@lazyingart/agintiflow 0.20.212 → 0.20.213

package/docs/supervision-campaign-ledger.md

@@ -96,3 +96,37 @@ 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`.
+
+### GitHub maintenance hidden acceptance and goal-scoped evidence
+
+`github-safe-maintenance-012` started from a normal maintenance prompt rather
+than a checker-shaped instruction. The agent repaired and verified the target,
+but its first completion omitted the exact `docs/maintenance-handoff.md` file
+required by independent acceptance. The same retained session inspected the
+external failure, created the missing handoff, committed and pushed target
+commit `36fa6c0`, and left `main` clean and synchronized. The hidden
+`github_maintenance_contract.py` checker then passed.
+
+Supervision of the repair exposed four runtime defects that could affect other
+profiles:
+
+- A test command wrapped as `command; echo "EXIT:$?"` could have shell exit zero
+  even when the real command failed. Explicit final `EXIT`, `STATUS`, or
+  `RESULT` probes are now parsed; a missing or nonzero marker is failing
+  evidence.
+- A genuinely new continuation could inherit completed artifact, SCS,
+  project-verification, and repair state from the prior goal. New goals now
+  clear only goal-scoped execution evidence while preserving conversation,
+  durable goal history, and goal-keyed research memory.
+- An acceptance sentence listing screenshots, PDFs, reports, or app launches
+  "as appropriate" could force irrelevant visual work. Optional evidence
+  examples no longer become mandatory categories.
+- Merely naming a read-only checker such as `contract.py` could force a file
+  artifact. File evidence now requires both mutation intent and a workspace
+  file/source target; a virtual canvas artifact remains artifact evidence only.
+
+The patched source resumed `aginti-github-maintenance-001`, invoked the hidden
+checker exactly once, performed no file, canvas, commit, or push side effect,
+and completed with a clean repository-state check. The full npm suite and the
+focused dynamic-budget, SCS/model-role, and web-canvas regressions pass for
+AgInTiFlow `0.20.213`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lazyingart/agintiflow",
-  "version": "0.20.212",
+  "version": "0.20.213",
   "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-dynamic-step-budget.js

@@ -26,6 +26,7 @@ import {
   recordProjectVerificationOutcome,
   recordExactOutputProgress,
   recordStaticDiscoveryProgress,
+  resetGoalScopedRuntimeState,
   resetStaticDiscoveryAfterContextLoss,
   rememberCompletedDeepResearch,
   repeatedNoProgressToolBlock,
@@ -37,6 +38,7 @@ import {
 import {
   augmentScsTaskContractWithProjectVerification,
   buildScsEvidenceLedger,
+  deriveScsTaskContract,
   evaluateScsEvidence,
 } from "../src/scs-evidence.js";
 import {
@@ -72,6 +74,128 @@ function toolMessage(payload) {
 
 try {
   assert(normalizeDynamicStepsMode("off") === "off", "dynamic mode off did not normalize");
+  const failedExitProbeState = { meta: {} };
+  const failedExitProbeResult = {
+    toolName: "run_command",
+    ok: true,
+    exitCode: 0,
+    args: { command: 'python -m unittest discover -s tests; echo "EXIT:$?"' },
+    stdout: "external contract did not pass\nEXIT:1\n",
+    stderr: "",
+  };
+  recordProjectVerificationOutcome(failedExitProbeState, failedExitProbeResult, {
+    commandCwd: workspace,
+    taskProfile: "qa",
+  });
+  assert(
+    failedExitProbeResult.projectTest?.passed === false &&
+      failedExitProbeResult.projectTest?.explicitExitStatus === 1,
+    "a wrapped nonzero test status was recorded as passing"
+  );
+  const missingExitProbeState = { meta: {} };
+  const missingExitProbeResult = {
+    toolName: "run_command",
+    ok: true,
+    exitCode: 0,
+    args: { command: 'python -m unittest discover -s tests; printf "EXIT=%s\\n" "$?"' },
+    stdout: "external contract output ended before the status marker\n",
+    stderr: "",
+  };
+  recordProjectVerificationOutcome(missingExitProbeState, missingExitProbeResult, {
+    commandCwd: workspace,
+    taskProfile: "qa",
+  });
+  assert(
+    missingExitProbeResult.projectTest?.passed === false &&
+      missingExitProbeResult.projectTest?.explicitExitStatus === null,
+    "a missing wrapped test status was recorded as passing"
+  );
+  const passingExitProbeState = { meta: {} };
+  const passingExitProbeResult = {
+    toolName: "run_command",
+    ok: true,
+    exitCode: 0,
+    args: { command: 'python -m unittest discover -s tests; echo "EXIT=$?"' },
+    stdout: "contract checks passed\nEXIT=0\n",
+    stderr: "",
+  };
+  recordProjectVerificationOutcome(passingExitProbeState, passingExitProbeResult, {
+    commandCwd: workspace,
+    taskProfile: "qa",
+  });
+  assert(
+    passingExitProbeResult.projectTest?.passed === true &&
+      passingExitProbeResult.projectTest?.explicitExitStatus === 0,
+    "a wrapped zero test status was not accepted"
+  );
+
+  const newGoalState = {
+    meta: {
+      artifactProgress: { complete: true },
+      completionEvidenceRepair: { attempts: 1 },
+      dataProjectWorkflow: { ready: true },
+      durableEvidenceCategories: ["file", "visual"],
+      durableGitActions: ["commit"],
+      durableGitEvidence: [{ action: "commit", goalRevision: 1 }],
+      failedTestRecoveryPacket: { content: "old failure" },
+      goalContract: { revision: 2 },
+      projectVerification: { mutationRevision: 4 },
+      scs: { taskContract: { exactOutputPaths: ["old-output.md"] } },
+      completedDeepResearch: [{ goalKey: "retained-other-goal" }],
+    },
+  };
+  const removedGoalState = resetGoalScopedRuntimeState(newGoalState);
+  assert(removedGoalState.includes("artifactProgress"), "new goal did not clear stale artifact progress");
+  assert(!newGoalState.meta.projectVerification, "new goal retained stale project verification");
+  assert(!newGoalState.meta.scs, "new goal retained the previous SCS task contract");
+  assert(!newGoalState.meta.durableEvidenceCategories, "new goal inherited completed evidence categories");
+  assert(newGoalState.meta.goalContract?.revision === 2, "new goal reset its durable goal contract");
+  assert(newGoalState.meta.completedDeepResearch?.length === 1, "new goal discarded goal-keyed research cache");
+
+  const optionalVisualContract = deriveScsTaskContract({
+    goal: "Repair the repository, verify it, commit, and push the intentional work.",
+    taskProfile: "github",
+    acceptanceCriteria: [
+      "Do not rely only on chat summaries; verify files, commands, screenshots, PDFs, reports, or app launches as appropriate.",
+    ],
+  });
+  assert(
+    !optionalVisualContract.requiredEvidence.some((item) => item.category === "visual"),
+    "an optional evidence example forced irrelevant visual validation"
+  );
+  const explicitVisualContract = deriveScsTaskContract({
+    goal: "Capture and inspect a screenshot of the repaired interface.",
+    taskProfile: "website",
+  });
+  assert(
+    explicitVisualContract.requiredEvidence.some((item) => item.category === "visual"),
+    "an explicit screenshot request lost visual validation"
+  );
+  const readOnlyCheckerContract = deriveScsTaskContract({
+    goal:
+      "Re-run /tmp/acceptance/github_maintenance_contract.py once, verify the repository, and do not edit, commit, or push anything.",
+    taskProfile: "github",
+  });
+  assert(
+    !readOnlyCheckerContract.requiredEvidence.some((item) => item.category === "file"),
+    "a read-only checker path was mistaken for a requested file change"
+  );
+  const canvasOnlyContract = deriveScsTaskContract({
+    goal: "Create a canvas artifact preview for this smoke test.",
+  });
+  assert(
+    canvasOnlyContract.requiredEvidence.some((item) => item.category === "artifact") &&
+      !canvasOnlyContract.requiredEvidence.some((item) => item.category === "file"),
+    "a virtual canvas artifact was mistaken for a workspace-file mutation"
+  );
+  const sourceRepairContract = deriveScsTaskContract({
+    goal: "Fix src/runtime.py and verify the focused tests.",
+    taskProfile: "python",
+  });
+  assert(
+    sourceRepairContract.requiredEvidence.some((item) => item.category === "file"),
+    "an explicit source repair lost its file-change evidence gate"
+  );
   assert(
     completionEvidenceNeedsCommand({ missingProjectCommands: ["python analysis.py"] }),
     "a pending canonical command did not reopen command execution"

package/src/agent-runner.js

@@ -2423,6 +2423,28 @@ async function finishWithDirectAnswer({ config, state, store, observers, session
   };
 }
 
+export function resetGoalScopedRuntimeState(state = {}) {
+  state.meta = state.meta || {};
+  const keys = [
+    "artifactProgress",
+    "completionEvidenceRepair",
+    "dataProjectWorkflow",
+    "durableEvidenceCategories",
+    "durableGitActions",
+    "durableGitEvidence",
+    "failedTestRecoveryPacket",
+    "projectVerification",
+    "scs",
+  ];
+  const removed = [];
+  for (const key of keys) {
+    if (!(key in state.meta)) continue;
+    delete state.meta[key];
+    removed.push(key);
+  }
+  return removed;
+}
+
 async function applyContinuationPrompt(state, config, observers) {
   if (!config.resume || !config.goal) return null;
 
@@ -2439,6 +2461,9 @@ async function applyContinuationPrompt(state, config, observers) {
   state.meta = state.meta || {};
   const preserveTaskBoundary = preservesCurrentTaskBoundary(state, config.goal);
   const goalUpdate = updateGoalContract(state, config.goal, { preserveTaskBoundary });
+  if (!preserveTaskBoundary) {
+    resetGoalScopedRuntimeState(state);
+  }
   if (
     preserveTaskBoundary &&
     continuationAddsConcreteRequirement(config.goal) &&
@@ -2961,6 +2986,19 @@ function commandReportsTestFailure(result = {}) {
   );
 }
 
+function explicitExitProbeStatus(command = "", result = {}) {
+  const normalizedCommand = normalizeProjectCommand(command);
+  const hasExitProbe = /(?:^|;)\s*(?:echo|printf)\b[^;&|]*(?:EXIT|STATUS|RESULT)[^;&|]*\$\?[^;&|]*$/i.test(
+    normalizedCommand
+  );
+  if (!hasExitProbe) return { present: false, status: null };
+
+  const output = `${String(result.stdout || "")}\n${String(result.stderr || "")}`;
+  const matches = [...output.matchAll(/(?:^|\n)\s*(?:EXIT|STATUS|RESULT)(?:_CODE)?\s*[:=]\s*(-?\d+)\s*(?=\n|$)/gim)];
+  if (!matches.length) return { present: true, status: null };
+  return { present: true, status: Number(matches.at(-1)[1]) };
+}
+
 function actionableTestWarnings(result = {}) {
   const output = redactSensitiveText(`${String(result.stderr || "")}\n${String(result.stdout || "")}`);
   const warnings = [];
@@ -3133,17 +3171,24 @@ export function recordProjectVerificationOutcome(state = {}, toolResult = {}, co
 
   if (toolName === "run_command") {
     const command = normalizeProjectCommand(toolResult.args?.command || "");
+    const exitProbe = explicitExitProbeStatus(command, toolResult);
     const run = {
       command,
       at: now,
-      ok: toolResult.ok !== false && Number(toolResult.exitCode ?? 0) === 0,
+      ok:
+        toolResult.ok !== false &&
+        Number(toolResult.exitCode ?? 0) === 0 &&
+        (!exitProbe.present || exitProbe.status === 0),
       mutationRevision: verification.mutationRevision,
+      ...(exitProbe.present ? { explicitExitStatus: exitProbe.status } : {}),
     };
     verification.commandRuns = [...verification.commandRuns, run].slice(-40);
     toolResult.projectMutationRevision = verification.mutationRevision;
     if (isSubstantiveTestCommand(command)) {
       const zeroTests = commandReportsZeroTests(toolResult);
-      const reportedFailure = commandReportsTestFailure(toolResult);
+      const reportedFailure =
+        commandReportsTestFailure(toolResult) ||
+        (exitProbe.present && exitProbe.status !== 0);
       const qualityWarnings = actionableTestWarnings(toolResult);
       const passed = run.ok && !zeroTests && !reportedFailure && qualityWarnings.length === 0;
       const failedEvidence = passed ? {} : compactFailedTestEvidence(toolResult, config);

package/src/scs-evidence.js

@@ -487,6 +487,31 @@ function codeProfileRequiresCommand(goal = "") {
   return substantiveCodeWork && !simpleDocumentWrite;
 }
 
+function goalRequestsWorkspaceMutation(goal = "") {
+  const text = normalizedText(goal);
+  return (
+    /\b(?:append|build|convert|copy|create|delete|edit|fix|generate|implement|modify|move|patch|refactor|remove|rename|repair|replace|rewrite|save|update|write)\b/.test(
+      text
+    ) ||
+    /创建|写入|编辑|修复|实现|修改|更新|生成|保存|复制|移动|转换|删除|重命名|替换|追加/.test(text)
+  );
+}
+
+function goalRequestsFileMutation(goal = "") {
+  const text = normalizedText(goal);
+  if (!goalRequestsWorkspaceMutation(text)) return false;
+  return (
+    /\b(?:code|codebase|document(?:ation)?|file|notes?|path|readme|repo(?:sitory)?|script|source|workspace)\b/.test(
+      text
+    ) ||
+    /(?:^|[\s`'"(])(?:\.{0,2}\/|\/)?[a-z0-9_.-]+(?:\/[a-z0-9_.{}-]+)+/i.test(text) ||
+    /\.(?:c|cc|cpp|cs|css|csv|go|h|hpp|html?|java|js|jsx|json|kt|md|mjs|php|py|rb|rs|sh|swift|tex|ts|tsx|txt|ya?ml)\b/i.test(
+      text
+    ) ||
+    /文件|文档|代码|代码库|仓库|脚本|源码|路径|工作区|说明书|笔记/.test(text)
+  );
+}
+
 function profileRequirementsForGoal(taskProfile = "", goal = "") {
   const profile = String(taskProfile || "").toLowerCase();
   const defaults = PROFILE_REQUIREMENTS[profile] || [];
@@ -510,8 +535,14 @@ function profileRequirementsForGoal(taskProfile = "", goal = "") {
     "aaps",
   ]);
   if (!codeLikeProfiles.has(profile)) return defaults;
-  if (codeProfileRequiresCommand(goal)) return defaults;
-  return defaults.filter((category) => category !== "command");
+  let requirements = [...defaults];
+  if (!goalRequestsWorkspaceMutation(goal)) {
+    requirements = requirements.filter((category) => category !== "file");
+  }
+  if (!codeProfileRequiresCommand(goal)) {
+    requirements = requirements.filter((category) => category !== "command");
+  }
+  return requirements;
 }
 
 function isReadOnlyReadinessTask(goal = "") {
@@ -550,18 +581,16 @@ function inferRequirementCategories(goal = "", taskProfile = "", acceptanceCrite
       ""
     )
     .replace(/\bfigure\s+out\b/gi, "");
+  const mandatoryEvidenceText = artifactSignalText.replace(
+    /[^.\n]{0,240}\b(?:as appropriate|if appropriate|when useful|where applicable)\b/gi,
+    " "
+  );
   const profile = String(taskProfile || "").toLowerCase();
   const categories = new Set(
     goalRequiresEvidence(positiveGoal, "") ? profileRequirementsForGoal(taskProfile, positiveGoal) : []
   );
 
-  if (
-    textHas(
-      text,
-      /\b(file|path|workspace|edit|patch|fix|repair|refactor|convert|copy|move|remove|delete|source|script|code)\b|\.(?:md|txt|js|jsx|ts|tsx|mjs|cjs|json|ya?ml|py|tex|html|css|svg|csv)\b|\b(?:markdown|json|yaml|html|css|tex|latex)\s+file\b|\bfile\s+(?:as|in)\s+(?:markdown|json|yaml|html|css|tex|latex)\b/
-    ) ||
-    /文件|写入文件|编辑|修复|转换|复制|移动|删除|脚本|代码/.test(text)
-  ) {
+  if (goalRequestsFileMutation(positiveGoal)) {
     categories.add("file");
   }
   const directCommandSignal =
@@ -577,13 +606,13 @@ function inferRequirementCategories(goal = "", taskProfile = "", acceptanceCrite
   if (directCommandSignal || (validationSignal && codeProfileRequiresCommand(positiveGoal))) {
     categories.add("command");
   }
-  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)) {
+  if (textHas(mandatoryEvidenceText, /\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(mandatoryEvidenceText)) {
     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)) {
+  if (textHas(mandatoryEvidenceText, /\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(mandatoryEvidenceText)) {
     categories.add("browser");
   }
-  if (textHas(text, /\b(screenshot|visible|visual|see|inspect image|open image|read_image|thumbnail)\b/) || /截图|可见|缩略图/.test(text)) {
+  if (textHas(mandatoryEvidenceText, /\b(screenshot|visible|visual|see|inspect image|open image|read_image|thumbnail)\b/) || /截图|可见|缩略图/.test(mandatoryEvidenceText)) {
     categories.add("visual");
   }
   if (