@llblab/pi-actors 0.37.1 → 0.38.0

package/dist/lib/tools-response.js

@@ -30,6 +30,23 @@ export function compactNextActions(actions) {
 function formatFailureCount(value) {
     return Array.isArray(value) ? value.length : undefined;
 }
+function compactPolicyAxis(label, value) {
+    const axis = asRecord(value);
+    const source = typeof axis.source === "string" ? axis.source : undefined;
+    if (!source || source === "unused")
+        return undefined;
+    const renderedValue = typeof axis.value === "string" && axis.value.trim()
+        ? `:${axis.value.trim()}`
+        : "";
+    return `${label}=${source}${renderedValue}`;
+}
+function compactModelPolicy(value) {
+    const policy = asRecord(value);
+    return [
+        compactPolicyAxis("model", policy.model),
+        compactPolicyAxis("thinking", policy.thinking),
+    ].filter((token) => Boolean(token));
+}
 export function actorRunNextActions(run) {
     const id = String(run ?? "").trim();
     if (!id)
@@ -46,6 +63,7 @@ export function compactAsyncRunStatus(value) {
     const result = asRecord(status.result);
     const run = String(status.run ?? "<unknown>");
     const tokens = [`run=${run}`, `status=${String(status.status ?? "unknown")}`];
+    tokens.push(...compactModelPolicy(status.model_policy ?? progress.model_policy));
     if (status.tool)
         tokens.push(`tool=${String(status.tool)}`);
     if (status.recipe)
@@ -222,9 +240,12 @@ export function compactRecipeRegistry(summary) {
     const recommendations = Array.isArray(summary.recommendations)
         ? summary.recommendations.length
         : 0;
+    const currentPolicy = Array.isArray(summary.active)
+        ? summary.active.filter((entry) => entry.current_policy).length
+        : 0;
     const nextActions = Array.isArray(summary.next_actions)
         ? summary.next_actions
         : [];
-    return `\nrecipes active=${active} drafts=${drafts} shadowed=${shadowed} invalid=${invalid} disabled=${disabled} recommendations=${recommendations} diagnostics=${diagnostics}${compactNextActions(nextActions)}`;
+    return `\nrecipes active=${active} drafts=${drafts} shadowed=${shadowed} invalid=${invalid} disabled=${disabled} current_policy=${currentPolicy} recommendations=${recommendations} diagnostics=${diagnostics}${compactNextActions(nextActions)}`;
 }
 export const DEFAULT_INSPECT_LINES = Limits.DEFAULT_INSPECT_LINES;

package/dist/lib/tools-spawn.d.ts

@@ -3,7 +3,8 @@
  * Zones: actor launch, draft recipe capture, launch diagnostics
  * Owns the public spawn execution path for run-backed actors
  */
-export interface SpawnToolContext {
+import * as ModelContext from "./model-context.ts";
+export interface SpawnToolContext extends ModelContext.CurrentModelContext {
     cwd: string;
     sessionManager?: {
         getSessionId?: () => string;

package/dist/lib/tools-spawn.js

@@ -7,6 +7,7 @@ import { mkdirSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 import * as AsyncRuns from "./async-runs.js";
 import * as Messages from "./messages.js";
+import * as ModelContext from "./model-context.js";
 import * as Paths from "./paths.js";
 import * as RecipesDiscovery from "./recipes-discovery.js";
 import * as Rooms from "./rooms.js";
@@ -123,7 +124,7 @@ export function createSpawnToolDefinition() {
                             template: input.template,
                         }
                         : {}),
-                    values: asRecord(input.values),
+                    values: ModelContext.withCurrentModelValues(asRecord(input.values), ctx),
                     ...(input.artifacts &&
                         typeof input.artifacts === "object" &&
                         !Array.isArray(input.artifacts)

package/dist/recipes/lens-swarm.json

@@ -10,6 +10,10 @@
     "model:string",
     "thinking:string",
     "tools:string",
+    "subagent_ttl_ms:int",
+    "reviewer_concurrency:string",
+    "min_successful_reviewers:int",
+    "merge_policy:string",
     "claim:string",
     "evidence_policy:string",
     "risk_policy:string",
@@ -21,12 +25,17 @@
       "architecture",
       "operator UX"
     ],
-    "thinking": "off",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
     "tools": "",
+    "subagent_ttl_ms": 600000,
+    "reviewer_concurrency": "",
+    "min_successful_reviewers": 1,
+    "merge_policy": "Preserve partial reviewer reports, branch status, and degraded confidence markers. If reviewer evidence is below threshold, do not invent consensus.",
     "claim": "The reviewed scope has clear, evidence-based findings, risks, and recommended next actions.",
     "evidence_policy": "Cite inspected files, command outputs, or explicit uncertainty for every material claim.",
     "risk_policy": "Preserve minority high-impact risks and ideas; separate confirmed issues from hypotheses.",
-    "output_format": "Markdown sections: Summary, Consensus Findings, Minority Findings, Verification, Judge Notes, Risks, Next Actions."
+    "output_format": "Markdown sections: Status (complete/degraded/insufficient_data), Summary, Consensus Findings, Minority Findings, Verification, Judge Notes, Risks, Next Actions."
   },
   "mailbox": {
     "accepts": [
@@ -55,6 +64,10 @@
         "judge_model": "{model}",
         "thinking": "{thinking}",
         "tools": "{tools}",
+        "subagent_ttl_ms": "{subagent_ttl_ms}",
+        "reviewer_concurrency": "{reviewer_concurrency}",
+        "min_successful_reviewers": "{min_successful_reviewers}",
+        "merge_policy": "{merge_policy}",
         "evidence_policy": "{evidence_policy}",
         "risk_policy": "{risk_policy}",
         "output_format": "{output_format}"

package/dist/recipes/pipeline-release-readiness.json

@@ -20,7 +20,12 @@
     "verifier_model:string",
     "merger_model:string",
     "judge_model:string",
-    "tools:string"
+    "thinking:string",
+    "tools:string",
+    "subagent_ttl_ms:int",
+    "reviewer_concurrency:string",
+    "min_successful_reviewers:int",
+    "merge_policy:string"
   ],
   "defaults": {
     "scope": ".",
@@ -36,7 +41,16 @@
       "documentation",
       "packaged skills"
     ],
-    "tools": ""
+    "reviewer_model": "{current_model}",
+    "verifier_model": "{current_model}",
+    "merger_model": "{current_model}",
+    "judge_model": "{current_model}",
+    "thinking": "{current_thinking}",
+    "tools": "",
+    "subagent_ttl_ms": 600000,
+    "reviewer_concurrency": "",
+    "min_successful_reviewers": 1,
+    "merge_policy": "Preserve partial reviewer reports, branch status, and degraded confidence markers. If reviewer evidence is below threshold, do not invent consensus."
   },
   "mailbox": {
     "accepts": [
@@ -90,8 +104,13 @@
         "verifier_model": "{verifier_model}",
         "merger_model": "{merger_model}",
         "judge_model": "{judge_model}",
+        "thinking": "{thinking}",
         "tools": "{tools}",
-        "output_format": "Markdown sections: Release Verdict, Blocking Issues, Package/Docs Risks, Validation Evidence, Publish Notes, Required Follow-up."
+        "subagent_ttl_ms": "{subagent_ttl_ms}",
+        "reviewer_concurrency": "{reviewer_concurrency}",
+        "min_successful_reviewers": "{min_successful_reviewers}",
+        "merge_policy": "{merge_policy}",
+        "output_format": "Markdown sections: Status (complete/degraded/insufficient_data), Release Verdict, Blocking Issues, Package/Docs Risks, Validation Evidence, Publish Notes, Required Follow-up."
       }
     },
     {

package/dist/recipes/pipeline-review-readiness.json

@@ -11,7 +11,12 @@
     "verifier_model:string",
     "merger_model:string",
     "judge_model:string",
-    "tools:string"
+    "thinking:string",
+    "tools:string",
+    "subagent_ttl_ms:int",
+    "reviewer_concurrency:string",
+    "min_successful_reviewers:int",
+    "merge_policy:string"
   ],
   "defaults": {
     "lenses": [
@@ -21,7 +26,16 @@
       "operator UX"
     ],
     "release_gate": "The scope is safe to ship or hand to the next release step.",
-    "tools": ""
+    "reviewer_model": "{current_model}",
+    "verifier_model": "{current_model}",
+    "merger_model": "{current_model}",
+    "judge_model": "{current_model}",
+    "thinking": "{current_thinking}",
+    "tools": "",
+    "subagent_ttl_ms": 600000,
+    "reviewer_concurrency": "",
+    "min_successful_reviewers": 1,
+    "merge_policy": "Preserve partial reviewer reports, branch status, and degraded confidence markers. If reviewer evidence is below threshold, do not invent consensus."
   },
   "mailbox": {
     "accepts": [
@@ -45,8 +59,13 @@
       "verifier_model": "{verifier_model}",
       "merger_model": "{merger_model}",
       "judge_model": "{judge_model}",
+      "thinking": "{thinking}",
       "tools": "{tools}",
-      "output_format": "Markdown sections: Ship Verdict, Blocking Findings, Nonblocking Risks, Evidence, Required Follow-up."
+      "subagent_ttl_ms": "{subagent_ttl_ms}",
+      "reviewer_concurrency": "{reviewer_concurrency}",
+      "min_successful_reviewers": "{min_successful_reviewers}",
+      "merge_policy": "{merge_policy}",
+      "output_format": "Markdown sections: Status (complete/degraded/insufficient_data), Ship Verdict, Blocking Findings, Nonblocking Risks, Evidence, Required Follow-up."
     }
   }
 }

package/dist/recipes/subagent-judge.json

@@ -13,7 +13,8 @@
     "rubric": "Judge evidence preservation, severity calibration, consensus purity, internal consistency, and merge bias.",
     "evidence": "Use the provided raw outputs or artifact paths; mark any missing evidence.",
     "output_format": "Markdown sections: Verdict, Issues, Evidence Preservation, Bias Risks, Required Fixes.",
-    "thinking": "medium",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
     "tools": ""
   },
   "mailbox": {

package/dist/recipes/subagent-merge.json

@@ -3,6 +3,7 @@
   "args": [
     "inputs:string",
     "mode:enum(consensus-first,risk-first)",
+    "merge_policy:string",
     "risk_policy:string",
     "output_format:string",
     "model:string",
@@ -11,9 +12,11 @@
   ],
   "defaults": {
     "mode": "consensus-first",
+    "merge_policy": "Preserve partial reviewer evidence, branch status, and degraded confidence markers.",
     "risk_policy": "Preserve minority severe findings and mark merger-added claims explicitly.",
     "output_format": "Markdown sections: Summary, Consensus, Minority Findings, Contradictions, Next Actions.",
-    "thinking": "medium",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
     "tools": ""
   },
   "mailbox": {
@@ -27,5 +30,5 @@
       "run.failed"
     ]
   },
-  "template": "pi -p --model {model} --thinking {thinking} {tools?--tools:--no-tools} {tools} Merge these subagent outputs using mode {mode}: {inputs}. Risk policy: {risk_policy}. Output format: {output_format}"
+  "template": "pi -p --model {model} --thinking {thinking} {tools?--tools:--no-tools} {tools} Merge these subagent outputs using mode {mode}: {inputs}. Merge policy: {merge_policy}. Risk policy: {risk_policy}. Output format: {output_format}"
 }

package/dist/recipes/subagent-normalize.json

@@ -11,7 +11,8 @@
   "defaults": {
     "format": "Markdown sections: Summary, Findings, Evidence, Risks, Next Actions.",
     "preservation_policy": "Do not change meaning, severity, evidence, or uncertainty while normalizing.",
-    "thinking": "off",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
     "tools": ""
   },
   "mailbox": {

package/dist/recipes/subagent-preflight.json

@@ -0,0 +1,29 @@
+{
+  "async": true,
+  "args": [
+    "stage:string",
+    "model:string",
+    "thinking:string",
+    "tools:string",
+    "output_format:string"
+  ],
+  "defaults": {
+    "stage": "subagent",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
+    "tools": "",
+    "output_format": "Reply exactly: ACTOR_PREFLIGHT_OK"
+  },
+  "mailbox": {
+    "accepts": [
+      "control.kill"
+    ],
+    "emits": [
+      "preflight.completed",
+      "command.done",
+      "run.done",
+      "run.failed"
+    ]
+  },
+  "template": "pi -p --model {model} --thinking {thinking} {tools?--tools:--no-tools} {tools} Preflight check for stage {stage}. Confirm this model, thinking level, and tool policy can start before expensive fanout. Do not inspect files. {output_format}"
+}

package/dist/recipes/subagent-review-coordinator.json

@@ -1,6 +1,7 @@
 {
   "async": true,
   "imports": {
+    "preflight": "subagent-preflight.json",
     "reviewer": "subagent-review.json",
     "verifier": "subagent-verify.json",
     "merger": "subagent-merge.json",
@@ -17,6 +18,10 @@
     "judge_model:string",
     "thinking:string",
     "tools:string",
+    "subagent_ttl_ms:int",
+    "reviewer_concurrency:string",
+    "min_successful_reviewers:int",
+    "merge_policy:string",
     "evidence_policy:string",
     "risk_policy:string",
     "output_format:string"
@@ -28,11 +33,19 @@
       "operator UX"
     ],
     "claim": "The reviewed scope is ready for the next implementation or release step.",
-    "thinking": "off",
+    "thinking": "{current_thinking}",
     "tools": "",
+    "subagent_ttl_ms": 600000,
+    "reviewer_concurrency": "",
+    "min_successful_reviewers": 1,
+    "merge_policy": "Preserve partial reviewer reports, branch status, and degraded confidence markers. If reviewer evidence is below threshold, do not invent consensus.",
     "evidence_policy": "Cite inspected files, command output, or explicit uncertainty for every material claim.",
     "risk_policy": "Preserve minority high-impact risks and separate confirmed issues from hypotheses.",
-    "output_format": "Markdown sections: Summary, Consensus Findings, Minority Findings, Verification, Judge Notes, Risks, Next Actions."
+    "output_format": "Markdown sections: Status (complete/degraded/insufficient_data), Summary, Consensus Findings, Minority Findings, Verification, Judge Notes, Risks, Next Actions.",
+    "reviewer_model": "{current_model}",
+    "verifier_model": "{current_model}",
+    "merger_model": "{current_model}",
+    "judge_model": "{current_model}"
   },
   "mailbox": {
     "accepts": [
@@ -49,12 +62,65 @@
     ]
   },
   "template": [
+    {
+      "parallel": true,
+      "failure": "root",
+      "template": [
+        {
+          "label": "preflight:reviewer",
+          "name": "preflight",
+          "timeout": "{subagent_ttl_ms}",
+          "values": {
+            "stage": "reviewer",
+            "model": "{reviewer_model}",
+            "thinking": "{thinking}",
+            "tools": "{tools}"
+          }
+        },
+        {
+          "label": "preflight:verifier",
+          "name": "preflight",
+          "timeout": "{subagent_ttl_ms}",
+          "values": {
+            "stage": "verifier",
+            "model": "{verifier_model}",
+            "thinking": "{thinking}",
+            "tools": "{tools}"
+          }
+        },
+        {
+          "label": "preflight:merger",
+          "name": "preflight",
+          "timeout": "{subagent_ttl_ms}",
+          "values": {
+            "stage": "merger",
+            "model": "{merger_model}",
+            "thinking": "{thinking}",
+            "tools": "{tools}"
+          }
+        },
+        {
+          "label": "preflight:judge",
+          "name": "preflight",
+          "timeout": "{subagent_ttl_ms}",
+          "values": {
+            "stage": "judge",
+            "model": "{judge_model}",
+            "thinking": "{thinking}",
+            "tools": "{tools}"
+          }
+        }
+      ]
+    },
     {
       "parallel": true,
       "repeat": "{lenses.length}",
+      "concurrency": "{reviewer_concurrency}",
+      "min_successful": "{min_successful_reviewers}",
       "failure": "branch",
       "template": {
         "name": "reviewer",
+        "timeout": "{subagent_ttl_ms}",
         "values": {
           "scope": "{scope}",
           "lens": "{lenses[index]}",
@@ -68,9 +134,10 @@
     },
     {
       "name": "verifier",
+      "timeout": "{subagent_ttl_ms}",
       "values": {
         "claim": "{claim}",
-        "evidence": "Use previous reviewer outputs from stdin and named scope: {scope}.",
+        "evidence": "Use previous reviewer outputs from stdin and named scope: {scope}. Respect the parallel_status header and verify only against usable reviewer evidence.",
         "model": "{verifier_model}",
         "thinking": "{thinking}",
         "tools": "{tools}",
@@ -79,32 +146,37 @@
     },
     {
       "name": "merger",
+      "timeout": "{subagent_ttl_ms}",
       "values": {
         "inputs": "Use previous reviewer and verifier outputs from stdin.",
         "mode": "consensus-first",
+        "merge_policy": "{merge_policy}",
         "model": "{merger_model}",
-        "thinking": "medium",
+        "thinking": "{thinking}",
         "tools": "{tools}",
         "risk_policy": "{risk_policy}"
       }
     },
     {
       "name": "judge",
+      "timeout": "{subagent_ttl_ms}",
       "values": {
         "report": "Use merged output from stdin.",
-        "evidence": "Use previous reviewer, verifier, and merger outputs from stdin.",
+        "evidence": "Use previous reviewer, verifier, and merger outputs from stdin. Preserve complete/degraded/insufficient_data status.",
         "model": "{judge_model}",
-        "thinking": "medium",
+        "thinking": "{thinking}",
         "tools": "{tools}"
       }
     },
     {
       "name": "normalizer",
+      "timeout": "{subagent_ttl_ms}",
       "values": {
         "input": "Use merged and judged output from stdin.",
         "format": "{output_format}",
+        "preservation_policy": "Preserve branch status and explicitly mark Status as complete, degraded, or insufficient_data.",
         "model": "{merger_model}",
-        "thinking": "off",
+        "thinking": "{thinking}",
         "tools": "{tools}"
       }
     }

package/dist/recipes/subagent-review.json

@@ -17,7 +17,8 @@
     "evidence_policy": "Cite inspected files, command output, or explicit uncertainty for every material claim.",
     "risk_policy": "Preserve minority high-impact risks and separate confirmed issues from hypotheses.",
     "output_format": "Markdown sections: Findings, Evidence, Risks, Next Actions.",
-    "thinking": "off",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
     "tools": ""
   },
   "mailbox": {

package/dist/recipes/subagent-verify.json

@@ -14,7 +14,8 @@
     "acceptance": "Separate proven, disproven, unknown, and missing evidence.",
     "evidence_policy": "Do not infer beyond provided evidence or inspected artifacts.",
     "output_format": "Markdown sections: Verdict, Evidence, Gaps, Confidence.",
-    "thinking": "off",
+    "model": "{current_model}",
+    "thinking": "{current_thinking}",
     "tools": ""
   },
   "mailbox": {

package/dist/scripts/async-runner.mjs

@@ -8,7 +8,7 @@
  * chasing a one-off lib entrypoint domain.
  */
 
-import { appendFileSync, existsSync, readFileSync } from "node:fs";
+import { appendFileSync, existsSync, mkdirSync, readFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath, pathToFileURL } from "node:url";
 
@@ -25,8 +25,10 @@ async function importRuntimeModule(name) {
   );
 }
 
-const { appendRecipeContextToPiArgs } =
+const { appendRecipeContextToPiArgs, materializePiPrintPromptArg } =
   await importRuntimeModule("recipes-context");
+const { buildReviewPreflightDiagnostic, formatReviewPreflightDiagnostic } =
+  await importRuntimeModule("preflight-diagnostics");
 const { execCommandTemplate } = await importRuntimeModule("command-templates");
 const { executeRegisteredTool } = await importRuntimeModule("execution");
 const { writeJsonAtomic } = await importRuntimeModule("file-state");
@@ -75,6 +77,7 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
 
   function progress(phase, extra = {}) {
     writeJsonAtomic(progressPath, {
+      ...(meta.model_policy ? { model_policy: meta.model_policy } : {}),
       phase,
       updatedAt: new Date().toISOString(),
       ...extra,
@@ -83,6 +86,7 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
 
   let activeSubagents = 0;
   let completedSubagents = 0;
+  let promptCounter = 0;
   const subagentFailures = [];
 
   function getCommandDoneDelivery(result) {
@@ -97,18 +101,65 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
     });
   }
 
+  function promptFilePath() {
+    promptCounter += 1;
+    const dir = join(stateDir, "prompts");
+    mkdirSync(dir, { recursive: true });
+    return join(dir, `command-${String(promptCounter).padStart(3, "0")}.md`);
+  }
+
+  function readPromptText(promptFile) {
+    if (!promptFile) return undefined;
+    try {
+      return readFileSync(promptFile, "utf8");
+    } catch {
+      return undefined;
+    }
+  }
+
   async function observedExec(command, args, options) {
-    const commandDetail = formatCommandDetail(command, args);
-    const execArgs = appendRecipeContextToPiArgs(
+    const contextArgs = appendRecipeContextToPiArgs(
       command,
       args,
       meta.recipe_context_records,
       options?.actorRecipeContext,
     );
+    const materialized = materializePiPrintPromptArg(
+      command,
+      contextArgs,
+      promptFilePath,
+    );
+    const execArgs = materialized.args;
+    const commandDetail = formatCommandDetail(command, execArgs);
     activeSubagents += 1;
-    event("command.start", { activeSubagents, command: commandDetail });
+    event("command.start", {
+      activeSubagents,
+      command: commandDetail,
+      ...(materialized.promptFile ? { prompt_file: materialized.promptFile } : {}),
+      ...(materialized.promptBytes ? { prompt_bytes: materialized.promptBytes } : {}),
+    });
     progressRunning();
-    const result = await execCommandTemplate(command, execArgs, options);
+    let result = await execCommandTemplate(command, execArgs, options);
+    const preflightDiagnostic = result.code !== 0
+      ? buildReviewPreflightDiagnostic({
+          args: execArgs,
+          code: result.code,
+          killed: result.killed,
+          ...(materialized.promptFile ? { promptFile: materialized.promptFile } : {}),
+          promptText: readPromptText(materialized.promptFile),
+          stderr: result.stderr,
+          stdout: result.stdout,
+        })
+      : undefined;
+    if (preflightDiagnostic) {
+      result = {
+        ...result,
+        stderr: [
+          result.stderr,
+          formatReviewPreflightDiagnostic(preflightDiagnostic),
+        ].filter(Boolean).join("\n"),
+      };
+    }
     activeSubagents = Math.max(0, activeSubagents - 1);
     completedSubagents += 1;
     if (result.code !== 0) {
@@ -116,6 +167,8 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
         code: result.code,
         command: commandDetail,
         killed: result.killed,
+        ...(materialized.promptFile ? { prompt_file: materialized.promptFile } : {}),
+        ...(preflightDiagnostic ? { preflight: preflightDiagnostic } : {}),
       });
     }
     event("command.done", {
@@ -123,6 +176,9 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
       code: result.code,
       command: commandDetail,
       killed: result.killed,
+      ...(materialized.promptFile ? { prompt_file: materialized.promptFile } : {}),
+      ...(materialized.promptBytes ? { prompt_bytes: materialized.promptBytes } : {}),
+      ...(preflightDiagnostic ? { preflight: preflightDiagnostic } : {}),
     });
     outbox(
       "command.done",
@@ -134,6 +190,9 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
         code: result.code,
         command: commandDetail,
         killed: result.killed,
+        ...(materialized.promptFile ? { prompt_file: materialized.promptFile } : {}),
+        ...(materialized.promptBytes ? { prompt_bytes: materialized.promptBytes } : {}),
+        ...(preflightDiagnostic ? { preflight: preflightDiagnostic } : {}),
       },
       getCommandDoneDelivery(result),
       result.code === 0 ? "info" : "error",
@@ -163,6 +222,7 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
       code: result.details.code,
       command: result.details.command,
       killed: result.details.killed,
+      ...(meta.model_policy ? { model_policy: meta.model_policy } : {}),
       truncated: result.details.truncated,
       completedAt: new Date().toISOString(),
     });
@@ -173,15 +233,29 @@ export async function runAsyncRunner(stateDir = process.argv[2]) {
     event("run.done", { code: result.details.code });
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error);
+    const details = error && typeof error === "object" ? error.details : undefined;
     appendFileSync(stderrPath, `${message}\n`);
     writeJsonAtomic(resultPath, {
-      code: 1,
+      code: typeof details?.code === "number" ? details.code : 1,
       error: message,
-      killed: false,
+      killed: Boolean(details?.killed),
+      ...(Array.isArray(details?.branches) ? { branches: details.branches } : {}),
+      ...(details?.failureReason ? { failure_reason: details.failureReason } : {}),
+      ...(meta.model_policy ? { model_policy: meta.model_policy } : {}),
+      ...(details?.softQuorum ? { soft_quorum: details.softQuorum } : {}),
       completedAt: new Date().toISOString(),
     });
-    progress("failed", { completed: 0, failures: [{ message }] });
-    event("run.failed", { error: message });
+    progress("failed", {
+      completed: 0,
+      failures: Array.isArray(details?.branches) && details.branches.length > 0
+        ? details.branches
+        : [{ message }],
+      ...(details?.failureReason ? { failureReason: details.failureReason } : {}),
+    });
+    event("run.failed", {
+      error: message,
+      ...(details?.failureReason ? { failure_reason: details.failureReason } : {}),
+    });
     throw error;
   }
 }

package/dist/scripts/conformance.mjs

@@ -14,6 +14,7 @@ import { fileURLToPath } from "node:url";
 const conformanceSuites = [
   "tests/protocol-examples.test.ts",
   "tests/recipes-discovery.test.ts",
+  "tests/review-swarm-dogfood.test.ts",
   "tests/registry.test.ts",
   "tests/runtime.test.ts",
   "tests/async-runs.test.ts",