/(^|\/)(test|tests|__tests__)\//.test(path4))
+    return true;
+  const basename4 = path4.slice(path4.lastIndexOf("/") + 1);
+  if (basename4.includes(".test.") || basename4.includes(".spec."))
+    return true;
+  if (/_test\.[^.]+$/.test(basename4))
+    return true;
+  return false;
+}
+function findHeadings(lines) {
+  const headings = [];
+  for (let lineIndex = 0; lineIndex < lines.length; lineIndex++) {
+    const match = /^(#{1,6})\s(.*)$/.exec(lines[lineIndex]);
+    if (match) {
+      headings.push({ lineIndex, level: match[1].length, text: match[2].trim() });
+    }
+  }
+  return headings;
+}
+function sectionBody(lines, heading, allHeadings) {
+  const next = allHeadings.find((h) => h.lineIndex > heading.lineIndex);
+  const end = next ? next.lineIndex : lines.length;
+  return lines.slice(heading.lineIndex + 1, end);
+}
+function isNonEmptyBody(body) {
+  return body.some((line) => line.trim().length > 0);
+}
+function stripEnclosingBackticks(token) {
+  if (token.length >= 2 && token.startsWith("`") && token.endsWith("`")) {
+    return token.slice(1, -1);
+  }
+  return token;
+}
+function parseScopeBullets(body) {
+  const paths = [];
+  for (const line of body) {
+    const match = /^[-*]\s+(.+)$/.exec(line);
+    if (!match)
+      continue;
+    const firstToken = match[1].trim().split(/\s+/)[0];
+    paths.push(stripEnclosingBackticks(firstToken));
+  }
+  return paths;
+}
+function validateTaskBrief(content) {
+  const lines = content.split(/\r?\n/);
+  const headings = findHeadings(lines);
+  const violations = [];
+  let scopeBody = null;
+  for (const { key, canonical } of REQUIRED_SECTIONS) {
+    const heading = headings.find((h) => (h.level === 2 || h.level === 3) && h.text.toLowerCase() === key);
+    const body = heading ? sectionBody(lines, heading, headings) : null;
+    if (!heading || !isNonEmptyBody(body)) {
+      violations.push(`missing or empty required section "${canonical}"`);
+      continue;
+    }
+    if (key === "scope") {
+      scopeBody = body;
+    }
+  }
+  const scopeFiles = scopeBody ? parseScopeBullets(scopeBody) : [];
+  if (scopeBody && scopeFiles.length === 0) {
+    violations.push('"## Scope" must declare the files the task touches as a bullet list (one file per bullet)');
+  }
+  const nonTestScopeCount = scopeFiles.filter((path4) => !isTestFilePath(path4)).length;
+  if (nonTestScopeCount > QUICK_TASK_SCOPE_CAP) {
+    violations.push(`scope declares ${nonTestScopeCount} non-test files; the cap is ${QUICK_TASK_SCOPE_CAP}`);
+  }
+  return {
+    ok: violations.length === 0,
+    violations,
+    scopeFiles,
+    nonTestScopeCount
+  };
+}
+var QUICK_TASK_SCOPE_CAP, REQUIRED_SECTIONS;
+var init_task_brief = __esm({
+  "../broker/dist/runtime/task-brief.js"() {
+    "use strict";
+    QUICK_TASK_SCOPE_CAP = 5;
+    REQUIRED_SECTIONS = [
+      { key: "task", canonical: "## Task" },
+      { key: "approved approach", canonical: "## Approved approach" },
+      { key: "scope", canonical: "## Scope" },
+      { key: "acceptance checks", canonical: "## Acceptance checks" }
+    ];
+  }
+});
+
 // ../broker/dist/storage/repositories/workspace-repository.js
 function upsertWorkspace(db, input) {
   db.prepare(`
@@ -6824,6 +6957,7 @@ var init_dist2 = __esm({
     init_relay_evaluator_diagnostics_repository();
     init_relay_turn_event_diagnostics_repository();
     init_workflow_registry();
+    init_task_brief();
     init_workspace_repository();
     init_session_attachment_repository();
     init_broker_daemon_repository();
@@ -14494,8 +14628,8 @@ function compareSemver(a, b) {
 var EZIO_PROVENANCE = {
   ezioCliVersion: "0.4.0",
   ezioGitSha: "77a3b48",
-  builtAt: "2026-07-03T06:53:50.854Z",
-  whisperVersion: "0.12.1"
+  builtAt: "2026-07-04T05:15:42.289Z",
+  whisperVersion: "0.13.0"
 };
 
 // src/ezio-provenance-types.ts
@@ -19220,6 +19354,7 @@ function launchSessions(input) {
 // src/commands/workflow/start.ts
 init_dist();
 init_dist2();
+import { readFileSync as readFileSync12 } from "node:fs";
 var EVALUATOR_README_HINT = 'See the README "Evaluator configuration" section for details.';
 async function runWorkflowStart(deps) {
   const daemon = getBrokerDaemonByCollab(deps.broker.db, deps.collabId);
@@ -19243,6 +19378,23 @@ async function runWorkflowStart(deps) {
       );
     }
   }
+  if (deps.workflowType === "quick-task") {
+    const read = deps.readTaskBrief ?? ((p) => readFileSync12(p, "utf8"));
+    let briefContent;
+    try {
+      briefContent = read(deps.specPath);
+    } catch {
+      throw new Error(
+        `Task brief not readable at ${deps.specPath}. Check the path and try again.`
+      );
+    }
+    const brief = validateTaskBrief(briefContent);
+    if (!brief.ok) {
+      throw new Error(
+        "Task brief failed the quick-task scope gate:\n" + brief.violations.map((v) => `  - ${v}`).join("\n") + "\nSplit the task or use spec-driven-development."
+      );
+    }
+  }
   const def = getWorkflowDefinition(deps.workflowType);
   const boundAgents = deps.broker.control.listSessionBindings(deps.collabId).filter((b) => b.bindingState === "bound").map((b) => b.agentType).filter((t) => agentTypes.includes(t));
   let resolved;
@@ -19481,7 +19633,7 @@ async function runSkillInstall(input) {
 init_dist2();
 
 // src/runtime/cli-package-info.ts
-import { readFileSync as readFileSync12 } from "node:fs";
+import { readFileSync as readFileSync13 } from "node:fs";
 import { dirname as dirname9 } from "node:path";
 import { fileURLToPath as fileURLToPath7 } from "node:url";
 var PKG_CANDIDATES = ["../../package.json", "../package.json"];
@@ -19489,7 +19641,7 @@ function readWhisperPackage() {
   for (const rel of PKG_CANDIDATES) {
     try {
       const url = new URL(rel, import.meta.url);
-      const pkg = JSON.parse(readFileSync12(url, "utf8"));
+      const pkg = JSON.parse(readFileSync13(url, "utf8"));
       if (pkg.name === "ai-whisper" && typeof pkg.version === "string") {
         return { url, version: pkg.version };
       }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-whisper",
-  "version": "0.12.1",
+  "version": "0.13.0",
   "description": "Terminal-first relay for paired AI coding agents (Claude + Codex), driven by structured workflows.",
   "license": "Apache-2.0",
   "repository": {
@@ -53,12 +53,12 @@
     "@types/react": "^19.2.14",
     "ink-testing-library": "^4.0.0",
     "@ai-whisper/adapter-ai-ezio": "0.0.0",
-    "@ai-whisper/adapter-antigravity": "0.0.0",
+    "@ai-whisper/adapter-claude": "0.0.0",
     "@ai-whisper/adapter-codex": "0.0.0",
-    "@ai-whisper/companion-core": "0.0.0",
+    "@ai-whisper/adapter-antigravity": "0.0.0",
     "@ai-whisper/broker": "0.0.0",
-    "@ai-whisper/adapter-claude": "0.0.0",
-    "@ai-whisper/shared": "0.0.0"
+    "@ai-whisper/shared": "0.0.0",
+    "@ai-whisper/companion-core": "0.0.0"
   },
   "files": [
     "dist",

package/skills/ai-whisper-bugfix/SKILL.md

@@ -46,7 +46,8 @@ Parse the JSON. The expected shape is:
   "agents": [
     { "agentType": "codex",  "bindingState": "bound" | "pending_attach" | "unbound" | null },
     { "agentType": "claude", "bindingState": "bound" | "pending_attach" | "unbound" | null },
-    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null }
+    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "agy",    "bindingState": "bound" | "pending_attach" | "unbound" | null }
   ],
   "recovery": { "state": "normal" | "recovery_required" | "recovered" },
   "evaluator": { "ready": true | false, "status": "ready" | "missing_anthropic_key" | "invalid_config" | "disabled" | "unknown" }
@@ -57,17 +58,17 @@ Required for readiness:
 - `daemon !== null`
 - `status === "active"`
 - `recovery.state === "normal"`
-- **EXACTLY TWO agents bound** — among `codex`, `claude`, and `ezio`, exactly two
-  must have `bindingState === "bound"` (the implementer + reviewer pair). **`ezio`
-  is a replacement role**: it stands in for `codex` or `claude`, so do NOT require
-  `codex` and `claude` specifically. A pair of `ezio` + `claude`, `ezio` + `codex`,
-  or `codex` + `claude` all pass. (The displaced slot reads `null`/`unbound` and
-  that is expected when `ezio` replaces it.)
+- **EXACTLY TWO agents bound** — among the supported agent types (`codex`, `claude`,
+  `ezio`, `agy`), exactly two must have `bindingState === "bound"` (the implementer +
+  reviewer pair). **`ezio` and `agy` are replacement roles**: either stands in for
+  `codex` or `claude`, so do NOT require `codex` and `claude` specifically — any
+  pair of two distinct supported agents passes. (The displaced slots read `null`/`unbound`
+  and that is expected when a replacement agent takes a seat.)
 - `evaluator.status` is NOT `"missing_anthropic_key"` or `"invalid_config"` (i.e., `ready`, `disabled`, and `unknown` all pass this gate; only the two true-misconfiguration statuses block)
 
 If the JSON has `{ "error": "no_collab_for_cwd", ... }`:
 
-> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` — in another), then re-run this skill.
+> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` / `agy` — in another), then re-run this skill.
 
 If `recovery.state === "recovery_required"`:
 
@@ -75,14 +76,14 @@ If `recovery.state === "recovery_required"`:
 
 If `recovery.state === "recovered"`:
 
-> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect codex` and `whisper collab reconnect claude`, then re-run this skill.
+> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect <agent>` for each bound agent, then re-run this skill.
 
 If FEWER than two agents are bound (count `bindingState === "bound"` across
-`codex`/`claude`/`ezio`):
+the supported agent types):
 
 > Only <N> agent(s) bound (<list bound agentTypes>). A workflow needs two — an
-> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio>`)
-> in a separate terminal, then re-run this skill. `ezio` may replace `codex` or `claude`.
+> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio|agy>`)
+> in a separate terminal, then re-run this skill. `ezio` or `agy` may replace `codex` or `claude`.
 
 (Do NOT append permission flags — mount already spawns the agent in full-permission mode; passing `--dangerously-skip-permissions` / `--dangerously-bypass-approvals-and-sandbox` again can crash the agent on a duplicate-argument error.)
 

package/skills/ai-whisper-deliberation/SKILL.md

@@ -50,7 +50,8 @@ Parse the JSON. The expected shape is:
   "agents": [
     { "agentType": "codex",  "bindingState": "bound" | "pending_attach" | "unbound" | null },
     { "agentType": "claude", "bindingState": "bound" | "pending_attach" | "unbound" | null },
-    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null }
+    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "agy",    "bindingState": "bound" | "pending_attach" | "unbound" | null }
   ],
   "recovery": { "state": "normal" | "recovery_required" | "recovered" },
   "evaluator": { "ready": true | false, "status": "ready" | "missing_anthropic_key" | "invalid_config" | "disabled" | "unknown" }
@@ -61,17 +62,17 @@ Required for readiness:
 - `daemon !== null`
 - `status === "active"`
 - `recovery.state === "normal"`
-- **EXACTLY TWO agents bound** — among `codex`, `claude`, and `ezio`, exactly two
-  must have `bindingState === "bound"` (the implementer + reviewer pair). **`ezio`
-  is a replacement role**: it stands in for `codex` or `claude`, so do NOT require
-  `codex` and `claude` specifically. A pair of `ezio` + `claude`, `ezio` + `codex`,
-  or `codex` + `claude` all pass. (The displaced slot reads `null`/`unbound` and
-  that is expected when `ezio` replaces it.)
+- **EXACTLY TWO agents bound** — among the supported agent types (`codex`, `claude`,
+  `ezio`, `agy`), exactly two must have `bindingState === "bound"` (the implementer +
+  reviewer pair). **`ezio` and `agy` are replacement roles**: either stands in for
+  `codex` or `claude`, so do NOT require `codex` and `claude` specifically — any
+  pair of two distinct supported agents passes. (The displaced slots read `null`/`unbound`
+  and that is expected when a replacement agent takes a seat.)
 - `evaluator.status` is NOT `"missing_anthropic_key"` or `"invalid_config"` (i.e., `ready`, `disabled`, and `unknown` all pass this gate; only the two true-misconfiguration statuses block)
 
 If the JSON has `{ "error": "no_collab_for_cwd", ... }`:
 
-> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` — in another), then re-run this skill.
+> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` / `agy` — in another), then re-run this skill.
 
 If `recovery.state === "recovery_required"`:
 
@@ -79,14 +80,14 @@ If `recovery.state === "recovery_required"`:
 
 If `recovery.state === "recovered"`:
 
-> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect codex` and `whisper collab reconnect claude`, then re-run this skill.
+> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect <agent>` for each bound agent, then re-run this skill.
 
 If FEWER than two agents are bound (count `bindingState === "bound"` across
-`codex`/`claude`/`ezio`):
+the supported agent types):
 
 > Only <N> agent(s) bound (<list bound agentTypes>). A workflow needs two — an
-> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio>`)
-> in a separate terminal, then re-run this skill. `ezio` may replace `codex` or `claude`.
+> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio|agy>`)
+> in a separate terminal, then re-run this skill. `ezio` or `agy` may replace `codex` or `claude`.
 
 (Do NOT append permission flags — mount already spawns the agent in full-permission mode; passing `--dangerously-skip-permissions` / `--dangerously-bypass-approvals-and-sandbox` again can crash the agent on a duplicate-argument error.)
 

package/skills/ai-whisper-plan-execution/SKILL.md

@@ -45,7 +45,7 @@ Read the plan once; extract every task with its full text. Then:
 
 Invariants: **reviewer tier ≥ implementer tier, always.** On the **2nd failed review of the same task**, bump the task implementer one tier and pass the failure history into the new dispatch.
 
-## Path B — no subagent dispatch (e.g. Codex, ezio)
+## Path B — no subagent dispatch (e.g. Codex, ezio, agy)
 
 Execute the plan inline: task-by-task in plan order, run per-task verification, commit as the plan specifies, and hand back per the workflow prompt with the mode line `Execution mode: inline (no subagent dispatch)`. Path A does not apply to you — nothing in this skill blocks or changes your handback.
 

package/skills/ai-whisper-quick-task/SKILL.md

@@ -0,0 +1,215 @@
+---
+name: ai-whisper-quick-task
+description: Kick off the quick-task workflow on a small, already-approved unit of work. Use when the user says things like "run quick task on <path or task>", "kick off quick-task with <path or task>", "/aiw-quick-task <path or task>", "$aiw-quick-task <path or task>", or otherwise asks to start the quick-task workflow on a small, ready-to-execute task.
+---
+
+# ai-whisper-quick-task
+
+Kick off the ai-whisper quick-task workflow on a small, already-approved unit of work. This skill is fire-and-forget: it verifies the collab is ready, runs `whisper workflow start`, and exits. **Do NOT continue polling or narrating after kickoff** — continuous activity from the calling agent keeps it busy, which blocks the broker's idle detection and stalls the workflow. The dashboard (`whisper collab dashboard`) is the inspection surface during the run.
+
+Unlike `ai-whisper-sdd`, which kicks off against a spec someone already wrote, this skill also **writes the task brief itself** (Steps 1–2) from an approach approved earlier in the conversation.
+
+## When to invoke
+
+Match phrases like:
+- *"run quick task on docs/notes.md"* / *"kick off quick-task with @docs/notes.md"*
+- *"run quick task: fix the off-by-one in the pagination helper"* (task described inline, no file yet)
+- *"/aiw-quick-task docs/notes.md"* (Claude picker form)
+- *"$aiw-quick-task docs/notes.md"* (Codex picker form)
+
+If the user references the task ambiguously and no approach has been discussed yet, that's fine — proceed to Step 1 and work it out there. Do not guess at scope or approach.
+
+## Steps
+
+### 1. Executability pre-check
+
+Quick-task is for work that is executable right away — a small, scoped change to a codebase you already understand, with an approach already agreed. Before doing anything else, check whether the task actually needs:
+- **research** — unknowns that require investigation before a design is even possible,
+- **unsettled design decisions** — multiple viable approaches not yet chosen between, or
+- **schema/contract migrations** — breaking changes to a shared data shape, API, or wire format.
+
+If any of these apply, REFUSE this workflow and recommend spec-driven-development (or deliberation first, if the idea itself still needs sharpening):
+
+> This needs more upfront design than quick-task is for. I'd recommend spec-driven-development instead — write a spec and kick off SDD (or run deliberation first if the idea needs sharpening).
+
+If the approach was never explicitly approved by the user in this conversation, ask ONCE for approval before proceeding:
+
+> Before I kick off quick-task: here's the approach I'd take — <approach>. Approve it and I'll write the brief and start the workflow.
+
+Never invent an approach and kick off in the same breath. The human approves the approach in chat; the workflow's implementer must not redesign it.
+
+### 2. Write or resolve the brief
+
+If the user gave a path to an existing brief, resolve it to an absolute path and verify it's a readable file via the Read tool. If not readable:
+
+> Task brief `<path>` is not readable. Check the path and try again.
+
+Otherwise, write the brief yourself from the approach approved in Step 1. Create the `.ai-whisper/tasks/` directory first if it doesn't already exist, and make sure `.ai-whisper/` is gitignored — this workflow has no broker setup step to do it for you, so do it here (create only if absent; never clobber an existing file):
+
+```bash
+mkdir -p .ai-whisper/tasks
+[ -f .ai-whisper/.gitignore ] || printf '*\n' > .ai-whisper/.gitignore
+```
+
+Then write the brief to `.ai-whisper/tasks/<YYYY-MM-DD>-<slug>.md` (slug derived from the task title, kebab-case).
+
+Embed this template VERBATIM:
+
+```markdown
+# Task: <short title>
+
+## Task
+<what + why, 2–5 lines>
+
+## Approved approach
+<the approach the human approved in chat — the implementer must not redesign it>
+
+## Scope
+- `path/to/file-a.ts`
+- `path/to/file-b.ts`
+- `test/file-a.test.ts`
+
+## Acceptance checks
+- <how the reviewer verifies: commands to run, expected behavior>
+```
+
+`whisper workflow start` enforces a hard gate on this brief: all four sections above (`## Task`, `## Approved approach`, `## Scope`, `## Acceptance checks`) are required and must be non-empty; `## Scope` must list every file the task touches, one bullet per file; and at most 5 non-test files (test files are uncounted).
+
+### 3. Verify collab readiness
+
+Run:
+
+```bash
+whisper collab status --json
+```
+
+Parse the JSON. The expected shape is:
+
+```json
+{
+  "collabId": "collab_xyz",
+  "workspaceRoot": "/path",
+  "status": "active",
+  "daemon": { "host": "127.0.0.1", "port": 4311, "pid": 12345 },
+  "agents": [
+    { "agentType": "codex",  "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "claude", "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "agy",    "bindingState": "bound" | "pending_attach" | "unbound" | null }
+  ],
+  "recovery": { "state": "normal" | "recovery_required" | "recovered" },
+  "evaluator": { "ready": true | false, "status": "ready" | "missing_anthropic_key" | "invalid_config" | "disabled" | "unknown" }
+}
+```
+
+Required for readiness:
+- `daemon !== null`
+- `status === "active"`
+- `recovery.state === "normal"`
+- **EXACTLY TWO agents bound** — among the supported agent types (`codex`, `claude`,
+  `ezio`, `agy`), exactly two must have `bindingState === "bound"` (the implementer +
+  reviewer pair). **`ezio` and `agy` are replacement roles**: either stands in for
+  `codex` or `claude`, so do NOT require `codex` and `claude` specifically — any
+  pair of two distinct supported agents passes. (The `agents` array may list all
+  supported types; the displaced slots read `null`/`unbound` and that is expected
+  when a replacement agent takes a seat.)
+- `evaluator.status` is NOT `"missing_anthropic_key"` or `"invalid_config"` (i.e., `ready`, `disabled`, and `unknown` all pass this gate; only the two true-misconfiguration statuses block)
+
+If the JSON has `{ "error": "no_collab_for_cwd", ... }`:
+
+> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` / `agy` — in another), then re-run this skill.
+
+If `recovery.state === "recovery_required"`:
+
+> The collab is in recovery_required state. Run `whisper collab recover`, then re-run this skill.
+
+If `recovery.state === "recovered"`:
+
+> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect <agent>` for each bound agent, then re-run this skill.
+
+If FEWER than two agents are bound (count `bindingState === "bound"` across
+the supported agent types):
+
+> Only <N> agent(s) bound (<list bound agentTypes>). A workflow needs two — an
+> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio|agy>`)
+> in a separate terminal, then re-run this skill. `ezio` or `agy` may replace `codex` or `claude`.
+
+(Do NOT append permission flags — mount already spawns the agent in full-permission mode; passing `--dangerously-skip-permissions` / `--dangerously-bypass-approvals-and-sandbox` again can crash the agent on a duplicate-argument error.)
+
+If `evaluator.status === "missing_anthropic_key"` (i.e., `evaluator.ready === false` AND status is `missing_anthropic_key`):
+
+> The evaluator has no Anthropic API key. Create `~/.ai-whisper/auth.json` with `{ "ANTHROPIC_API_KEY": "sk-ant-..." }` (chmod 600), then restart the daemon (`whisper collab stop` and re-mount, or restart the broker), and re-run this skill. See the README "Evaluator configuration" section.
+
+If `evaluator.status === "invalid_config"` (i.e., `evaluator.ready === false` AND status is `invalid_config`):
+
+> The evaluator config is malformed. Fix the JSON in `~/.ai-whisper/auth.json` or `~/.ai-whisper/config.json`, then restart the daemon and re-run this skill. See the README "Evaluator configuration" section.
+
+If `evaluator.status === "disabled"`: this means the orchestrator is intentionally off — it is NOT a misconfiguration and does NOT block this skill gate. Proceed to step 4; `workflow start` will surface the orchestrator-disabled error itself.
+
+(Note: `evaluator.ready` is `false` for `missing_anthropic_key`, `invalid_config`, AND `disabled`; it is `true` only for `ready` and `unknown`. That's why this gate keys off `status` rather than `ready` — so `disabled` does not block the skill while the two true-misconfiguration statuses do.)
+
+### 4. Kick off the workflow
+
+Run:
+
+```bash
+whisper workflow start --type=quick-task --spec=<resolved-absolute-path>
+```
+
+(No `--implementer` / `--reviewer` needed — the agent triggering this skill becomes the implementer and the other agent the reviewer. Pass `--implementer <agent> --reviewer <agent>` only to override.)
+
+If the scope gate rejects the brief, `whisper workflow start` exits non-zero and prints every violation together. Relay these violations to the user VERBATIM and STOP — do NOT silently rewrite or shrink the `## Scope` list to squeeze under the cap. The user decides whether to split the task into smaller pieces or escalate to spec-driven-development.
+
+Otherwise, parse the workflowId from stdout (format: `Workflow started: <workflowId>`).
+
+### 5. Report and exit
+
+Print exactly:
+
+> Workflow `<workflowId>` started. Track progress with `whisper collab dashboard`.
+
+Then stop. Do NOT poll `whisper workflow inspect`. Do NOT continue narrating. The workflow runs in the broker driver; your job is done.
+
+## Why fire-and-forget
+
+The broker's relay handoff system uses **idle detection** to know when an agent is ready to receive the next handoff. If this skill polled the workflow's status every few seconds, the calling agent (you) would emit output continuously, the broker would never see you as idle, and the workflow's first handoff couldn't be delivered to you — the workflow stalls. Kick off and exit; observation belongs to the dashboard.
+
+## Duo roleplay
+
+The mount may have assigned you a movie-duo character for this collab session — check the `AI_WHISPER_CHARACTER` / `AI_WHISPER_CHARACTER_ROLE` env vars, or the `[ai-whisper duo]` brief injected at session start, to find out. If so, staying in character is welcome — but **conversational prose only**: chat, status updates, banter with the operator or your teammate.
+
+Never let character flavor into code, commit messages, PR descriptions, or file contents. The reviewer/evaluator protocol output (verdict labels, approve/findings/escalate) stays protocol-exact regardless of who you're playing.
+
+## Resume / cancel
+
+If the user asks to resume a halted workflow, run:
+
+```bash
+whisper workflow resume <workflowId>
+```
+
+If they ask to cancel:
+
+```bash
+whisper workflow cancel <workflowId>
+```
+
+Same fire-and-forget shape: invoke, report one line, exit.
+
+## Pausing the workflow (operator control)
+
+If the user interrupts you mid-workflow and asks to pause it (e.g. "pause the workflow, I need to fix X"):
+
+1. Find the active workflow id: `whisper workflow list`.
+2. Run `whisper workflow pause <workflowId>`.
+3. Acknowledge and **stop working** — do not start the next change.
+
+The operator edits artifacts while paused, then resumes:
+
+```bash
+whisper workflow resume <workflowId> --message "what I changed and why"
+```
+
+On resume the agents receive a notice listing the changed files plus the operator note, and must re-read those files before continuing.
+
+Provider gotcha: the Codex CLI **exits its session** on Ctrl+C at an idle prompt (a mid-task Ctrl+C only interrupts the running task). The user typically interrupts a *busy* agent before issuing the pause instruction — do not assume Ctrl+C is a safe no-op.

package/skills/ai-whisper-ralph/SKILL.md

@@ -46,7 +46,8 @@ Parse the JSON. The expected shape is:
   "agents": [
     { "agentType": "codex",  "bindingState": "bound" | "pending_attach" | "unbound" | null },
     { "agentType": "claude", "bindingState": "bound" | "pending_attach" | "unbound" | null },
-    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null }
+    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "agy",    "bindingState": "bound" | "pending_attach" | "unbound" | null }
   ],
   "recovery": { "state": "normal" | "recovery_required" | "recovered" },
   "evaluator": { "ready": true | false, "status": "ready" | "missing_anthropic_key" | "invalid_config" | "disabled" | "unknown" }
@@ -57,17 +58,17 @@ Required for readiness:
 - `daemon !== null`
 - `status === "active"`
 - `recovery.state === "normal"`
-- **EXACTLY TWO agents bound** — among `codex`, `claude`, and `ezio`, exactly two
-  must have `bindingState === "bound"` (the implementer + reviewer pair). **`ezio`
-  is a replacement role**: it stands in for `codex` or `claude`, so do NOT require
-  `codex` and `claude` specifically. A pair of `ezio` + `claude`, `ezio` + `codex`,
-  or `codex` + `claude` all pass. (The displaced slot reads `null`/`unbound` and
-  that is expected when `ezio` replaces it.)
+- **EXACTLY TWO agents bound** — among the supported agent types (`codex`, `claude`,
+  `ezio`, `agy`), exactly two must have `bindingState === "bound"` (the implementer +
+  reviewer pair). **`ezio` and `agy` are replacement roles**: either stands in for
+  `codex` or `claude`, so do NOT require `codex` and `claude` specifically — any
+  pair of two distinct supported agents passes. (The displaced slots read `null`/`unbound`
+  and that is expected when a replacement agent takes a seat.)
 - `evaluator.status` is NOT `"missing_anthropic_key"` or `"invalid_config"` (i.e., `ready`, `disabled`, and `unknown` all pass this gate; only the two true-misconfiguration statuses block)
 
 If the JSON has `{ "error": "no_collab_for_cwd", ... }`:
 
-> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` — in another), then re-run this skill.
+> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` / `agy` — in another), then re-run this skill.
 
 If `recovery.state === "recovery_required"`:
 
@@ -75,14 +76,14 @@ If `recovery.state === "recovery_required"`:
 
 If `recovery.state === "recovered"`:
 
-> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect codex` and `whisper collab reconnect claude`, then re-run this skill.
+> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect <agent>` for each bound agent, then re-run this skill.
 
 If FEWER than two agents are bound (count `bindingState === "bound"` across
-`codex`/`claude`/`ezio`):
+the supported agent types):
 
 > Only <N> agent(s) bound (<list bound agentTypes>). A workflow needs two — an
-> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio>`)
-> in a separate terminal, then re-run this skill. `ezio` may replace `codex` or `claude`.
+> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio|agy>`)
+> in a separate terminal, then re-run this skill. `ezio` or `agy` may replace `codex` or `claude`.
 
 (Do NOT append permission flags — mount already spawns the agent in full-permission mode; passing `--dangerously-skip-permissions` / `--dangerously-bypass-approvals-and-sandbox` again can crash the agent on a duplicate-argument error.)
 

package/skills/ai-whisper-sdd/SKILL.md

@@ -43,7 +43,8 @@ Parse the JSON. The expected shape is:
   "agents": [
     { "agentType": "codex",  "bindingState": "bound" | "pending_attach" | "unbound" | null },
     { "agentType": "claude", "bindingState": "bound" | "pending_attach" | "unbound" | null },
-    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null }
+    { "agentType": "ezio",   "bindingState": "bound" | "pending_attach" | "unbound" | null },
+    { "agentType": "agy",    "bindingState": "bound" | "pending_attach" | "unbound" | null }
   ],
   "recovery": { "state": "normal" | "recovery_required" | "recovered" },
   "evaluator": { "ready": true | false, "status": "ready" | "missing_anthropic_key" | "invalid_config" | "disabled" | "unknown" }
@@ -54,18 +55,18 @@ Required for readiness:
 - `daemon !== null`
 - `status === "active"`
 - `recovery.state === "normal"`
-- **EXACTLY TWO agents bound** — among `codex`, `claude`, and `ezio`, exactly two
-  must have `bindingState === "bound"` (the implementer + reviewer pair). **`ezio`
-  is a replacement role**: it stands in for `codex` or `claude`, so do NOT require
-  `codex` and `claude` specifically. A pair of `ezio` + `claude`, `ezio` + `codex`,
-  or `codex` + `claude` all pass. (The `agents` array may list all three types;
-  the displaced slot reads `null`/`unbound` and that is expected when `ezio`
-  replaces it.)
+- **EXACTLY TWO agents bound** — among the supported agent types (`codex`, `claude`,
+  `ezio`, `agy`), exactly two must have `bindingState === "bound"` (the implementer +
+  reviewer pair). **`ezio` and `agy` are replacement roles**: either stands in for
+  `codex` or `claude`, so do NOT require `codex` and `claude` specifically — any
+  pair of two distinct supported agents passes. (The `agents` array may list all
+  supported types; the displaced slots read `null`/`unbound` and that is expected
+  when a replacement agent takes a seat.)
 - `evaluator.status` is NOT `"missing_anthropic_key"` or `"invalid_config"` (i.e., `ready`, `disabled`, and `unknown` all pass this gate; only the two true-misconfiguration statuses block)
 
 If the JSON has `{ "error": "no_collab_for_cwd", ... }`:
 
-> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` — in another), then re-run this skill.
+> No collab found in this workspace. Mount any **two** agents (e.g. `whisper collab mount ezio` in one terminal and `whisper collab mount codex` — or `claude` / `agy` — in another), then re-run this skill.
 
 If `recovery.state === "recovery_required"`:
 
@@ -73,14 +74,14 @@ If `recovery.state === "recovery_required"`:
 
 If `recovery.state === "recovered"`:
 
-> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect codex` and `whisper collab reconnect claude`, then re-run this skill.
+> The collab has been recovered and still needs reconnect. Run `whisper collab reconnect <agent>` for each bound agent, then re-run this skill.
 
 If FEWER than two agents are bound (count `bindingState === "bound"` across
-`codex`/`claude`/`ezio`):
+the supported agent types):
 
 > Only <N> agent(s) bound (<list bound agentTypes>). A workflow needs two — an
-> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio>`)
-> in a separate terminal, then re-run this skill. `ezio` may replace `codex` or `claude`.
+> implementer and a reviewer. Mount another agent (`whisper collab mount <codex|claude|ezio|agy>`)
+> in a separate terminal, then re-run this skill. `ezio` or `agy` may replace `codex` or `claude`.
 
 (Do NOT append permission flags — mount already spawns the agent in full-permission mode; passing `--dangerously-skip-permissions` / `--dangerously-bypass-approvals-and-sandbox` again can crash the agent on a duplicate-argument error.)