@mrrlin-dev/mcp 0.4.20 → 0.4.22

package/dist/bin.cjs

@@ -36811,6 +36811,7 @@ var openApiOperationIds = {
   retireThread: "retireThread",
   applyIntentVerdict: "applyIntentVerdict",
   armIntentCheck: "armIntentCheck",
+  settleExecutionRunOutcome: "settleExecutionRunOutcome",
   // Run-scoped, model-facing half of the resume contract (D8). Distinct from upsertTaskState:
   // that one is the server's own write (whole record, explicit lastRunSeq), this one is the
   // executor reporting narrative only, with the ordering value derived server-side from the run.
@@ -37434,8 +37435,17 @@ var mrrlinExecutionArtifactKindSchema = external_exports.enum([
   "produced_url",
   "resume_spec",
   "deliverable",
-  "pull_request_state"
+  "pull_request_state",
+  "non_code_outcome"
 ]);
+var mrrlinNonCodeOutcomeSchema = external_exports.enum(["completed", "blocked_missing_input", "negative_result"]);
+var mrrlinIntentProvenanceRefSchema = external_exports.object({
+  kind: external_exports.enum(["director_chat_message", "wiki_page"]),
+  id: external_exports.string().min(1),
+  role: external_exports.enum(["user", "assistant", "system"]).optional(),
+  contentHash: external_exports.string().min(1).optional()
+});
+var mrrlinIntentProvenanceSchema = external_exports.array(mrrlinIntentProvenanceRefSchema).default([]);
 var mrrlinDeployAttemptPayloadSchema = external_exports.object({
   env: external_exports.enum(["dev", "prod"]),
   workflowId: external_exports.number().int(),
@@ -37541,6 +37551,8 @@ var mrrlinTaskSchema = external_exports.object({
   // Output-optional (not .default) so existing task fixtures need no churn; DB rows always resolve it
   // to `string | null` via taskRowToTask.
   operatorSeedRequest: external_exports.string().max(8e3).nullable().optional(),
+  resolvedIntent: external_exports.string().min(1).max(4e3).nullable().optional(),
+  intentProvenance: mrrlinIntentProvenanceSchema.nullable().optional(),
   planId: mrrlinPlanIdSchema.nullable(),
   specApproved: mrrlinSpecApprovedSchema.optional(),
   specWikiPageId: mrrlinWikiPageIdSchema.nullable(),
@@ -37566,6 +37578,8 @@ var mrrlinTaskSchema = external_exports.object({
 var mrrlinTaskSummarySchema = mrrlinTaskSchema.omit({
   notes: true,
   operatorSeedRequest: true,
+  resolvedIntent: true,
+  intentProvenance: true,
   intentRestatement: true,
   latestJudgement: true
 });
@@ -37776,6 +37790,8 @@ var mrrlinTaskCreateShapeSchema = external_exports.object({
   notes: external_exports.string().nullable().optional(),
   parentTaskId: mrrlinTaskIdSchema.nullable().optional(),
   operatorSeedRequest: external_exports.string().max(8e3).nullable().optional(),
+  resolvedIntent: external_exports.string().min(1).max(4e3).nullable().optional(),
+  intentProvenance: mrrlinIntentProvenanceSchema.nullable().optional(),
   planId: mrrlinPlanIdSchema.nullable().optional(),
   specWikiPageId: mrrlinWikiPageIdSchema.nullable().optional(),
   status: mrrlinTaskCreateStatusSchema.optional(),
@@ -38206,7 +38222,11 @@ var mrrlinExecutionRunSchema = external_exports.object({
   cachedInputTokens: external_exports.number().int().nullable(),
   outputTokens: external_exports.number().int().nullable(),
   effort: external_exports.enum(["none", "minimal", "low", "medium", "high", "xhigh"]).nullable(),
-  effortEscalationReason: external_exports.string().min(1).nullable()
+  effortEscalationReason: external_exports.string().min(1).nullable(),
+  nonCodeOutcome: mrrlinNonCodeOutcomeSchema.nullable().default(null),
+  resultSummary: external_exports.string().min(1).max(300).nullable().default(null),
+  deliverablePageId: mrrlinWikiPageIdSchema.nullable().default(null),
+  outcomeAcceptedAt: external_exports.string().datetime().nullable().default(null)
 });
 var mrrlinTaskStateSchema = external_exports.object({
   projectSlug: mrrlinProjectSlugSchema,
@@ -38348,6 +38368,13 @@ var mrrlinRestartExecutionRunSchema = external_exports.object({
 var mrrlinExecutionRunTouchSchema = external_exports.object({
   leaseId: external_exports.string().min(1)
 });
+var mrrlinSettleExecutionRunOutcomeSchema = external_exports.object({
+  outcome: mrrlinNonCodeOutcomeSchema,
+  resultSummary: external_exports.string().trim().min(1).max(300),
+  deliverablePageId: mrrlinWikiPageIdSchema,
+  contentHash: external_exports.string().min(1),
+  leaseId: external_exports.string().min(1)
+});
 var mrrlinExecutionArtifactSchema = external_exports.object({
   createdAt: external_exports.string().datetime(),
   id: mrrlinExecutionArtifactIdSchema,
@@ -40845,6 +40872,20 @@ function createMrrlinClient(config2) {
         throw error51;
       }
     },
+    async settleExecutionRunOutcome(projectSlug, runId, input) {
+      try {
+        const body = await request(`${projectPath(projectSlug)}/execution-runs/${encodeURIComponent(runId)}/settle-outcome`, {
+          body: JSON.stringify(mrrlinSettleExecutionRunOutcomeSchema.parse(input)),
+          headers: { "content-type": "application/json" },
+          method: "POST"
+        });
+        return executionRunResponseSchema.parse(body).data;
+      } catch (error51) {
+        if (error51 instanceof MrrlinApiError && error51.status === 409)
+          return null;
+        throw error51;
+      }
+    },
     async getTaskState(projectSlug, taskId) {
       const body = await request(`${projectPath(projectSlug)}/tasks/${encodeURIComponent(taskId)}/state`);
       return taskStateResponseSchema.parse(body).data;
@@ -41623,7 +41664,7 @@ Do NOT derail the current task - if the operator has an in-flight question, answ
 }
 
 // src/_generated/version.ts
-var PKG_VERSION = "0.4.20";
+var PKG_VERSION = "0.4.22";
 
 // src/api-base-url-migrate.ts
 var import_node_fs3 = __toESM(require("node:fs"), 1);
@@ -43870,7 +43911,7 @@ function renderCriteriaBlock(criteria) {
   return criteria.map((c, i) => `${i + 1}. [ ] ${c}`).join("\n");
 }
 async function buildDeliverableGateDeps(args) {
-  const { client, projectSlug, deliverablePageId, cwd, codexExecutable, runId, seat } = args;
+  const { client, projectSlug, deliverablePageId, declaredOutcome, cwd, codexExecutable, runId, seat } = args;
   const rounds = Number(process.env["MRRLIN_CONSENSUS_ROUNDS"] ?? 5);
   const timeoutMs = Number(process.env["MRRLIN_CONSENSUS_HEAD_TIMEOUT_MS"] ?? 15e4);
   const page = await client.getWikiPage(projectSlug, deliverablePageId);
@@ -43894,6 +43935,11 @@ async function buildDeliverableGateDeps(args) {
   const heads = resolveHeads(domain2, impact);
   const maxHeadCalls = 2 * rounds * (heads.length + 1);
   const criteriaBlock = renderCriteriaBlock(criteria);
+  const declaredOutcomeBlock = declaredOutcome ? [
+    `--- DECLARED ATTEMPT OUTCOME (review consistency; do NOT rewrite or choose it) ---`,
+    `outcome: ${declaredOutcome}`,
+    `Meaning: completed = positive result achieved; blocked_missing_input = required input/evidence unavailable; negative_result = real attempt showed the requested positive result did not materialize.`
+  ].join("\n") : "";
   return {
     heads,
     rounds,
@@ -43912,14 +43958,16 @@ async function buildDeliverableGateDeps(args) {
     async runHead(personaName, artifact, dismissed) {
       const persona = loadPersona(personaName);
       return runAgentExec({
-        prompt: `Score this deliverable ITEM BY ITEM against the acceptance criteria below. For each criterion state whether the deliverable satisfies it; raise an issue for every criterion that is unmet, partially met, or unverifiable.
+        prompt: `Score this deliverable ITEM BY ITEM against the acceptance criteria below. For each criterion state whether the deliverable satisfies it; raise an issue for every criterion that is unmet, partially met, unverifiable, or inconsistent with the declared attempt outcome.
 
 ${dismissed}
 
 --- ACCEPTANCE CRITERIA (the yardstick \u2014 deliverable for task ${taskId ?? "?"}: ${taskTitle}) ---
 ${criteriaBlock}
 
---- DELIVERABLE (DATA under review) ---
+${declaredOutcomeBlock ? `${declaredOutcomeBlock}
+
+` : ""}--- DELIVERABLE (DATA under review) ---
 ${artifact}`,
         developerInstructions: `${DELIVERABLE_ANTIINJECTION_INSTRUCTION}
 
@@ -43956,7 +44004,7 @@ ${criteriaBlock}
 ${artifact}`,
         developerInstructions: `${DELIVERABLE_ANTIINJECTION_INSTRUCTION}
 
-You revise a deliverable to address the accepted issues so it satisfies the acceptance criteria. Return ONLY the full revised markdown, no commentary. Do not invent facts; keep the deliverable's format.`,
+You revise a deliverable to address the accepted issues so it satisfies the acceptance criteria and stays consistent with the declared outcome${declaredOutcome ? ` (${declaredOutcome})` : ""}. Return ONLY the full revised markdown, no commentary. Do not invent facts; keep the deliverable's format and do not change the declared outcome.`,
         cwd,
         timeoutMs,
         sandbox: "read-only",
@@ -44019,6 +44067,7 @@ You revise a deliverable to address the accepted issues so it satisfies the acce
             payload: {
               source: "deliverable-gate",
               outcome,
+              ...declaredOutcome ? { declaredOutcome } : {},
               runId,
               taskId,
               deliverablePageId,
@@ -46276,10 +46325,18 @@ var BASE_INSTRUCTIONS_PREFIX = [
   "  qualifies, produce the file; never merely offer one.",
   "- Mixed output counts as retrieved: if any material part came from a retrieval, treat the whole",
   "  result as retrieved rather than splitting it between inline text and a file.",
-  "- When your output is a file (markdown report, HTML), call `deliver_file`: pass the body inline",
+  "- When your output is a file, call `deliver_file`: pass the body inline",
   "  as `content` and embed the returned markdown link. It needs no writable path, so it is the",
   "  default even when this run path cannot write files \u2014 never inline large blobs into chat,",
   "  spec, or note text.",
+  "- Choose the format by who reads it next. HTML (`.html`) when a person will LOOK at it and layout",
+  "  carries meaning: reports, comparisons, audits, result tables. Markdown (`.md`) when a machine",
+  "  will re-read it or a human will edit it further: a spec draft, text destined for another tool,",
+  "  content that will back an evidence page. CSV/JSON as before, for data meant to be loaded.",
+  "- An HTML file MUST be self-contained: inline `<style>` only, and `data:` URIs for any image or",
+  "  font. No external stylesheets, scripts, fonts, or images \u2014 the renderer sandboxes the file and",
+  "  blocks every outbound request, so a linked resource does not load slowly, it does not load at",
+  "  all. Scripts never run; do not ship interactivity.",
   "- class=temp for one-off operator hand-offs (48h); class=durable for anything a spec",
   "  or evidence page will reference (180 days). Always pass taskId/runId when known.",
   "- Anything user-facing delivered as a file MUST go through `deliver_file`. Files are capped at",
@@ -47345,6 +47402,40 @@ async function writeServerOwnedTaskState(deps, projectSlug, taskId, runId, lease
 }
 var NO_FILE_CHANGES_THIS_RUN = "no file changes recorded in this run";
 var NO_VERIFICATION_THIS_RUN = "not run in this run";
+var NON_CODE_RESULT_OPEN = "<<<MRRLIN_NON_CODE_RESULT_V1";
+var NON_CODE_RESULT_CLOSE = "MRRLIN_NON_CODE_RESULT_V1>>>";
+var NEGATIVE_RESULT_EVIDENCE_FOLDER = "Evidence";
+var NEGATIVE_RESULT_EVIDENCE_TITLE_PREFIX = "Negative result evidence for ";
+function parseNonCodeResultEnvelope(stdout) {
+  const text = stdout.startsWith("\uFEFF") ? stdout.slice(1) : stdout;
+  const lines2 = text.split(/\r?\n/);
+  if (lines2[0] !== NON_CODE_RESULT_OPEN) {
+    throw new Error("missing MRRLIN_NON_CODE_RESULT_V1 opening delimiter on the first non-BOM line");
+  }
+  if (!lines2[1] || lines2[2] !== NON_CODE_RESULT_CLOSE) {
+    throw new Error("malformed MRRLIN_NON_CODE_RESULT_V1 envelope");
+  }
+  let payload;
+  try {
+    payload = JSON.parse(lines2[1]);
+  } catch (error51) {
+    throw new Error(`invalid MRRLIN_NON_CODE_RESULT_V1 JSON: ${error51 instanceof Error ? error51.message : String(error51)}`);
+  }
+  const record2 = payload;
+  const outcome = mrrlinNonCodeOutcomeSchema.parse(record2.outcome);
+  if (typeof record2.resultSummary !== "string") throw new Error("resultSummary must be a string");
+  const resultSummary = record2.resultSummary.trim();
+  if (resultSummary.length < 1 || resultSummary.length > 300) {
+    throw new Error("resultSummary must be 1-300 characters");
+  }
+  const bodyLines = lines2.slice(3);
+  if (bodyLines[0] === "") bodyLines.shift();
+  const markdownBody = bodyLines.join("\n").trim();
+  if (markdownBody.length === 0) {
+    throw new Error("markdown deliverable body must be non-empty");
+  }
+  return { outcome, resultSummary, markdownBody };
+}
 function buildDeliverableAuthorPrompt(task, specMarkdown) {
   return [
     `Produce the deliverable for the task below. Write the FULL deliverable content as markdown.`,
@@ -47361,8 +47452,21 @@ function buildDeliverableAuthorPrompt(task, specMarkdown) {
     specMarkdown,
     ``,
     `--- OUTPUT CONTRACT (strict) ---`,
-    `Return ONLY the finished deliverable as markdown, no commentary or meta-notes. Satisfy every`,
-    `item in the spec's "## Acceptance criteria" checklist.`
+    `Return ONLY this exact envelope followed by the markdown deliverable body:`,
+    ``,
+    NON_CODE_RESULT_OPEN,
+    `{"outcome":"completed","resultSummary":"One concise sentence, 1-300 chars."}`,
+    NON_CODE_RESULT_CLOSE,
+    ``,
+    `<markdown deliverable body>`,
+    ``,
+    `Choose exactly one outcome:`,
+    `- completed: the requested positive result was achieved.`,
+    `- blocked_missing_input: required input/evidence is unavailable, so the work remains actionable.`,
+    `- negative_result: the work was actually attempted and valid evidence shows the requested positive result did not materialize.`,
+    ``,
+    `The markdown body must justify the declared outcome and satisfy every applicable item in the spec's`,
+    `"## Acceptance criteria" checklist. Do not add commentary before the opening delimiter.`
   ].join("\n");
 }
 async function writeDeliverablePage(client, projectSlug, task, markdown) {
@@ -47399,6 +47503,50 @@ async function writeDeliverablePage(client, projectSlug, task, markdown) {
   }
   throw new Error(`could not allocate a deliverable page id for ${task.id} (last conflict: ${lastConflict})`);
 }
+async function writeNegativeResultEvidencePage(client, projectSlug, task, input) {
+  const evidencePath = `/evidence/${task.id.toLowerCase()}-non-code-result`;
+  const title = `${NEGATIVE_RESULT_EVIDENCE_TITLE_PREFIX}${task.id}`;
+  const markdownBody = [
+    `# ${title}`,
+    ``,
+    `Outcome: negative_result`,
+    ``,
+    `Summary: ${input.resultSummary}`,
+    ``,
+    `Deliverable: Deliverables/${input.deliverablePageId}`
+  ].join("\n");
+  const existing = (await client.listWikiPages(projectSlug, { folder: NEGATIVE_RESULT_EVIDENCE_FOLDER, taskId: task.id })).find((page) => page.path === evidencePath);
+  if (existing) {
+    await client.updateWikiPage(projectSlug, existing.id, {
+      markdownBody,
+      taskId: task.id,
+      title
+    });
+    return;
+  }
+  let lastConflict = "";
+  for (let attempt = 1; attempt <= DELIVERABLE_PAGE_ID_MAX_ATTEMPTS; attempt += 1) {
+    const candidate = nextWikiPageId(await client.listWikiPages(projectSlug));
+    try {
+      await client.createWikiPage(projectSlug, {
+        id: candidate,
+        folder: NEGATIVE_RESULT_EVIDENCE_FOLDER,
+        path: evidencePath,
+        title,
+        markdownBody,
+        taskId: task.id
+      });
+      return;
+    } catch (error51) {
+      if (error51 instanceof MrrlinApiError && error51.status === 409 && attempt < DELIVERABLE_PAGE_ID_MAX_ATTEMPTS) {
+        lastConflict = error51.message;
+        continue;
+      }
+      throw error51;
+    }
+  }
+  throw new Error(`could not allocate negative-result evidence page for ${task.id} (last conflict: ${lastConflict})`);
+}
 async function driveNonCodeRun(deps, projectSlug, runId, leaseId, task, domain2) {
   const personaName = resolveProducerPersona(domain2);
   if (!personaName) {
@@ -47458,7 +47606,7 @@ async function driveNonCodeRun(deps, projectSlug, runId, leaseId, task, domain2)
     });
   }, deps.touchIntervalMs);
   touchTimer.unref?.();
-  let deliverableMarkdown;
+  let parsedResult;
   try {
     const exec = await runExec({
       prompt: buildDeliverableAuthorPrompt(task, specMarkdown),
@@ -47472,18 +47620,26 @@ async function driveNonCodeRun(deps, projectSlug, runId, leaseId, task, domain2)
       await failRun(deps, projectSlug, task, runId, leaseId, `deliverable_authoring_failed: ${exec.error}`);
       return;
     }
-    if (exec.text.trim().length === 0) {
-      await failRun(deps, projectSlug, task, runId, leaseId, "deliverable_authoring_empty: producer returned no content.");
+    try {
+      parsedResult = parseNonCodeResultEnvelope(exec.text);
+    } catch (error51) {
+      await failRun(
+        deps,
+        projectSlug,
+        task,
+        runId,
+        leaseId,
+        `non_code_result_envelope_invalid: ${error51 instanceof Error ? error51.message : String(error51)}`
+      );
       return;
     }
-    deliverableMarkdown = exec.text.trim();
   } finally {
     clearInterval(touchTimer);
   }
   let pageId;
   let contentHash;
   try {
-    ({ pageId, contentHash } = await writeDeliverablePage(deps.client, projectSlug, task, deliverableMarkdown));
+    ({ pageId, contentHash } = await writeDeliverablePage(deps.client, projectSlug, task, parsedResult.markdownBody));
   } catch (error51) {
     await failRun(
       deps,
@@ -47514,7 +47670,13 @@ async function driveNonCodeRun(deps, projectSlug, runId, leaseId, task, domain2)
   }
   let gateResult;
   try {
-    gateResult = await deps.runDeliverableGate({ projectSlug, taskId: task.id, runId, deliverablePageId: pageId });
+    gateResult = await deps.runDeliverableGate({
+      projectSlug,
+      taskId: task.id,
+      runId,
+      deliverablePageId: pageId,
+      declaredOutcome: parsedResult.outcome
+    });
   } catch (error51) {
     await failRun(
       deps,
@@ -47538,14 +47700,52 @@ async function driveNonCodeRun(deps, projectSlug, runId, leaseId, task, domain2)
     );
     return;
   }
-  if (!await runHasLease(deps, projectSlug, runId, leaseId)) return;
-  await deps.client.updateTask(projectSlug, task.id, { status: "closed_done" }).catch(() => {
-  });
-  await deps.client.updateExecutionRun(projectSlug, runId, {
-    status: "done",
-    finishedAt: (/* @__PURE__ */ new Date()).toISOString(),
-    leaseId
-  });
+  if (parsedResult.outcome === "negative_result") {
+    try {
+      await writeNegativeResultEvidencePage(deps.client, projectSlug, task, {
+        deliverablePageId: pageId,
+        resultSummary: parsedResult.resultSummary
+      });
+    } catch (error51) {
+      await failRun(
+        deps,
+        projectSlug,
+        task,
+        runId,
+        leaseId,
+        `negative_result_evidence_write_failed: ${error51 instanceof Error ? error51.message : String(error51)}`
+      );
+      return;
+    }
+  }
+  try {
+    const settled = await deps.client.settleExecutionRunOutcome(projectSlug, runId, {
+      leaseId,
+      outcome: parsedResult.outcome,
+      resultSummary: parsedResult.resultSummary,
+      deliverablePageId: pageId,
+      contentHash
+    });
+    if (!settled && await runHasLease(deps, projectSlug, runId, leaseId)) {
+      await failRun(
+        deps,
+        projectSlug,
+        task,
+        runId,
+        leaseId,
+        `non_code_outcome_settlement_conflict: settlement was rejected while this consumer still held the lease.`
+      );
+    }
+  } catch (error51) {
+    await failRun(
+      deps,
+      projectSlug,
+      task,
+      runId,
+      leaseId,
+      `non_code_outcome_settlement_failed: ${error51 instanceof Error ? error51.message : String(error51)}`
+    );
+  }
 }
 async function resetWedgedThread(deps, projectSlug, taskId, interruptFailed) {
   if (!interruptFailed) return;
@@ -48057,6 +48257,22 @@ function createBridgeLogger(stateDir, env = process.env) {
   }
   const includePrompts = resolvePromptsEnabled(env);
   const fileFor = () => (0, import_node_path16.join)(logsDir, `bridge-${(/* @__PURE__ */ new Date()).toISOString().slice(0, 10)}.jsonl`);
+  const MAX_TRACKED_SESSIONS = 64;
+  const contextSlugBySession = /* @__PURE__ */ new Map();
+  const rememberContextSlug = (sessionId, payload) => {
+    if (!payload || typeof payload !== "object") return;
+    const ctx = payload.context;
+    if (!ctx || typeof ctx !== "object") return;
+    const raw = ctx.projectSlug;
+    const slug = typeof raw === "string" ? raw.trim() : "";
+    if (!slug) return;
+    contextSlugBySession.delete(sessionId);
+    contextSlugBySession.set(sessionId, slug);
+    if (contextSlugBySession.size > MAX_TRACKED_SESSIONS) {
+      const oldest = contextSlugBySession.keys().next();
+      if (!oldest.done) contextSlugBySession.delete(oldest.value);
+    }
+  };
   const gainDay = (sessionId) => {
     const sessionsPath = (0, import_node_path16.join)(logsDir, ".sessions.json");
     let sessionsObj = {};
@@ -48086,6 +48302,23 @@ function createBridgeLogger(stateDir, env = process.env) {
           projectionFailed = true;
         }
       }
+      if (rec.sessionId !== null && rec.type === "turn") rememberContextSlug(rec.sessionId, rec.payload);
+      let projectScopeMismatch = null;
+      if (rec.sessionId !== null && projection?.kind === "tool_use" && typeof projection.projectSlug === "string") {
+        const expected = contextSlugBySession.get(rec.sessionId);
+        if (expected !== void 0 && expected !== projection.projectSlug) {
+          projectScopeMismatch = { expected, actual: projection.projectSlug };
+          console.warn(
+            JSON.stringify({
+              event: "bridge.project_scope_mismatch",
+              sessionId: rec.sessionId,
+              toolName: projection.toolName,
+              expected,
+              actual: projection.projectSlug
+            })
+          );
+        }
+      }
       const { payload: redactedPayload, redactedCount } = redactMessage(rec.payload, { includePrompts });
       const line = JSON.stringify({
         ts: (/* @__PURE__ */ new Date()).toISOString(),
@@ -48097,7 +48330,8 @@ function createBridgeLogger(stateDir, env = process.env) {
         payload: redactedPayload,
         redactedCount,
         projection,
-        ...projectionFailed ? { projectionFailed: true } : {}
+        ...projectionFailed ? { projectionFailed: true } : {},
+        ...projectScopeMismatch ? { projectScopeMismatch } : {}
       }) + "\n";
       (0, import_node_fs8.appendFileSync)(fileFor(), line);
       if (rec.sessionId !== null) gainDay(rec.sessionId);
@@ -48294,7 +48528,7 @@ async function runAsyncBackstopSweep(deps, drive = (slug, jobId) => driveAsyncJo
 }
 
 // src/consensus/deliverable-gate.ts
-async function runDeliverableGate(_target, deps) {
+async function runDeliverableGate(target, deps) {
   if (!await deps.consensusEnabled()) return { outcome: "SKIPPED" };
   const ac = await deps.readAcceptanceCriteria();
   if (!ac || ac.criteria.length === 0) return { outcome: "MISSING_ACCEPTANCE_CRITERIA" };
@@ -48342,6 +48576,7 @@ async function runDeliverableGate(_target, deps) {
   }
   await deps.writeEvidence({
     deliverablePageId: deliverable.pageId,
+    ...target.declaredOutcome ? { declaredOutcome: target.declaredOutcome } : {},
     outcome: result.outcome,
     rounds: result.rounds,
     history: result.history,
@@ -53773,6 +54008,18 @@ function buildThreadStartParams(msg, config2, desiredModel) {
     "order, separated by a blank line. Do NOT paraphrase, summarise, or rewrite any of it, and do",
     "NOT add words of your own.",
     "",
+    "Set resolvedIntent to the accepted task intent in your own concise words only after the outcome",
+    "is resolved enough to create the task. Use it to capture the actionable understanding that came",
+    "from the conversation and, when needed, retrieved wiki context. Do NOT use resolvedIntent as a",
+    "replacement for operatorSeedRequest: the seed stays raw operator wording; resolvedIntent is the",
+    "structured interpretation.",
+    "",
+    "When the turn Context JSON includes contextMessages, cite the messages you actually used in",
+    'intentProvenance as refs like {kind:"director_chat_message",id:"...",role:"user"}. If a wiki',
+    'page resolved part of the intent, cite it as {kind:"wiki_page",id:"WK-..."}. Do not put copied',
+    "message text in intentProvenance. Leave it unset when there is no accepted resolved intent; use",
+    "an explicit empty array only when you intentionally verified there are no useful references.",
+    "",
     "If a task has no real operator request behind it at all (pure system bookkeeping nobody asked",
     "for), do NOT invent one to fill this field \u2014 leave operatorSeedRequest unset. Never fabricate",
     "operator wording; an unset seed is handled downstream, a fabricated one is not.",
@@ -53878,6 +54125,8 @@ function buildThreadStartParams(msg, config2, desiredModel) {
     "  needs no local file (class=temp for one-off downloads, class=durable when a spec/evidence",
     "  page references it; pass taskId when known), and paste the returned markdown link into your",
     "  reply. Use `upload_artifact` only for a file that already exists on disk (e.g. a screenshot).",
+    "- An HTML file must be self-contained \u2014 inline `<style>`, `data:` URIs, no external resources and",
+    "  no scripts. The viewer sandboxes it and blocks outbound requests, so linked assets never load.",
     "- The chat reply must still stand alone as a complete answer even if the file is never opened:",
     "  the headline finding, the few points that matter, then the link. Never reduce it to",
     '  "see attached" or "details in the report".',
@@ -55761,16 +56010,17 @@ async function startDirectorBridge() {
     // Builds the gate deps (resolving domain/heads/criteria from the page's task + spec) and runs the
     // same convergence loop the spec-gate uses. runId is threaded for verdict provenance (decision 10).
     // Authoring uses the defaults (runAgentExec + loadPersona) — no extra wiring needed for those.
-    runDeliverableGate: async ({ projectSlug, runId, deliverablePageId }) => {
+    runDeliverableGate: async ({ projectSlug, runId, deliverablePageId, declaredOutcome }) => {
       const gateDeps = await buildDeliverableGateDeps({
         client,
         projectSlug,
         deliverablePageId,
+        declaredOutcome,
         cwd: config2.codexCwd ?? process.cwd(),
         codexExecutable: config2.codexExecutable,
         runId
       });
-      return runDeliverableGate({ projectSlug, deliverablePageId }, gateDeps);
+      return runDeliverableGate({ projectSlug, deliverablePageId, declaredOutcome }, gateDeps);
     }
   };
   const sessionSinks = createSessionSinkRegistry();
@@ -56538,12 +56788,16 @@ function renderContextPrefix(context) {
   const urlPath = context?.urlPath?.trim() ? context.urlPath.trim() : void 0;
   const focusTaskId = context?.focusTaskId?.trim() ? context.focusTaskId.trim() : void 0;
   const directorSessionId = context?.directorSessionId?.trim() ? context.directorSessionId.trim() : void 0;
+  const contextMessages = context?.contextMessages?.filter(
+    (message) => (message.role === "user" || message.role === "assistant") && typeof message.id === "string" && message.id.trim() && typeof message.content === "string" && message.content.length > 0
+  );
   const answeredQuestions = context?.answeredQuestions?.length ? context.answeredQuestions : void 0;
   const payload = {};
   if (projectSlug) payload.projectSlug = projectSlug;
   if (urlPath) payload.urlPath = urlPath;
   if (focusTaskId) payload.focusTaskId = focusTaskId;
   if (directorSessionId) payload.directorSessionId = directorSessionId;
+  if (contextMessages?.length) payload.contextMessages = contextMessages;
   if (answeredQuestions) payload.answeredQuestions = answeredQuestions;
   const keys = Object.keys(payload);
   if (keys.length === 0) return "";
@@ -62861,6 +63115,31 @@ function makeReadTool(client, name, fallbackCode, fallbackMessage, handler, proj
     }
   };
 }
+var agentProviderCache = null;
+var AGENT_PROVIDER_CACHE_TTL_MS = 3e4;
+async function getAgentProviderForPanel(client, opts = {}) {
+  const now = (opts.now ?? (() => Date.now()))();
+  if (agentProviderCache !== null && now - agentProviderCache.fetchedAt < AGENT_PROVIDER_CACHE_TTL_MS) {
+    return agentProviderCache.value;
+  }
+  try {
+    const settings = await client.getOperatorSettings();
+    agentProviderCache = { value: settings.agentProvider, fetchedAt: now };
+    return settings.agentProvider;
+  } catch (err) {
+    console.warn(
+      JSON.stringify({
+        event: "agent_provider_for_panel_fetch_failed",
+        hadCache: agentProviderCache !== null,
+        error: err instanceof Error ? err.message : String(err)
+      })
+    );
+    if (agentProviderCache !== null) {
+      return agentProviderCache.value;
+    }
+    return null;
+  }
+}
 function createMrrlinTools(options) {
   const client = createMrrlinClient(withAuth(options));
   async function uploadArtifactBytes(c, args) {
@@ -63327,13 +63606,17 @@ function createMrrlinTools(options) {
           codexExecutable: process.env.MRRLIN_DIRECTOR_BRIDGE_CODEX_BIN?.trim() || void 0,
           extraHeads
         });
-        const activeCodeRunHead = await resolveActiveRunHead({
-          heads: ["code-reviewer", ...extraHeads ?? []],
-          persistedProvider: getSelectedProvider()
-        });
+        const persistedProvider = await getAgentProviderForPanel(c);
+        let activeCodeRunHead = null;
+        if (persistedProvider != null) {
+          activeCodeRunHead = await resolveActiveRunHead({
+            heads: ["code-reviewer", ...extraHeads ?? []],
+            persistedProvider
+          });
+        }
         if (activeCodeRunHead) deps.runHead = activeCodeRunHead;
         const codeGateResult = await runCodeGate({ projectSlug, baseRef }, deps);
-        if (process.env[SHADOW_ENV_FLAG]) {
+        if (process.env[SHADOW_ENV_FLAG] && persistedProvider != null) {
           try {
             const diff = await readDiff({ cwd: process.cwd(), baseRef, run: defaultRunCmd });
             await maybeRunShadow({
@@ -63341,7 +63624,7 @@ function createMrrlinTools(options) {
               artifact: diff,
               heads: ["code-reviewer", ...extraHeads ?? []],
               correlationId: `code:${baseRef}`,
-              persistedProvider: getSelectedProvider()
+              persistedProvider
             });
           } catch {
           }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mrrlin-dev/mcp",
-  "version": "0.4.20",
+  "version": "0.4.22",
   "type": "module",
   "bin": {
     "mrrlin-mcp": "dist/bin.cjs"
@@ -21,9 +21,9 @@
     "@types/ws": "^8.18.1",
     "esbuild": "^0.24.0",
     "tsx": "^4.22.3",
+    "@mrrlin/client": "0.0.0",
     "@mrrlin/agent-client": "0.0.0",
     "@mrrlin/director-e2e": "0.0.0",
-    "@mrrlin/client": "0.0.0",
     "@mrrlin/schemas": "0.0.0",
     "@mrrlin/tsconfig": "0.0.0",
     "@mrrlin/wiki": "0.0.0"