@awak-app/simy-cli 0.5.69 → 0.5.71

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@awak-app/simy-cli",
-  "version": "0.5.69",
+  "version": "0.5.71",
   "description": "Local SIMY executor for non-coding Local Tasks and Agentic Loop runs.",
   "type": "module",
   "bin": {

package/src/agent.js

@@ -875,7 +875,7 @@ export async function startAgent({
       runReview: async (request, provider) => {
         const scope = await authorizeDesktopWork();
         const beforeDispatch = () => authorizeDesktopWork(scope);
-        return (dependencies.runSqmLocalReview || ((value, selected, options) => runStructuredProviderReview(selected, value, options)))(request, provider, { beforeDispatch });
+        return (dependencies.runSqmLocalReview || ((value, selected, options) => runStructuredProviderReview(selected, value, options)))(request, provider, { beforeDispatch, readOnlyAnalysis: true });
       },
       onChange: notifyDesktopStatusChange,
       getWorkspaceUrl: () => {

package/src/cli-contract.js

@@ -13,7 +13,7 @@ import {
 } from "./durable-local-task-steps-contract.js";
 import { buildExecutionCapabilityContract } from "./execution-capability-contract.js";
 
-export const CLI_VERSION = "0.5.69";
+export const CLI_VERSION = "0.5.71";
 export const CLI_API_CONTRACT_VERSION = 10;
 
 export function withCliContract(

package/src/sqm/review-worker.js

@@ -14,16 +14,19 @@ const DEFAULT_LEASE_SECONDS = 300;
 // Claim can retrieve and freeze organization-scoped Slack evidence (45s budget).
 const DEFAULT_REQUEST_TIMEOUT_MS = 75_000;
 const DEFAULT_EXECUTION_TIMEOUT_MS = 240_000;
-// Pin SQM independently of the user's interactive model and CLI defaults.
-export const SQM_CODEX_MODEL = "gpt-6.1-sol";
+// Keep analysis independent of interactive config. Availability varies by
+// account and CLI; only an explicit model rejection may try the next model.
+export const SQM_CODEX_MODELS = Object.freeze(["gpt-6-astra", "gpt-6.1-sol"]);
+export const SQM_CODEX_MODEL = SQM_CODEX_MODELS[0];
 
 export class SqmLocalReviewError extends Error {
-  constructor(message, { status = null, code = "sqm_local_review_error", stage = null, jobId = null, failureRecorded = false } = {}) {
+  constructor(message, { status = null, code = "sqm_local_review_error", stage = null, jobId = null, failureRecorded = false, validationFeedback = "" } = {}) {
     super(message);
     this.name = "SqmLocalReviewError";
     this.status = status;
     this.code = code;
     this.stage = stage; this.jobId = jobId; this.failureRecorded = failureRecorded;
+    this.validationFeedback = typeof validationFeedback === "string" ? validationFeedback.slice(0, 2000) : "";
   }
 }
 
@@ -71,6 +74,7 @@ export function createSqmLocalReviewApi({
           status: response.status,
           code: machineCode(payload?.code || payload?.detail?.code),
           stage: payload?.detail?.stage, jobId: payload?.detail?.job_id, failureRecorded: payload?.detail?.failure_recorded === true,
+          validationFeedback: payload?.detail?.validation_feedback,
         });
       }
       return payload;
@@ -124,6 +128,23 @@ export function createSqmLocalReviewWorker({
     lease_token: claim.lease_token,
   });
 
+  // Repair only a rejected model result, under the same live lease. Network,
+  // authorization, and persistence failures must not repeat provider execution.
+  const validatedReview = async (request, provider, code, submit, setStage) => {
+    let nextRequest = request;
+    for (let attempt = 0; ; attempt += 1) {
+      setStage(false);
+      const review = await runReview(nextRequest, provider);
+      setStage(true);
+      try {
+        return { review, response: await submit(review) };
+      } catch (error) {
+        if (attempt >= 1 || error.status !== 422 || error.code !== code || !error.validationFeedback) throw error;
+        nextRequest = { ...request, prompt: `${request.prompt}\n\nThe previous result failed validation. Return a corrected complete JSON object using only the supplied evidence. Fix these constraints:\n${error.validationFeedback}` };
+      }
+    }
+  };
+
   const processClaim = async (claim, provider, historyScope, workspaceUrl) => {
     activeJobId = claim.job_id;
     const priorIndex = history.findIndex(item => item.id === claim.job_id);
@@ -142,25 +163,21 @@ export function createSqmLocalReviewWorker({
     heartbeatTimer.unref?.();
     let stage = "analysis";
     try {
-      const analysis = await runReview(claim.request, provider);
-      stage = "advance";
-      const advanced = await api.request({
+      const { review: analysis, response: advanced } = await validatedReview(claim.request, provider, "invalid_sqm_analysis_output", analysis => api.request({
         action: "advance",
         ...leaseBody(claim, provider),
         analysis_output: analysis.output,
         executor_model: analysis.model,
-      });
+      }), submitting => { stage = submitting ? "advance" : "analysis"; });
       if (advanced.completed) { await updateTask(claim.job_id, "completed", historyScope); return advanced; }
       stage = "adjudication";
-      const adjudication = await runReview(advanced.request, provider);
-      stage = "complete";
-      const completed = await api.request({
+      const { response: completed } = await validatedReview(advanced.request, provider, "invalid_sqm_adjudication_output", adjudication => api.request({
         action: "complete",
         ...leaseBody(claim, provider),
         analysis_output: analysis.output,
         adjudication_output: adjudication.output,
         executor_model: adjudication.model || analysis.model,
-      });
+      }), submitting => { stage = submitting ? "complete" : "adjudication"; });
       await updateTask(claim.job_id, "completed", historyScope);
       return completed;
     } catch (error) {
@@ -253,7 +270,7 @@ export async function runStructuredProviderReview(
     executionTimeoutMs = DEFAULT_EXECUTION_TIMEOUT_MS,
     environment = process.env,
     signal,
-    readOnlyAnalysis = false,
+    readOnlyAnalysis = true,
     terminationGraceMs = 1_000,
     beforeDispatch = async () => {},
   } = {},
@@ -278,38 +295,54 @@ export async function runStructuredProviderReview(
       "Return only JSON that strictly matches the provided output schema.",
       request.prompt,
     ].join("\n\n");
-    await beforeDispatch();
-    if (signal?.aborted) throw new SqmLocalReviewError('Local analysis was cancelled.', { code: 'analysis_cancelled' });
-    const stdout = await spawnReview(
-      inspection.executable,
-      provider === "claude" ? [
-        "--print", "--output-format", "json",
-        "--json-schema", JSON.stringify(request.output_schema),
-        "--no-session-persistence", "--tools", "",
-        "--setting-sources", "", "--settings", '{"disableAllHooks":true}',
-        "--strict-mcp-config", "--mcp-config", '{"mcpServers":{}}',
-        ...(readOnlyAnalysis ? ["--system-prompt", request.system, "--max-turns", "1"] : []),
-      ] : [
-        "exec",
-        "--model",
-        SQM_CODEX_MODEL,
-        "--skip-git-repo-check",
-        "--sandbox",
-        "read-only",
-        "--ephemeral",
-        "--ignore-user-config",
-        "--ignore-rules",
-        ...(readOnlyAnalysis ? codexAnalysisArguments(request.system) : []),
-        "--output-schema",
-        schemaPath,
-        "--output-last-message",
-        outputPath,
-        "-",
-      ],
-      prompt,
-      { spawnImpl, executionTimeoutMs, label, cwd: workDir, environment, provider, signal, readOnlyAnalysis,
-        terminationGraceMs: Number.isInteger(terminationGraceMs) ? Math.max(1, Math.min(terminationGraceMs, 5_000)) : 1_000 },
-    );
+    const deadline = Date.now() + executionTimeoutMs;
+    const models = provider === "codex" ? SQM_CODEX_MODELS : [null];
+    let stdout;
+    let selectedModel;
+    for (const model of models) {
+      await beforeDispatch();
+      if (signal?.aborted) throw new SqmLocalReviewError('Local analysis was cancelled.', { code: 'analysis_cancelled' });
+      const remaining = deadline - Date.now();
+      if (remaining <= 0) throw new SqmLocalReviewError('Local analysis timed out.', { code: 'analysis_timeout' });
+      // Never consume a partial/stale output file left by a rejected model.
+      await rm(outputPath, { force: true });
+      try {
+        stdout = await spawnReview(
+          inspection.executable,
+          provider === "claude" ? [
+            "--print", "--output-format", "json",
+            "--json-schema", JSON.stringify(request.output_schema),
+            "--no-session-persistence", "--tools", "",
+            "--setting-sources", "", "--settings", '{"disableAllHooks":true}',
+            "--strict-mcp-config", "--mcp-config", '{"mcpServers":{}}',
+            ...(readOnlyAnalysis ? ["--system-prompt", request.system, "--max-turns", "1"] : []),
+          ] : [
+            "exec",
+            "--model",
+            model,
+            "--skip-git-repo-check",
+            "--sandbox",
+            "read-only",
+            "--ephemeral",
+            "--ignore-user-config",
+            "--ignore-rules",
+            ...(readOnlyAnalysis ? codexAnalysisArguments(request.system) : []),
+            "--output-schema",
+            schemaPath,
+            "--output-last-message",
+            outputPath,
+            "-",
+          ],
+          prompt,
+          { spawnImpl, executionTimeoutMs: remaining, label, cwd: workDir, environment, provider, signal, readOnlyAnalysis, model,
+            terminationGraceMs: Number.isInteger(terminationGraceMs) ? Math.max(1, Math.min(terminationGraceMs, 5_000)) : 1_000 },
+        );
+        selectedModel = model;
+        break;
+      } catch (error) {
+        if (error.code !== "provider_model_unsupported" || model === models.at(-1)) throw error;
+      }
+    }
     const envelope = provider === "claude" ? JSON.parse(stdout) : null;
     if (envelope?.is_error || (envelope && envelope.subtype !== "success")) {
       if (readOnlyAnalysis && envelope?.is_error === true && providerSignInRequired(envelope.result)) {
@@ -325,7 +358,7 @@ export async function runStructuredProviderReview(
     }
     return {
       output,
-      model: provider === "codex" ? SQM_CODEX_MODEL : `${provider}-local/${inspection.installed_version || "unknown"}`,
+      model: provider === "codex" ? selectedModel : `${provider}-local/${inspection.installed_version || "unknown"}`,
     };
   } catch (error) {
     if (error instanceof SqmLocalReviewError || error?.code === 'desktop_access_required') throw error;
@@ -350,7 +383,7 @@ function codexAnalysisArguments(system) {
     '-c', `developer_instructions=${JSON.stringify(system)}`];
 }
 
-function spawnReview(command, args, prompt, { spawnImpl, executionTimeoutMs, label, cwd, environment, provider, signal, readOnlyAnalysis, terminationGraceMs }) {
+function spawnReview(command, args, prompt, { spawnImpl, executionTimeoutMs, label, cwd, environment, provider, signal, readOnlyAnalysis, terminationGraceMs, model }) {
   return new Promise((resolve, reject) => {
     if (signal?.aborted) return reject(new Error('Local analysis was cancelled'));
     const child = spawnImpl(command, args, {
@@ -361,6 +394,7 @@ function spawnReview(command, args, prompt, { spawnImpl, executionTimeoutMs, lab
     });
     let stdout = "";
     let stderr = "";
+    let stderrPending = "";
     let settled = false;
     let terminalError = null;
     let killTimer = null;
@@ -396,9 +430,16 @@ function spawnReview(command, args, prompt, { spawnImpl, executionTimeoutMs, lab
         for (const line of stdout.split('\n').filter(Boolean)) {
           let event;
           try { event = JSON.parse(line); } catch { continue; }
-          if (['error', 'turn.failed'].includes(event.type) && providerSignInRequired(event.message ?? event.error?.message)) {
-            halt(new SqmLocalReviewError('Sign in to the local AI app again.', { code: 'provider_sign_in_required' }));
-            return;
+          if (['error', 'turn.failed'].includes(event.type)) {
+            const message = event.message ?? event.error?.message;
+            if (providerSignInRequired(message)) {
+              halt(new SqmLocalReviewError('Sign in to the local AI app again.', { code: 'provider_sign_in_required' }));
+              return;
+            }
+            if (providerModelUnsupported(message, model)) {
+              halt(modelUnsupportedError(model));
+              return;
+            }
           }
           if (/command_execution|mcp_tool_call|web_search|file_change|tool_call/.test(event.item?.type || '')) {
             halt(new Error('Read-only analysis attempted to use a tool'));
@@ -414,9 +455,32 @@ function spawnReview(command, args, prompt, { spawnImpl, executionTimeoutMs, lab
     child.stderr?.setEncoding?.("utf8");
     child.stderr?.on?.("data", (chunk) => {
       if (!readOnlyAnalysis) stderr = `${stderr}${chunk}`.slice(-4000);
+      if (provider === "codex") {
+        stderrPending = `${stderrPending}${chunk}`.slice(-8000);
+        const lines = stderrPending.split(/\r?\n/);
+        stderrPending = lines.pop() || "";
+        for (const line of lines) checkStderr(line);
+      }
     });
+    const checkStderr = line => {
+      // Native CLI diagnostics only; never match warnings or assistant text.
+      const match = line.match(/^\s*ERROR:\s*(.*)$/);
+      if (!match) return;
+      let message = match[1];
+      try {
+        const event = JSON.parse(message);
+        if (event.type !== "error") return;
+        message = event.error?.message ?? event.message;
+      } catch { /* Older CLIs print a plain native ERROR line. */ }
+      if (providerSignInRequired(message)) {
+        halt(new SqmLocalReviewError('Sign in to the local AI app again.', { code: 'provider_sign_in_required' }));
+      } else if (providerModelUnsupported(message, model)) {
+        halt(modelUnsupportedError(model));
+      }
+    };
     child.on("error", finish);
     child.on("close", (code) => {
+      if (provider === "codex") checkStderr(stderrPending);
       if (readOnlyAnalysis && provider === 'claude' && !terminalError) {
         let envelope;
         try { envelope = JSON.parse(stdout); } catch { /* Non-JSON failure remains unavailable. */ }
@@ -433,10 +497,22 @@ function spawnReview(command, args, prompt, { spawnImpl, executionTimeoutMs, lab
   });
 }
 
+function providerModelUnsupported(message, model) {
+  if (typeof message !== "string" || typeof model !== "string") return false;
+  const rejection = message.trim().match(/^The ['"]([^'"]+)['"] model is not supported(?: when using Codex with a ChatGPT account)?\.?$/i);
+  return rejection?.[1] === model;
+}
+
+function modelUnsupportedError(model) {
+  return new SqmLocalReviewError(`The local Codex account does not support ${model}. Update Codex or check the models available to this account.`, {
+    code: "provider_model_unsupported", status: 400,
+  });
+}
+
 function providerSignInRequired(message) {
   // Match explicit provider authentication failures only, never general 403s
   // (which can be permissions/quotas) or model-generated prose/tool output.
-  return typeof message === 'string' && /(?:not logged in|please (?:log|sign) in|invalid api key|authentication token (?:is )?(?:invalid|expired)|refresh token (?:has )?expired)/i.test(message.slice(0, 4000));
+  return typeof message === 'string' && /(?:\b401\s+Unauthorized\b|not logged in|please (?:log|sign) in|invalid api key|authentication token (?:is )?(?:invalid|expired)|refresh token (?:has )?expired)/i.test(message.slice(0, 4000));
 }
 
 function safeReviewEnvironment(environment, provider) {
@@ -496,6 +572,7 @@ function reviewFailure(error, stage, provider) {
 function formatFailure(failure, failureReported = null) {
   if (!failure) return "SQM analysis failed. No failure diagnostics were recorded.";
   const hint = failure.code === "provider_sign_in_required" ? "Sign in to the local AI app again."
+    : failure.code === "provider_model_unsupported" ? "The local Codex account rejected the configured SQM models. Update Codex or check this account's available models."
     : failure.code === "analysis_timeout" ? "The local AI analysis exceeded its time limit."
     : failure.code === "sqm_source_loading_failed" ? "SIMY could not load evidence for this analysis."
     : failure.status === 403 || failure.code === "desktop_access_required" ? "Check SQM and Developer access for the selected organization."