@lazyingart/agintiflow 0.20.210 → 0.20.212

package/docs/supervision-campaign-ledger.md

@@ -75,3 +75,24 @@ The incident established two general contracts:
   canvas/download filename derived from the task title and source purpose.
   Internal collision identifiers are short suffixes, never the leading or only
   visible filename information.
+
+### QA repair continuity and evidence intent
+
+`qa-incident-metrics-001` passed after two reusable runtime fixes. A normal,
+underspecified QA prompt led the DeepSeek-backed agent to reproduce and diagnose
+compound-duration parsing, percentile interpolation/mutation, and deterministic
+summary-order defects. The first partial patch advanced the mutation revision,
+but the runtime then forgot the retained failing test and prematurely reduced
+the tool surface to test-only mode. Source-next restored the failed-test repair
+state until a fresh current-revision test passed, allowing the same durable
+session to finish the coherent patch, add regressions, run 15 tests at 100%
+statement coverage, clean debris, and commit `667891f`.
+
+Independent `pytest` and the hidden `qa_incident_metrics_contract.py` checker
+both passed. The run also exposed an evidence-intent false positive: `figure
+out` and `clean up generated test debris` were interpreted as a request for a
+canvas artifact. Evidence inference now excludes those non-production phrases
+while retaining the artifact gate for real generated figures. Exact session
+evidence remains in
+`~/.agintiflow/sessions/aginti-qa-incident-metrics-001/events.jsonl` and the
+machine ledger records the run as `passed_after_fix`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lazyingart/agintiflow",
-  "version": "0.20.210",
+  "version": "0.20.212",
   "type": "module",
   "description": "AgInTiFlow is a project-aware agent workspace for hybrid wet-dry R&D, hardware-aware intelligence, software automation, and industrial workflows.",
   "license": "Apache-2.0",

package/scripts/smoke-cli-chat.js

@@ -74,6 +74,12 @@ function cellWidth(value = "") {
   return [...String(value || "").replace(/\x1b\[[0-9;?]*[ -/]*[@-~]/g, "")].reduce((sum, char) => sum + charCellWidth(char), 0);
 }
 
+function stripTerminalControls(value = "") {
+  return String(value || "")
+    .replace(/\x1b\][\s\S]*?(?:\x07|\x1b\\)/g, "")
+    .replace(/\x1b\[[0-9;?]*[ -/]*[@-~]/g, "");
+}
+
 function runChat(inputText) {
   return runCli(
     ["chat", "--provider", "mock", "--routing", "manual", "--profile", "code", "--headless", "--sandbox-mode", "host"],
@@ -596,7 +602,7 @@ try {
     animated: false,
     webAppUrl: "http://127.0.0.1:3210",
   }).join("\n");
-  if (!launchHeaderWithWeb.includes("webapp: http://127.0.0.1:3210")) {
+  if (!stripTerminalControls(launchHeaderWithWeb).includes("webapp: http://127.0.0.1:3210")) {
     throw new Error("launch header did not render the active webapp URL in the tagline row");
   }
   const launchHeaderWithWebError = buildLaunchHeaderLines({
@@ -749,17 +755,21 @@ try {
     throw new Error("terminal prompt layout did not render live input queue and cwd footer");
   }
   const hintPromptLayout = buildPromptLayout("/mo", 3, 90, 24, { suggestions: ["/models", "/model"], suggestionIndex: 1 });
-  const hintText = hintPromptLayout.renderedRows
-    .map((line) => line.replace(/\x1b\[[0-9;?]*[ -/]*[@-~]/g, ""))
-    .join("\n");
-  if (!hintText.includes("  user> /mo") || !hintText.includes("hint    /models  >/model")) {
-    throw new Error("terminal prompt layout did not align user and hint text columns");
+  const hintRaw = hintPromptLayout.renderedRows.join("\n");
+  const hintText = stripTerminalControls(hintRaw);
+  const selectedHintRendered =
+    hintText.includes("hint    /models  >/model") ||
+    (hintText.includes("hint    /models  /model") && hintRaw.includes("\x1b[1m/model\x1b[0m"));
+  if (!hintText.includes("  user> /mo") || !selectedHintRendered) {
+    throw new Error(`terminal prompt layout did not align user and hint text columns: ${JSON.stringify(hintText)}`);
   }
   const exactHintLayout = buildPromptLayout("/model", 6, 90, 24);
-  const exactHintText = exactHintLayout.renderedRows
-    .map((line) => line.replace(/\x1b\[[0-9;?]*[ -/]*[@-~]/g, ""))
-    .join("\n");
-  if (!exactHintText.includes("hint    >/model") || exactHintText.includes("/models")) {
+  const exactHintRaw = exactHintLayout.renderedRows.join("\n");
+  const exactHintText = stripTerminalControls(exactHintRaw);
+  const exactHintSelected =
+    exactHintText.includes("hint    >/model") ||
+    (exactHintText.includes("hint    /model") && exactHintRaw.includes("\x1b[1m/model\x1b[0m"));
+  if (!exactHintSelected || exactHintText.includes("/models")) {
     throw new Error("exact slash commands should not show broader prefix matches");
   }
   if (classifyEscapeAction({ active: false }) !== "noop") {

package/scripts/smoke-context-budget-recovery.js

@@ -52,6 +52,17 @@ const artifactContract = deriveScsTaskContract({
 });
 assert.equal(artifactContract.requiresExternalEvidence, true, "real chat artifact work lost its evidence gate");
 assert.ok(artifactContract.requiredEvidence.some((item) => item.category === "artifact"));
+const scopedArtifactRootContract = deriveScsTaskContract({
+  goal:
+    'AGINTI_EVIDENCE_SCOPE_JSON: {"mode":"task","request":"Create result.txt with the exact requested content.","artifact_root":"/tmp/labcanvas-task-artifacts"}',
+  taskProfile: "chatops",
+});
+assert.equal(scopedArtifactRootContract.artifactRoot, "/tmp/labcanvas-task-artifacts");
+assert.deepEqual(
+  scopedArtifactRootContract.exactOutputPaths,
+  ["/tmp/labcanvas-task-artifacts/result.txt"],
+  "a bare task artifact filename was not resolved against the host-declared artifact root"
+);
 
 let capturedPayload = null;
 const client = {

package/scripts/smoke-model-roles.js

@@ -986,6 +986,14 @@ const generatedReviewContract = deriveScsTaskContract({
   ].join("\n"),
   taskProfile: "review",
 });
+const qaCleanupContract = deriveScsTaskContract({
+  goal: "Inspect the project, figure out what is actually wrong, fix the failing tests, clean up generated test debris, and commit only the intentional fix.",
+  taskProfile: "qa",
+});
+const generatedFigureContract = deriveScsTaskContract({
+  goal: "Generate a publication-ready figure that compares the two repair strategies.",
+  taskProfile: "design",
+});
 assert(
   !explainCodeContract.requiresExternalEvidence,
   "code profile alone should not force external evidence for a pure explanation"
@@ -1037,6 +1045,16 @@ assert(
     !generatedReviewContract.requiredEvidence.some((item) => item.category === "artifact"),
   "review profile should not turn review-format boilerplate into file/artifact production requirements"
 );
+assert(
+  qaCleanupContract.requiredEvidence.some((item) => item.category === "command") &&
+    qaCleanupContract.requiredEvidence.some((item) => item.category === "git") &&
+    !qaCleanupContract.requiredEvidence.some((item) => item.category === "artifact"),
+  "cleaning generated test debris should not invent an unrelated artifact-delivery requirement"
+);
+assert(
+  generatedFigureContract.requiredEvidence.some((item) => item.category === "artifact"),
+  "tightening QA intent phrases removed the real generated-figure artifact gate"
+);
 const fileOnlyLedger = buildScsEvidenceLedger({
   context: { events: [{ type: "file.changed", data: { path: "src/app.js" } }] },
 });

package/scripts/smoke-progressive-tool-selection.js

@@ -725,6 +725,47 @@ const constrainedInstructionWrite = failedTestRepairWithRequiredInstruction.find
 assertStrict.deepEqual(constrainedInstructionWrite.function.parameters.properties.path.enum, ["AGINTI.md"]);
 assertStrict.deepEqual(constrainedInstructionWrite.function.parameters.properties.mode.enum, ["create"]);
 
+const continuedFailedTestRepairRuntime = nextStepRuntimeConfig(
+  { provider: "localllm", taskProfile: "qa" },
+  {
+    meta: {
+      projectVerification: {
+        mutationRevision: 1,
+        discoveredTests: ["python -m pytest -q"],
+        requiredOutputs: [],
+        testRuns: [
+          {
+            command: "python -m pytest -q",
+            mutationRevision: 0,
+            passed: false,
+            failureSignature: "baseline-failure",
+          },
+        ],
+      },
+    },
+    messages: [],
+  }
+);
+assertStrict.equal(
+  continuedFailedTestRepairRuntime.testFailureRepairActive,
+  true,
+  "a partial repair forgot the retained failed test after the mutation revision advanced"
+);
+assertStrict.equal(
+  continuedFailedTestRepairRuntime.testFailureCommand,
+  "python -m pytest -q",
+  "a partial repair lost the retained failing test command"
+);
+sameNames(
+  selectProgressiveTools(allTools, {
+    config: continuedFailedTestRepairRuntime,
+    goal: "Continue the coherent repair, then retest.",
+    profile: "qa",
+  }),
+  ["read_file", "search_files", "apply_patch", "run_command", "finish"],
+  "a partial repair was forced into test-only mode before the coherent patch was complete"
+);
+
 const pendingTestTools = selectProgressiveTools(allTools, {
   config: {
     provider: "localllm",

package/src/agent-runner.js

@@ -4239,13 +4239,23 @@ export function nextStepRuntimeConfig(config = {}, state = {}) {
   };
   const verification = state.meta?.projectVerification || {};
   const mutationRevision = Number(verification.mutationRevision || 0);
-  const latestCurrentTest = [...(verification.testRuns || [])]
+  const testRuns = Array.isArray(verification.testRuns) ? verification.testRuns : [];
+  const latestCurrentTest = [...testRuns]
     .reverse()
     .find((run) => Number(run.mutationRevision || 0) === mutationRevision);
-  if (latestCurrentTest && latestCurrentTest.passed !== true) {
+  const latestRecordedTest = [...testRuns]
+    .reverse()
+    .find((run) => String(run?.command || "").trim());
+  const retainedFailedTest =
+    latestCurrentTest && latestCurrentTest.passed !== true
+      ? latestCurrentTest
+      : !latestCurrentTest && latestRecordedTest?.passed === false
+        ? latestRecordedTest
+        : null;
+  if (retainedFailedTest) {
     runtimeConfig.testFailureRepairActive = true;
-    runtimeConfig.testFailureCommand = String(latestCurrentTest.command || "");
-    runtimeConfig.testFailureSignature = String(latestCurrentTest.failureSignature || "");
+    runtimeConfig.testFailureCommand = String(retainedFailedTest.command || "");
+    runtimeConfig.testFailureSignature = String(retainedFailedTest.failureSignature || "");
     const completedOutputs = new Set(
       (state.meta?.artifactProgress?.completed || [])
         .map((item) => String(item || "").replace(/\\/g, "/").replace(/^\.\//, ""))
@@ -4256,10 +4266,7 @@ export function nextStepRuntimeConfig(config = {}, state = {}) {
       .filter((item) => /(?:^|\/)(?:AGINTI|AGENTS)\.md$/i.test(item))
       .slice(0, 8);
   } else if (!latestCurrentTest && mutationRevision > 0 && (verification.discoveredTests || []).length) {
-    const retainedTestCommand = [...(verification.testRuns || [])]
-      .reverse()
-      .map((run) => String(run.command || ""))
-      .find(Boolean);
+    const retainedTestCommand = String(latestRecordedTest?.command || "");
     if (retainedTestCommand) {
       runtimeConfig.testVerificationPending = true;
       runtimeConfig.testVerificationCommand = retainedTestCommand;

package/src/scs-evidence.js

@@ -544,6 +544,12 @@ function requiresSourceGrounding(goal = "") {
 function inferRequirementCategories(goal = "", taskProfile = "", acceptanceCriteria = []) {
   const positiveGoal = stripForbiddenLanguage(goal);
   const text = normalizedText(positiveGoal);
+  const artifactSignalText = text
+    .replace(
+      /\b(?:clean(?:\s+up)?|remove|delete|clear|purge)\b[^.\n;]{0,120}\b(?:generated|temporary|stale|test)?\s*(?:test\s+)?(?:debris|caches?|byproducts?)\b/gi,
+      ""
+    )
+    .replace(/\bfigure\s+out\b/gi, "");
   const profile = String(taskProfile || "").toLowerCase();
   const categories = new Set(
     goalRequiresEvidence(positiveGoal, "") ? profileRequirementsForGoal(taskProfile, positiveGoal) : []
@@ -571,7 +577,7 @@ function inferRequirementCategories(goal = "", taskProfile = "", acceptanceCrite
   if (directCommandSignal || (validationSignal && codeProfileRequiresCommand(positiveGoal))) {
     categories.add("command");
   }
-  if (textHas(text, /\b(artifact|canvas|pdf|image|video|screenshot|cover|plot|chart|figure|docx|archive|copy to|export|generated|generate|draft)\b/) || /输出|产物|图片|视频|截图|封面|生成/.test(text)) {
+  if (textHas(artifactSignalText, /\b(artifact|canvas|pdf|image|video|screenshot|cover|plot|chart|figure|docx|archive|copy to|export|generated|generate|draft)\b/) || /输出|产物|图片|视频|截图|封面|生成/.test(artifactSignalText)) {
     categories.add("artifact");
   }
   if (textHas(text, /\b(browser|chrome|chromium|cdp|devtools|playwright|selenium|web[- ]?ui|website|page|tab|composer|click|type|upload|attach|submit|form)\b/) || /浏览器|网页|页面|上传|提交|附件|资产库/.test(text)) {
@@ -670,13 +676,27 @@ function stripForbiddenLanguage(goal = "") {
     .replace(/禁止([^。\n；]+)/g, "");
 }
 
+function parseAgintiEvidenceScope(goal = "") {
+  const matches = [
+    ...String(goal || "").matchAll(/^AGINTI_EVIDENCE_SCOPE_JSON:\s*(\{[^\n]+\})\s*$/gm),
+  ];
+  const match = matches.at(-1);
+  if (!match) return null;
+  try {
+    const payload = JSON.parse(match[1]);
+    return payload && typeof payload === "object" ? payload : null;
+  } catch {
+    return null;
+  }
+}
+
 export function hasAgintiEvidenceScope(goal = "") {
-  return /^AGINTI_EVIDENCE_SCOPE_JSON:\s*\{[^\n]+\}\s*$/m.test(String(goal || ""));
+  return Boolean(parseAgintiEvidenceScope(goal));
 }
 
 export function scopedChatopsEvidenceGoal(goal = "", taskProfile = "") {
-  const match = String(goal || "").match(/^AGINTI_EVIDENCE_SCOPE_JSON:\s*(\{[^\n]+\})\s*$/m);
-  if (!match) {
+  const payload = parseAgintiEvidenceScope(goal);
+  if (!payload) {
     const text = String(goal || "");
     const lines = text
       .split(/\r?\n/)
@@ -694,22 +714,32 @@ export function scopedChatopsEvidenceGoal(goal = "", taskProfile = "") {
     }
     return text;
   }
-  try {
-    const payload = JSON.parse(match[1]);
-    if (!payload || typeof payload !== "object") return String(goal || "");
-    const mode = String(payload.mode || "").trim().toLowerCase();
-    if (["chat-response", "host-managed-response", "plan-response", "read-only-answer"].includes(mode)) {
-      return "Answer the current chat turn directly without external execution.";
-    }
-    const request = String(payload.request || "").trim();
-    return request || String(goal || "");
-  } catch {
-    return String(goal || "");
+  const mode = String(payload.mode || "").trim().toLowerCase();
+  if (["chat-response", "host-managed-response", "plan-response", "read-only-answer"].includes(mode)) {
+    return "Answer the current chat turn directly without external execution.";
   }
+  const request = String(payload.request || "").trim();
+  return request || String(goal || "");
+}
+
+function scopedArtifactRoot(goal = "") {
+  const payload = parseAgintiEvidenceScope(goal);
+  if (!payload || String(payload.mode || "").trim().toLowerCase() !== "task") return "";
+  return String(payload.artifact_root || "").trim();
+}
+
+function applyScopedArtifactRoot(items = [], artifactRoot = "") {
+  if (!artifactRoot) return items;
+  return items.map((item) => {
+    const value = String(item || "").trim();
+    if (!value || path.isAbsolute(value) || value.startsWith("~/") || /[\\/]/.test(value)) return value;
+    return path.join(artifactRoot, value);
+  });
 }
 
 export function deriveScsTaskContract({ goal = "", taskProfile = "", acceptanceCriteria = [] } = {}) {
   const evidenceGoal = scopedChatopsEvidenceGoal(goal, taskProfile);
+  const artifactRoot = scopedArtifactRoot(goal);
   const requirementCategories = inferRequirementCategories(evidenceGoal, taskProfile, acceptanceCriteria);
   const requiredToolCalls = inferRequiredToolCalls(evidenceGoal);
   const requiresExternalEvidence = requirementCategories.length > 0 || requiredToolCalls.length > 0 || goalRequiresEvidence(evidenceGoal, taskProfile);
@@ -718,8 +748,11 @@ export function deriveScsTaskContract({ goal = "", taskProfile = "", acceptanceC
     category,
     description: CATEGORY_LABELS[category] || category,
   }));
-  const exactOutputPaths = inferExactOutputPaths(evidenceGoal);
-  const exactInputPaths = inferExactInputPaths(evidenceGoal).filter((item) => !exactOutputPaths.includes(item));
+  const inferredOutputPaths = inferExactOutputPaths(evidenceGoal);
+  const exactOutputPaths = applyScopedArtifactRoot(inferredOutputPaths, artifactRoot);
+  const exactInputPaths = inferExactInputPaths(evidenceGoal).filter(
+    (item) => !inferredOutputPaths.includes(item) && !exactOutputPaths.includes(item)
+  );
   const declaredSourceRoots = inferDeclaredSourceRoots(evidenceGoal);
   return {
     version: 1,
@@ -729,6 +762,7 @@ export function deriveScsTaskContract({ goal = "", taskProfile = "", acceptanceC
     requiredEvidence,
     forbiddenActions: inferForbiddenActions(evidenceGoal),
     exactOutputPaths,
+    artifactRoot,
     exactInputPaths,
     declaredSourceRoots,
     readOnlyReadiness: isReadOnlyReadinessTask(evidenceGoal),
@@ -786,11 +820,12 @@ export function augmentScsTaskContractWithProjectVerification(contract = {}, sta
       .map((item) => String(item?.path || item || "").trim())
       .filter(Boolean)
   ).slice(0, 80);
-  const requiredOutputs = unique(
+  const requiredOutputs = unique(applyScopedArtifactRoot(
     (Array.isArray(verification.requiredOutputs) ? verification.requiredOutputs : [])
       .map((item) => String(item || "").trim())
-      .filter(Boolean)
-  ).slice(0, 64);
+      .filter(Boolean),
+    String(contract.artifactRoot || "")
+  )).slice(0, 64);
   const requiredProjectCommands = unique(
     (Array.isArray(verification.requiredCommands) ? verification.requiredCommands : [])
       .map(normalizeProjectCommand)