@aefree/pi-unity 0.12.0 → 0.13.0

package/CHANGELOG.md

@@ -7,6 +7,32 @@ and this project follows semantic versioning for public package releases.
 
 ## Unreleased
 
+## 0.13.0 - 2026-09-10
+
+### Added
+
+- Added advertised `get_runtime_pipeline_settings` inspection and the bounded `unity_pipeline_run_script` tool for Pipeline 0.6 ephemeral in-memory compilation/execution of one existing project C# file. Hotpatch is intentionally out of scope.
+
+### Fixed
+
+- Preserve bounded Unity CLI discovery warnings/info and treat warning-bearing discovery as uncertainty rather than a confirmed absent Pipeline or launch-safe signal. Force human CLI version formatting so inherited `UNITY_FORMAT` cannot corrupt version discovery.
+- Accept Pipeline 0.6 compact response envelopes and warnings without treating warnings as compiler errors. Surface ambiguous busy responses rather than blindly redispatching commands that may be blocked by a modal dialog.
+
+## 0.12.1 - 2026-09-10
+
+### Security
+
+- Update the development/test Pi baseline to 0.85.1 and refresh the lockfile to resolve patched `brace-expansion` 5.0.9 and `undici` 8.9.0. Consumer-managed Pi hosts must be updated separately; peer compatibility ranges are unchanged.
+
+### Fixed
+
+- Compare existing Windows project aliases by canonical filesystem path, preventing connected project discovery from treating short paths or junction aliases as a different closed project.
+- Preserve bounded, redacted CLI stdout/stderr on failed Pipeline inspections and evals, including actionable package-version errors; retain native failure, timeout uncertainty and single-dispatch behavior.
+- Clarify connected single-selector arguments and pre-dispatch rejection recipes; document serial independent-fixture calls with stop-on-uncertainty boundaries, backed by offline registered-tool contract tests (not model-planning or live-Editor evidence).
+- Validate normalized JSON test artifacts as standalone inspection evidence, separate inspection success from test outcome, and reject missing explicit paths or conflicting artifacts without substituting unrelated latest logs.
+- Mark rejected Pipeline eval and inspection as native Pi tool failures while retaining structured diagnostics and uncertain post-dispatch effects, without retry or fallback.
+- Reject contradictory combined XML counters and test-record evidence during artifact inspection, including self-closing failures and records beyond the display limit; keep skipped and inconclusive counters distinct and preserve valid partial records.
+
 ## 0.12.0 - 2026-08-27
 
 ### Changed

package/README.md

@@ -35,17 +35,20 @@ Use these tools with an already-open exact Unity project copy that has a reachab
 - `unity_pipeline_recompile` — recompile through Pipeline with exact-copy preflight, bounded polling, and compact compiler evidence.
 - `unity_run_tests` — one intent-oriented EditMode or PlayMode workflow. It reuses compatible connected Pipeline execution or selects isolated `unity test` when the exact project copy is closed.
 - `unity_pipeline_eval` — execute bounded project-specific C# through Pipeline's Roslyn REPL. It accepts `timeoutSeconds` from 1–86,400 seconds; a timeout is uncertain and does not cancel or retry Editor work.
-- `unity_pipeline_inspect` — dispatch supported package-owned inspection commands and return structured evidence.
+- `unity_pipeline_inspect` — dispatch supported package-owned inspection commands (including read-only `get_runtime_pipeline_settings`) and return structured evidence. Runtime settings are refused by Pipeline in Play Mode; pi-unity never exits Play Mode to read them.
+- `unity_pipeline_run_script` — compile one existing project `.cs` file in Pipeline's ephemeral in-memory mode and invoke a named static entry point; supports bounded JSON arguments and compile-only `dryRun`. It deliberately does not expose hotpatch.
 
 Connected recompilation follows Unity's Script Changes While Playing policy and never preemptively sends `editor_stop`. Connected tests may exit Play Mode through advertised `editor_stop` when necessary, then verify Edit Mode before dispatch. Play Mode exit is allowed by default; `/unity-playmode-exit allow|disallow|status` controls the current session.
 
-A timeout is uncertain: work may still be running. The tools do not silently cancel, retry, launch another Editor, or switch to batchmode. The sole exception is Pipeline 0.5's explicit initial-settling `Server Busy` rejection for `unity_pipeline_recompile` and `unity_run_tests`, which is known not to have dispatched a main-thread command and is retried only within the configured deadline.
+A timeout is uncertain: work may still be running. The tools do not silently cancel, retry, launch another Editor, or switch to batchmode. Pipeline 0.6 can report busy for a modal dialog as well as startup settling, so ambiguous busy responses are surfaced and never blindly retried.
 
 ### Editor and batchmode
 
 - `unity_open_editor` — open the Unity Editor GUI. Pass `automated: true` to add the Unity Editor `-automated` flag; this is distinct from the Unity CLI's own `--non-interactive` option.
 - `unity_launch_batchmode` — run a bounded batchmode command through Unity CLI or the direct Editor executable.
-- `unity_inspect_artifacts` — summarize existing Unity Test Framework XML and Unity logs without launching Unity.
+- `unity_inspect_artifacts` — validate existing normalized JSON test artifacts, Unity Test Framework XML and Unity logs without launching Unity. `details.status` describes inspection; `details.testOutcome` describes the tests, including failures and uncertainty. A valid failed-test artifact is a successful inspection, not a passing run.
+
+Pass `normalizedResultPath` for standalone JSON evidence. Any explicit artifact path disables implicit latest-file selection; missing requested files fail inspection. With all paths omitted, `latestFromLogs` discovers recent files but cannot establish current-run identity. Mixed JSON/XML evidence must agree; without an explicit native artifact link, matching counts alone leave correlation uncertain.
 
 Use `unity_run_tests` for all ordinary Unity Test Framework work. It writes a durable normalized JSON result under `Logs/`; isolated `unity test` runs retain requested native reports. Connected Pipeline is selected only for compatible requests. A reachable Editor is never closed automatically to obtain isolated-only features.
 
@@ -82,6 +85,7 @@ Operation-specific recovery belongs to the operational skill. `unity-debugging`
 | Existing failed-run artifacts | `unity_inspect_artifacts` |
 | Project-specific C# query or operation | `unity_pipeline_eval` |
 | Supported structured project inspection | `unity_pipeline_inspect` |
+| Explicitly requested existing C# builder script | `unity_pipeline_run_script` |
 | Open the GUI explicitly | `unity_open_editor` or `/unity-open` |
 
 Pass an explicit project `path` when multiple copies may be discovered. Pipeline routing compares canonical paths so similarly named copies are not treated as interchangeable.
@@ -95,7 +99,7 @@ The connected compile and test tools:
 - poll internally with fixed deadlines and bounded backoff;
 - reject malformed or semantically failing nested results;
 - require a known positive test count and zero failures before reporting a pass;
-- discard passing-test records while retaining bounded failure diagnostics;
+- keep routine tool output compact while preserving connected test records in durable normalized JSON evidence;
 - detect pre-existing or clearly displaced test runs when available correlation fields permit it.
 
 Another connected client is not a project lock. When Pipeline returns stable correlation fields, conflicting status is reported as displaced and uncertain. If Pipeline omits stable run identity, a competing same-mode, same-filter run may be indistinguishable from the requested run; the tool cannot prove exclusive ownership from shared Editor status alone.
@@ -111,6 +115,12 @@ Another connected client is not a project lock. When Pipeline returns stable cor
 
 Use `unity_pipeline_inspect` when a purpose-built structured command fits. Use eval for bounded project-specific work that matches the user's intent. Prefer typed tools when they provide stronger lifecycle, polling, validation, or recovery semantics.
 
+### Pipeline run_script
+
+`unity_pipeline_run_script` is arbitrary code execution, not a sandbox or read-only inspection. Use it only when the user explicitly asks to run that existing file; obtain explicit authorization for lifecycle, settings, asset, build, test, or destructive mutations. It validates the exact project copy and advertised command twice, accepts a single existing `.cs` file inside the project, uses only `mode: ephemeral`, and never launches, saves, cancels, retries, falls back, hotpatches, or exits Play Mode. `dryRun: true` compiles without loading or executing the assembly.
+
+Rejected eval and inspection results are native Pi tool failures (`isError: true`) via the documented `tool_result` middleware, with structured rejection codes and bounded diagnostics retained. Pre-dispatch rejection does not execute the command; a timeout or dispatch failure can leave effects uncertain and never triggers retry or fallback.
+
 ## Launch and process safeguards
 
 `unity_open_editor` and batchmode tools treat Unity CLI as the authoritative launcher: `unity open`, `unity run`, and `unity test` receive only the selected project and let Unity CLI read `ProjectVersion.txt`. The direct Editor executable is an exceptional compatibility fallback used only when Unity CLI is unavailable. Set `launcher` to `auto`, `unity-cli`, or `editor-executable` when explicit routing is needed.
@@ -226,7 +236,9 @@ The registry-clean `package-lock.json` is committed. Optional development packag
 
 ## Unity Pipeline project side effect
 
-Starting `com.unity.pipeline@0.3.1-exp.1` assigns `Application.runInBackground = true`, which Unity persists as `PlayerSettings.runInBackground` in `ProjectSettings/ProjectSettings.asset`. Review that tracked change alongside `manifest.json` and `packages-lock.json` when installing Pipeline in a Unity project.
+Legacy Pipeline releases (including `0.3.1-exp.1`) assigned `Application.runInBackground = true`, which Unity persisted as `PlayerSettings.runInBackground` in `ProjectSettings/ProjectSettings.asset`. Pipeline 0.6 restores the original value around server start/stop. Review tracked settings changes when installing older releases; Pipeline 0.6 reports non-automated Editor state as descriptor `info` rather than a console warning.
+
+Pipeline 0.6 records local editor eval usage in `Library/Pipeline/eval-usage.jsonl`; raw source is not stored unless the Pipeline **Store Eval Source** setting is enabled. `UNITY_NO_CLI_INVOKED_TELEMETRY=1` opts out only of Unity CLI's per-invocation telemetry event; analytics and crash-reporting controls are separate and unchanged. Pipeline compact HTTP responses are accepted whether returned directly or inside the CLI's documented `data` wrapper; this does not assert that CLI wrapping has changed. For installed CLI/skill information, use read-only `unity skill show --list` or `unity skill show --path <path>`.
 
 ## License
 

package/index.ts

@@ -7,9 +7,9 @@ import { setTimeout as delay } from "node:timers/promises";
 import { getKeybindings, Text, truncateToWidth } from "@earendil-works/pi-tui";
 import {
   buildUnityBatchmodeAgentText,
-  deriveUnityArtifactInspectionStatus,
   deriveUnityBatchmodeStatus,
   hasKnownPositiveExecutedTestCount,
+  hasConflictingUnityXmlTestEvidence,
   loadUnityBatchmodeArtifacts,
   parseUnityBatchmodeInvocation,
   parseUnityTestResultsXml,
@@ -20,7 +20,7 @@ import {
   type UnityParsedTestResults,
 } from "./src/unity-batchmode";
 import { formatPathForUser, hasUnityCommandLineFlag } from "./src/unity-core";
-import { createUnityCliBatchmodeReportArgs, createUnityCliEditorExitCommand, createUnityCliRunCommand, createUnityCliTestCommand, dispatchUnityPlanningInspection, haveSameKnownProcessIds, inspectUnityCliProjectCapabilities, listRunningUnityCliEditorsForProject, resolveUnityCliCommand, UNITY_PLANNING_READ_COMMANDS, type UnityCliProjectCapabilities } from "./src/unity-cli";
+import { createUnityCliBatchmodeReportArgs, createUnityCliEditorExitCommand, createUnityCliRunCommand, createUnityCliTestCommand, dispatchUnityPipelineRunScript, dispatchUnityPlanningInspection, haveSameKnownProcessIds, inspectUnityCliProjectCapabilities, listRunningUnityCliEditorsForProject, resolveUnityCliCommand, UNITY_PLANNING_READ_COMMANDS, type UnityCliProjectCapabilities } from "./src/unity-cli";
 import { launchUnityCliOpenDetached } from "./src/unity-launch";
 import { createUnityBatchmodeCommand, launchUnityEditorDetached, resolveUnityEditorPath } from "./src/unity-editor-fallback";
 import { loadPiUnitySettings, type PiUnitySettings } from "./src/pi-unity-settings";
@@ -29,6 +29,8 @@ import { assertUnityProjectNotBusy, evaluateUnityLaunchSafety, getUnityNativeLoc
 import { readUnityVersion, resolveUnityProjectCandidates, type UnityProjectCandidate } from "./src/unity-projects";
 import { createUnityTestBatchPlan, type UnityTestBatchPlan, type UnityTestPlatform } from "./src/unity-test-batch";
 import { applyUnityCliRetrySummary, compactUnityTestSummary, defaultUnityTestReportFormats, deriveUnityCliEffectiveReportPath, determineUnityTestOutcome, getUnityTestRouteRequirements, normalizeUnityRunTestsRequest, parseUnityCliRetrySummary, resolveUnityCliBackendReportPaths, writeNormalizedUnityTestArtifact, type NormalizedUnityTestResult, type UnityRunTestsRequest } from "./src/unity-tests";
+import { validateNormalizedUnityTestArtifact } from "./src/unity-artifact-inspection";
+import { UNITY_TEST_MAX_ARTIFACT_BYTES } from "./src/unity-tests";
 import { auditUnityGuidance, type UnityGuidanceAuditResult } from "./src/unity-guidance-audit";
 import { runUnityPipelineRecompile, runUnityPipelineTests, type UnityPipelineOperationDetails } from "./src/unity-pipeline";
 import {
@@ -68,7 +70,7 @@ const GUI_WARNING = "This launches the full Unity Editor GUI and is not the same
 const SINGLE_PROCESS_WARNING = "Unity allows only one process per project folder. GUI Editor and batchmode/headless both count as that one process.";
 
 type UnityToolDetails = {
-  mode: "gui" | "batchmode" | "status" | "artifacts" | "pipeline_inspection" | "pipeline_eval" | "pipeline" | "tests";
+  mode: "gui" | "batchmode" | "status" | "artifacts" | "pipeline_inspection" | "pipeline_eval" | "pipeline_run_script" | "pipeline" | "tests";
   projectRoot: string;
   unityVersion: string;
   editorPath: string;
@@ -83,6 +85,10 @@ type UnityToolDetails = {
   invocation?: UnityBatchmodeInvocation;
   artifacts?: UnityBatchmodeArtifacts;
   parsedTestResults?: UnityParsedTestResults | null;
+  testOutcome?: NormalizedUnityTestResult["outcome"];
+  normalizedResultPath?: string;
+  normalizedResult?: Pick<NormalizedUnityTestResult, "source" | "platform" | "outcome" | "summary"> & { testRecordCount: number };
+  evidenceWarnings?: string[];
   status?: "passed" | "failed" | "killed";
   launcher?: "unity-cli" | "editor-executable";
   cliArgs?: string[];
@@ -95,6 +101,7 @@ type UnityToolDetails = {
   cliCapabilities?: UnityCliProjectCapabilities;
   pipelineInspection?: { outcome: "dispatched"; command: string; output: string; truncated: boolean } | { outcome: "rejected"; code: string; message: string };
   pipelineEval?: { outcome: "dispatched"; command: string; output: string; truncated: boolean } | { outcome: "rejected"; code: string; message: string };
+  pipelineRunScript?: { outcome: "dispatched"; command: string; output: string; truncated: boolean } | { outcome: "rejected"; code: string; message: string };
   pipeline?: UnityPipelineOperationDetails;
 };
 
@@ -118,11 +125,13 @@ const LAUNCH_BATCHMODE_PARAMS = Type.Object({
   closeBlockingUnityProcess: Type.Optional(Type.Boolean({ default: false, description: "When true, pi-unity may close a running Unity process for the resolved project before launch, but only if piUnity.allowCloseRunningUnityProcess is enabled in Pi settings. The process is selected by project matching, not by model-supplied PID." })),
 }, { additionalProperties: false });
 
+const CONNECTED_TEST_SELECTOR_GUIDANCE = 'unity_run_tests connected execution supports one testFilters entry OR one testCategories entry per call. For independent, non-overlapping selections, use separate execution: "connected" calls with the same explicit path and platform; await and inspect each passing result before issuing the next. Stop the remaining sequence on failure or uncertainty; never retry, broaden the selection, close the Editor, or switch routes automatically. Do not split a mixed filter/category intersection into separate runs.';
+
 const RUN_TESTS_PARAMS = Type.Object({
   path: Type.Optional(Type.String({ description: "Unity project path, workspace copy root, or folder containing project copies." })),
   testPlatform: StringEnum(["EditMode", "PlayMode"] as const),
-  testFilters: Type.Optional(Type.Array(Type.String(), { maxItems: 50 })),
-  testCategories: Type.Optional(Type.Array(Type.String(), { maxItems: 50 })),
+  testFilters: Type.Optional(Type.Array(Type.String(), { maxItems: 50, description: "Connected: one test-name selector only; omit testCategories. For two independent fixtures, issue serial single-selector calls, inspecting each result first. No semicolon lists. Omitted/empty selectors select all tests, not a repair for rejected filters." })),
+  testCategories: Type.Optional(Type.Array(Type.String(), { maxItems: 50, description: "Connected: one category only; omit testFilters. Multiple categories require separate non-overlapping selections or deliberate isolated execution; never split a filter/category intersection into separate runs." })),
   execution: Type.Optional(StringEnum(["auto", "connected", "isolated"] as const, { default: "auto" })),
   isolatedLauncher: Type.Optional(StringEnum(["auto", "unity-cli", "editor-executable"] as const, { default: "auto" })),
   retries: Type.Optional(Type.Integer({ minimum: 0, maximum: 20, default: 0 })),
@@ -159,6 +168,15 @@ const PIPELINE_EVAL_PARAMS = Type.Object({
   timeoutSeconds: Type.Optional(Type.Integer({ minimum: 1, maximum: 86400, default: 12, description: "Connected eval deadline in seconds (maximum 24 hours). A timeout is uncertain and does not retry or cancel Unity work." })),
 }, { additionalProperties: false });
 
+const PIPELINE_RUN_SCRIPT_PARAMS = Type.Object({
+  path: Type.Optional(Type.String({ maxLength: 1000, description: "Unity project path, workspace copy root, or folder containing project copies." })),
+  file: Type.String({ minLength: 1, maxLength: 1000, description: "Existing .cs file under the selected project root. The file is compiled in memory; it is not written or imported." }),
+  entry: Type.Optional(Type.String({ minLength: 1, maxLength: 500, description: "Named static entry point. Omit only when Pipeline can select Main or an unambiguous public static method." })),
+  args: Type.Optional(Type.Array(Type.Any(), { maxItems: 32, description: "Bounded JSON arguments coerced by Pipeline to the static entry-point parameter types." })),
+  dryRun: Type.Optional(Type.Boolean({ default: false, description: "Compile only; do not load the assembly or invoke the entry point." })),
+  timeoutSeconds: Type.Optional(Type.Integer({ minimum: 1, maximum: 86400, default: 30, description: "Connected run_script deadline. A timeout is uncertain and does not cancel the script." })),
+}, { additionalProperties: false });
+
 const PIPELINE_INSPECTION_PARAMS = Type.Object({
   path: Type.Optional(Type.String({ description: "Unity project path, workspace copy root, or folder containing project copies." })),
   command: StringEnum(UNITY_PLANNING_READ_COMMANDS, { description: "An advertised package-owned Pipeline inspection command." }),
@@ -178,7 +196,7 @@ const INSPECT_ARTIFACTS_PARAMS = Type.Object({
   testResultsPath: Type.Optional(Type.String({ description: "Unity Test Framework XML results path. Relative paths are resolved against cwd and the Unity project root." })),
   normalizedResultPath: Type.Optional(Type.String({ description: "pi-unity normalized JSON test artifact path." })),
   logFilePath: Type.Optional(Type.String({ description: "Unity log file path. Relative paths are resolved against cwd and the Unity project root." })),
-  latestFromLogs: Type.Optional(Type.Boolean({ default: true, description: "When paths are omitted, inspect the newest .xml and .log files under the project's Logs folder." })),
+  latestFromLogs: Type.Optional(Type.Boolean({ default: true, description: "Only when all artifact paths are omitted, inspect the newest .json, .xml and .log files under Logs. Latest files are not proof of a shared run." })),
   maxLines: Type.Optional(Type.Integer({ minimum: 1, maximum: 500, default: 60, description: "Maximum log/output lines to include." })),
   maxChars: Type.Optional(Type.Integer({ minimum: 500, maximum: 20000, default: 6000, description: "Maximum log/output characters to include." })),
 });
@@ -705,7 +723,8 @@ async function buildArtifactInspectionReport(
   candidate: UnityProjectCandidate,
   params: { testResultsPath?: string; normalizedResultPath?: string; logFilePath?: string; latestFromLogs?: boolean; maxLines?: number; maxChars?: number },
 ): Promise<{ text: string; details: UnityToolDetails }> {
-  const useLatest = params.latestFromLogs !== false;
+  // An exact artifact request must not silently recruit unrelated latest evidence.
+  const useLatest = params.latestFromLogs !== false && ![params.testResultsPath, params.logFilePath, params.normalizedResultPath].some(value => value?.trim());
   const logsRoot = join(candidate.projectRoot, "Logs");
   const testResultsPath = resolveArtifactPath(ctx.cwd, candidate.projectRoot, params.testResultsPath)
     ?? (useLatest ? await findNewestFile(logsRoot, [".xml"]) : undefined);
@@ -713,12 +732,16 @@ async function buildArtifactInspectionReport(
     ?? (useLatest ? await findNewestFile(logsRoot, [".log", ".txt"]) : undefined);
   const normalizedResultPath = resolveArtifactPath(ctx.cwd, candidate.projectRoot, params.normalizedResultPath)
     ?? (useLatest ? await findNewestFile(logsRoot, [".json"]) : undefined);
-  let normalizedSummary: string | undefined;
+  let normalized: NormalizedUnityTestResult | undefined;
+  const evidenceErrors: string[] = [];
+  const evidenceWarnings: string[] = [];
   if (normalizedResultPath) {
     try {
-      const normalized = JSON.parse(await readFile(normalizedResultPath, "utf8")) as Partial<NormalizedUnityTestResult>;
-      if (normalized.schemaVersion === 1 && typeof normalized.outcome === "string") normalizedSummary = `Normalized test result: ${normalized.platform ?? "Unity"} ${normalized.outcome}; ${normalized.summary?.total ?? "unknown"} total.`;
-    } catch { normalizedSummary = `Normalized test result JSON could not be parsed: ${normalizedResultPath}`; }
+      if ((await stat(normalizedResultPath)).size > UNITY_TEST_MAX_ARTIFACT_BYTES) throw new Error("Normalized artifact exceeds its size limit.");
+      normalized = validateNormalizedUnityTestArtifact(JSON.parse(await readFile(normalizedResultPath, "utf8")));
+    } catch (error) {
+      evidenceErrors.push(`Normalized test result could not be loaded/validated: ${normalizedResultPath}: ${error instanceof Error ? error.message : String(error)}`);
+    }
   }
   const invocation: UnityBatchmodeInvocation = {
     isTestRun: Boolean(testResultsPath),
@@ -731,16 +754,51 @@ async function buildArtifactInspectionReport(
   if (testResultsPath && artifacts.testResultsXml && !parsedTestResults) {
     artifacts.warnings.push(`Unity test results XML could not be parsed: ${artifacts.testResultsPath ?? testResultsPath}`);
   }
-  const hasLoadedArtifacts = Boolean(artifacts.testResultsPath || artifacts.logFilePath);
-  const status = deriveUnityArtifactInspectionStatus(hasLoadedArtifacts, invocation, parsedTestResults);
+  if (testResultsPath && !parsedTestResults) evidenceErrors.push(`Requested XML evidence is missing or malformed: ${testResultsPath}`);
+  if (parsedTestResults && hasConflictingUnityXmlTestEvidence(parsedTestResults)) {
+    evidenceErrors.push("Conflicting XML evidence: invalid or inconsistent counts/records.");
+  }
+  if (logFilePath && artifacts.logText === undefined) evidenceErrors.push(`Requested log evidence is missing: ${logFilePath}`);
+  const hasLoadedArtifacts = Boolean(normalized || parsedTestResults || artifacts.logText !== undefined);
+  if (!hasLoadedArtifacts) evidenceErrors.push("No valid Unity artifacts were loaded.");
+  let testOutcome = normalized?.outcome ?? (parsedTestResults ? determineUnityTestOutcome({ ...parsedTestResults, failed: parsedTestResults.failedTests.length > 0 ? Math.max(1, parsedTestResults.failed ?? 0) : parsedTestResults.failed }) : undefined);
+  if (normalized && parsedTestResults) {
+    if (hasConflictingUnityXmlTestEvidence(parsedTestResults, normalized.summary)) evidenceErrors.push("Conflicting normalized/XML evidence: combined counts/records disagree.");
+    for (const key of ["total", "passed", "failed", "skipped", "inconclusive"] as const) {
+      if (normalized.summary[key] !== undefined && parsedTestResults[key] !== undefined && normalized.summary[key] !== parsedTestResults[key]) evidenceErrors.push(`Conflicting normalized/XML evidence: ${key} differs.`);
+    }
+    if ((normalized.outcome === "passed" || normalized.outcome === "passed_with_flakes") && parsedTestResults.failedTests.length > 0) evidenceErrors.push("Conflicting normalized/XML evidence: XML contains failed tests.");
+    const linkedXml = normalized.backendArtifacts?.nunit;
+    if (linkedXml && resolve(candidate.projectRoot, linkedXml) !== resolve(testResultsPath!)) evidenceErrors.push("Conflicting artifact identity: selected XML is not the normalized artifact's nunit path.");
+    if (!linkedXml) {
+      evidenceWarnings.push("Normalized JSON and XML have no shared run identity; matching counts alone do not correlate these files.");
+      testOutcome = "uncertain";
+    }
+  }
+  if (parsedTestResults?.testRecordCounts?.other && (testOutcome === "passed" || testOutcome === "passed_with_flakes")) {
+    evidenceWarnings.push("XML contains test records with unknown outcomes; these are not passing evidence.");
+    testOutcome = "uncertain";
+  }
+  if (useLatest) evidenceWarnings.push("Latest artifact selection does not establish current-run identity; use exact paths for a particular run.");
+  if (evidenceErrors.length) testOutcome = "uncertain";
+  // Inspection succeeded even when valid evidence reports test failure or uncertainty.
+  const status = evidenceErrors.length ? "failed" : "passed";
   const lines = [
     `Unity artifacts inspected for ${formatPathForUser(ctx.cwd, candidate.projectRoot)}; Unity CLI selects the project's declared Editor version when launching.`,
     testResultsPath ? `Requested test results: ${testResultsPath}` : "Requested test results: (none found)",
     logFilePath ? `Requested log file: ${logFilePath}` : "Requested log file: (none found)",
     normalizedResultPath ? `Requested normalized result: ${normalizedResultPath}` : "Requested normalized result: (none found)",
-    ...(normalizedSummary ? [normalizedSummary] : []),
+    `Inspection: ${status}. Test outcome: ${testOutcome ?? "not established (log only)"}.`,
+    ...(normalized ? [compactUnityTestSummary(normalized), `Normalized source: ${normalized.source}; selection (bounded): ${summarizeTextForAgent(JSON.stringify(normalized.selection), 1, 2000)}; ${normalized.tests.length} retained test record(s).`] : []),
+    ...evidenceErrors,
+    ...evidenceWarnings,
   ];
 
+  if (normalized) {
+    for (const test of normalized.tests.filter(test => !["passed", "success"].includes(test.status.toLowerCase())).slice(0, 8)) {
+      lines.push(`- ${test.name.slice(0, 1000)}: ${test.status.slice(0, 100)}${test.message ? ` — ${test.message.slice(0, 1000)}` : ""}`);
+    }
+  }
   if (parsedTestResults) {
     lines.push(...formatParsedTestResultsForAgent(parsedTestResults));
   }
@@ -767,6 +825,10 @@ async function buildArtifactInspectionReport(
       artifacts: compactUnityArtifacts(artifacts),
       parsedTestResults,
       status,
+      testOutcome,
+      normalizedResultPath,
+      normalizedResult: normalized ? { source: normalized.source, platform: normalized.platform, outcome: normalized.outcome, summary: { total: normalized.summary.total, passed: normalized.summary.passed, failed: normalized.summary.failed, skipped: normalized.summary.skipped, inconclusive: normalized.summary.inconclusive }, testRecordCount: normalized.tests.length } : undefined,
+      evidenceWarnings,
     },
   };
 }
@@ -1152,14 +1214,14 @@ async function runUnifiedUnityTests(
   const busy = (await listBlockingUnityProcesses(candidate.projectRoot)).processes.length > 0 || capabilities.matchingInstances.length > 0;
   let route: "connected" | "isolated";
   if (request.execution === "connected") {
-    if (requirements.requiresIsolation) throw new Error(`Connected execution cannot honor this request: ${requirements.reasons.join("; ")}.`);
+    if (requirements.requiresIsolation) throw new Error(`Connected execution cannot honor this request: ${requirements.reasons.join("; ")}. No tests were dispatched. ${CONNECTED_TEST_SELECTOR_GUIDANCE} Other isolated-only options still require a deliberate isolated-execution decision.`);
     if (!reachable) throw new Error("Connected execution requires an already-open exact-copy reachable Pipeline Editor; no Unity was launched.");
     route = "connected";
   } else if (request.execution === "isolated") {
     if (reachable && !request.closeBlockingUnityProcess) throw new Error("Isolated execution will not close a reachable Pipeline Editor automatically. Close it first or use the explicitly guarded close option.");
     route = "isolated";
   } else if (reachable) {
-    if (requirements.requiresIsolation) throw new Error(`This request requires isolated execution (${requirements.reasons.join("; ")}), but the exact project copy is open in reachable Pipeline. pi-unity will not close it automatically.`);
+    if (requirements.requiresIsolation) throw new Error(`This request requires isolated execution (${requirements.reasons.join("; ")}), but the exact project copy is open in reachable Pipeline. pi-unity will not close it automatically. No tests were dispatched. ${CONNECTED_TEST_SELECTOR_GUIDANCE} Other isolated-only options still require a deliberate isolated-execution decision.`);
     route = "connected";
   } else {
     if (capabilities.pipelineDiscovery === "timeout" || (capabilities.projectSupportsPipeline && capabilities.pipelineDiscovery !== "absent" && capabilities.pipelineDiscovery !== "available")) throw new Error("Pipeline discovery is uncertain for this exact project copy; refusing to start an isolated Editor until project state is known.");
@@ -1242,9 +1304,9 @@ function renderUnityPipelineResult(result: any, options: { expanded: boolean; is
     const counts = pipeline.counts;
     const passed = counts?.passed === undefined || counts?.total === undefined ? "tests completed" : `${counts.passed}/${counts.total} passed`;
     text = `${icon} ${theme.fg("toolTitle", theme.bold(`Unity ${pipeline.testPlatform ?? ""} tests`.trim()))} ${theme.fg("accent", passed)}${theme.fg("muted", ` • ${pipeline.elapsedSeconds.toFixed(1)}s`)}`;
-  } else if (details.mode === "pipeline_eval" || details.mode === "pipeline_inspection") {
-    const output = details.mode === "pipeline_eval" ? details.pipelineEval : details.pipelineInspection;
-    const label = details.mode === "pipeline_eval" ? "Unity Pipeline Eval" : "Unity Pipeline Inspection";
+  } else if (details.mode === "pipeline_eval" || details.mode === "pipeline_inspection" || details.mode === "pipeline_run_script") {
+    const output = details.mode === "pipeline_eval" ? details.pipelineEval : details.mode === "pipeline_run_script" ? details.pipelineRunScript : details.pipelineInspection;
+    const label = details.mode === "pipeline_eval" ? "Unity Pipeline Eval" : details.mode === "pipeline_run_script" ? "Unity Pipeline Run Script" : "Unity Pipeline Inspection";
     const summary = output?.outcome === "dispatched" ? output.output || "(no bounded output returned)" : output?.message || primaryText;
     text = `${icon} ${theme.fg("toolTitle", theme.bold(label))}\n  ${theme.fg("toolOutput", compactUnityRendererValue(summary, 240))}`;
   } else {
@@ -1285,7 +1347,9 @@ function renderUnityToolResult(result: any, expanded: boolean, theme: any): Text
           ? "Unity Pipeline Inspection"
           : details.mode === "pipeline_eval"
             ? "Unity Pipeline Eval"
-            : details.mode === "pipeline"
+            : details.mode === "pipeline_run_script"
+              ? "Unity Pipeline Run Script"
+              : details.mode === "pipeline"
               ? "Unity Pipeline"
               : getBatchmodeVariantLabel(details.args);
   const projectLabel = details.projectRoot ?? "(unknown project)";
@@ -1333,6 +1397,18 @@ function formatUnityGuidanceAudit(result: UnityGuidanceAuditResult): string {
 }
 
 export default function freeUnityPi(pi: ExtensionAPI) {
+  // Pi's documented tool_result patch marks native failure without discarding the
+  // structured rejection details (throwing from execute retains only error text).
+  // A top-level isError property returned from execute is NOT a native error.
+  pi.on("tool_result", (event) => {
+    const details = event.details as UnityToolDetails | undefined;
+    if ((event.toolName === "unity_pipeline_eval" && details?.mode === "pipeline_eval" && details.pipelineEval?.outcome === "rejected")
+      || (event.toolName === "unity_pipeline_inspect" && details?.mode === "pipeline_inspection" && details.pipelineInspection?.outcome === "rejected")
+      || (event.toolName === "unity_pipeline_run_script" && details?.mode === "pipeline_run_script" && details.pipelineRunScript?.outcome === "rejected")) {
+      return { isError: true };
+    }
+  });
+
   type ScopeRegistrations = Readonly<{
     artifactProfile?: Readonly<{ registry: ScopedRegistryV1; token: RegistrationToken }>;
     fileDiscoveryFilter?: Readonly<{ registry: ScopedRegistryV1; token: RegistrationToken }>;
@@ -1546,6 +1622,7 @@ export default function freeUnityPi(pi: ExtensionAPI) {
     promptSnippet: "Run Unity EditMode or PlayMode tests through one safe routed workflow with durable normalized evidence.",
     promptGuidelines: [
       "Use unity_run_tests for ordinary Unity Test Framework runs. It selects connected Pipeline only for compatible requests and isolated unity test only when the exact project copy is closed.",
+      CONNECTED_TEST_SELECTOR_GUIDANCE,
       "Do not use unity_launch_batchmode for ordinary tests; raw test flags there are an unsupported escape hatch.",
       "A reachable Editor is never closed merely to obtain isolated-only options. Requests needing retries, sharding, reruns, coverage, multiple selectors, or XML reports are rejected before dispatch when it is open.",
       "Timeout, malformed evidence, cancellation, or missing artifacts never cause a backend fallback or relaunch.",
@@ -1637,6 +1714,38 @@ export default function freeUnityPi(pi: ExtensionAPI) {
     },
   });
 
+  pi.registerTool({
+    name: "unity_pipeline_run_script",
+    label: "Unity Pipeline Run Script",
+    description: "Compile one existing project C# file in memory and invoke its advertised static entry point through Pipeline 0.6 run_script.",
+    promptSnippet: "Run one explicitly requested existing C# builder script through an already-open exact Unity Pipeline Editor.",
+    promptGuidelines: [
+      "unity_pipeline_run_script is arbitrary code execution, not read-only and not a sandbox. Use it only for explicit user intent; obtain explicit authorization for lifecycle, settings, asset, build, test, or destructive mutations.",
+      "It uses Pipeline's ephemeral mode only: no hotpatch, no source upload, no asset import, no domain reload, no retry, fallback, cancellation, launch, save, or Play Mode exit.",
+      "Use dryRun=true to compile without loading or executing the script. A failure, malformed result, or timeout never establishes success; a timeout may still have running script effects.",
+    ],
+    parameters: PIPELINE_RUN_SCRIPT_PARAMS,
+    async execute(_toolCallId, params, signal, _onUpdate, ctx) {
+      throwIfAborted(signal);
+      const { candidate } = await resolveProjectCandidate(ctx, params.path);
+      const file = isAbsolute(params.file) ? params.file : resolve(candidate.projectRoot, params.file);
+      const result = await dispatchUnityPipelineRunScript({
+        projectRoot: candidate.projectRoot,
+        unityVersion: await requireManualUnityVersion(candidate),
+        file,
+        entry: params.entry,
+        args: params.args,
+        dryRun: params.dryRun,
+      }, { execute: createPlanningUnityCliExecutor(pi), signal, timeout: (params.timeoutSeconds ?? 30) * 1000 });
+      const text = result.outcome === "dispatched"
+        ? `Unity Pipeline run_script ${params.dryRun ? "compile-only completed" : "completed"}.\n${result.output || "(no bounded output returned)"}`
+        : `Unity Pipeline run_script rejected: ${result.code}\n${result.message}`;
+      return { content: [{ type: "text", text }], details: { mode: "pipeline_run_script", projectRoot: candidate.projectRoot, unityVersion: await requireManualUnityVersion(candidate), editorPath: "", status: result.outcome === "dispatched" ? "passed" : "failed", pipelineRunScript: result } satisfies UnityToolDetails };
+    },
+    renderCall(args, theme, context) { return renderUnityPipelineCall("unity_pipeline_run_script", args, theme, context); },
+    renderResult(result, options, theme, context) { return renderUnityPipelineResult(result, options, theme, context); },
+  });
+
   pi.registerTool({
     name: "unity_pipeline_inspect",
     label: "Unity Pipeline Inspect",
@@ -1689,13 +1798,13 @@ export default function freeUnityPi(pi: ExtensionAPI) {
   pi.registerTool({
     name: "unity_inspect_artifacts",
     label: "Unity Inspect Artifacts",
-    description: "Summarize existing Unity log files and Unity Test Framework XML results without launching Unity.",
-    promptSnippet: "Inspect existing Unity logs or test result XML files without launching Unity.",
+    description: "Inspect existing Unity normalized JSON, Test Framework XML and logs without launching Unity. Inspection success is separate from test outcome.",
+    promptSnippet: "Inspect existing Unity normalized test evidence, XML or logs without launching Unity.",
     promptGuidelines: [
       "Use unity_inspect_artifacts after Unity failures when existing -testResults or -logFile artifacts need concise parsing without another Unity launch.",
       "Prefer unity_inspect_artifacts over ad hoc bash parsing of Unity XML/log files when paths are known or Logs/ contains recent artifacts.",
       "unity_inspect_artifacts does not launch Unity and is safe to use even when the Unity project is busy.",
-      "Treat selected test XML as passing evidence only when it is well formed, reports a known positive executed-test count, and reports no failures.",
+      "Inspect details.testOutcome, not inspection status, for test success. Passing evidence needs consistent positive passing counts; missing explicit paths and conflicting artifacts fail inspection. Latest files are not current-run identity.",
     ],
     parameters: INSPECT_ARTIFACTS_PARAMS,
     async execute(_toolCallId, params, signal, _onUpdate, ctx) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aefree/pi-unity",
-  "version": "0.12.0",
+  "version": "0.13.0",
   "type": "module",
   "exports": {
     ".": "./index.ts"
@@ -25,11 +25,11 @@
     "typebox": "1.3.8"
   },
   "peerDependencies": {
+    "@aefree/pi-file-discovery": "^0.1.0",
+    "@aefree/pi-project-artifacts": "^0.1.0",
     "@earendil-works/pi-ai": "*",
     "@earendil-works/pi-coding-agent": "*",
-    "@earendil-works/pi-tui": "*",
-    "@aefree/pi-project-artifacts": "^0.1.0",
-    "@aefree/pi-file-discovery": "^0.1.0"
+    "@earendil-works/pi-tui": "*"
   },
   "engines": {
     "node": ">=22.19.0"
@@ -48,11 +48,11 @@
   },
   "homepage": "https://github.com/aefreedman/pi-unity#readme",
   "devDependencies": {
-    "@earendil-works/pi-ai": "0.83.0",
-    "@earendil-works/pi-coding-agent": "0.83.0",
-    "@earendil-works/pi-tui": "0.83.0",
-    "@aefree/pi-project-artifacts": "^0.1.0",
     "@aefree/pi-file-discovery": "^0.1.0",
+    "@aefree/pi-project-artifacts": "^0.1.0",
+    "@earendil-works/pi-ai": "0.85.1",
+    "@earendil-works/pi-coding-agent": "0.85.1",
+    "@earendil-works/pi-tui": "0.85.1",
     "tsx": "^4.23.5"
   },
   "peerDependenciesMeta": {

package/skills/unity-pipeline-workflows/SKILL.md

@@ -16,7 +16,7 @@ Use one typed tool call for each supported connected operation:
 
 These tools resolve the exact copy, require advertised commands, inspect lifecycle state, dispatch once, validate identity, and poll internally with a fixed deadline. Do not recreate their wait loops with `bash`, `unity recompile_status`, or `unity test_status` calls.
 
-A timeout or malformed response is uncertain: the Unity operation may still be running. Do not cancel, retry, launch batchmode, close the Editor, or claim a result without a new user-authorized decision. The only automatic retry is Pipeline 0.5's explicit initial-settling `Server Busy` response for `unity_pipeline_recompile` or `unity_run_tests`, which confirms that a main-thread command was rejected before dispatch.
+A timeout, malformed response, or busy response is uncertain: the Unity operation may still be running or blocked by a modal dialog. Do not cancel, retry, launch batchmode, close the Editor, or claim a result without a new user-authorized decision. Pipeline 0.6 uses busy for modal blocks and no source-verified discriminator is available through this transport, so pi-unity does not blindly retry it.
 
 ## Preconditions and boundaries
 
@@ -41,12 +41,30 @@ Call `unity_run_tests` with:
 
 The tool treats `no_tests`, idle, and not-started statuses as safe inactivity, detects a pre-existing active connected test before dispatch, and stops rather than claiming or replacing active work. It captures returned mode/filter/run identity fields when available and stops as uncertain if status is clearly displaced by another run.
 
+### Two independent fixtures in an open Editor
+
+Keep the Editor open. For an authorized request for two known non-overlapping fixtures, use the same explicit exact-copy `path` and platform, one selector per call. For example, replace `./SyntheticGame` with the selected project path and issue this `unity_run_tests` call:
+
+```json
+{"path":"./SyntheticGame","testPlatform":"EditMode","execution":"connected","testFilters":["Synthetic.InventoryFixture"]}
+```
+
+Await the completed tool result and inspect its outcome and exact normalized artifact path. Only after a passing result, issue the second call (not in parallel or pre-queued in the same turn):
+
+```json
+{"path":"./SyntheticGame","testPlatform":"EditMode","execution":"connected","testFilters":["Synthetic.DialogueFixture"]}
+```
+
+Stop the remaining sequence on failure, timeout, cancellation, malformed evidence, displaced identity, or any uncertainty; report the first result and the remaining fixture as not run. Do not retry, close the Editor, switch to isolated execution, or broaden to a parent suite/category or empty selectors. Empty/omitted selectors select all tests. Keep each result's artifact separately; do not treat one fixture's evidence as covering the other.
+
+For a single category use `testCategories: ["SyntheticCategory"]` and omit `testFilters`. Never combine the two families or join selectors with semicolons. Multiple-selector rejection means no test was dispatched: use the serial recipe only for independent, non-overlapping selections without other isolated-only requirements. If overlap or intended filter/category intersection semantics are unclear, stop for clarification rather than split, duplicate, or broaden tests. Required XML, retries, coverage, or other isolated-only options still need a deliberate isolation decision, not argument stripping.
+
 ## Bounded raw CLI troubleshooting only
 
 Normally call the typed tools, not raw CLI commands. If a typed tool is unavailable in an older installed package and a user specifically authorizes troubleshooting, first use `unity_project_status` and require advertised `recompile_status` or `run_tests` and `test_status`. Use the documented asynchronous form with `--async_tests true`, one fixed deadline, and bounded backoff; parse object and stringified nested JSON. `Total: 0` with `result: running` is a valid nonterminal initiating response. Passing evidence reports successful completion, a known positive executed-test count, and zero failures; nested `success:false`, changed exact-copy identity, or polling timeout is non-passing uncertainty. Do not silently fall back to batchmode after uncertain connected dispatch. Connected work does not guarantee NUnit XML.
 
 ## When not to use connected tools
 
-Use `unity_run_tests` with `execution: "isolated"` for a closed project, intentional isolation/CI, multiple selectors, retries, sharding, coverage, or required NUnit/JUnit evidence. A single category is supported connected when Pipeline advertises the command; do not claim broader connected selector support. Do not use batchmode as an automatic fallback after an uncertain connected dispatch.
+Use `unity_run_tests` with `execution: "isolated"` for a closed project, intentional isolation/CI, multiple selectors in one run, retries, sharding, coverage, or required NUnit/JUnit evidence. Multiple independent fixtures alone do not require closing a reachable Editor: use the serial recipe above. A single category is supported connected when Pipeline advertises the command; do not claim broader connected selector support. Do not use batchmode as an automatic fallback after an uncertain connected dispatch.
 
-Use the typed compile/test tools when their polling and terminal evidence fit the task. Advertised Pipeline `eval` remains available through `unity_pipeline_eval` for bounded project-specific inspection or operations outside those typed workflows; its `timeoutSeconds` range is 1–86,400 seconds, and a timeout remains uncertain without cancellation or retry. It is an assistance surface, not a forbidden fallback or a substitute for the typed tools' completion protocol. Eval compiles arbitrary C# with Roslyn on the Editor main thread, so ordinary properties and local-variable snippets are valid; it is not expression-only or reliably statically read-only. Prefer typed tools for their stronger evidence, but let user intent and project guidance govern mutations. Lifecycle, persistent-setting, destructive, asset, scene-save, package, build, and test mutations require explicit authorization.
+Use the typed compile/test tools when their polling and terminal evidence fit the task. For an explicitly requested existing C# builder file, `unity_pipeline_run_script` uses Pipeline 0.6 ephemeral in-memory compilation with bounded JSON arguments or compile-only dry run; it never enables hotpatch and remains arbitrary code execution requiring explicit mutation authorization. Advertised Pipeline `eval` remains available through `unity_pipeline_eval` for bounded project-specific inspection or operations outside those typed workflows; its `timeoutSeconds` range is 1–86,400 seconds, and a timeout remains uncertain without cancellation or retry. It is an assistance surface, not a forbidden fallback or a substitute for the typed tools' completion protocol. Eval compiles arbitrary C# with Roslyn on the Editor main thread, so ordinary properties and local-variable snippets are valid; it is not expression-only or reliably statically read-only. Prefer typed tools for their stronger evidence, but let user intent and project guidance govern mutations. Lifecycle, persistent-setting, destructive, asset, scene-save, package, build, and test mutations require explicit authorization.

package/src/unity-artifact-inspection.ts

@@ -0,0 +1,59 @@
+import { isAbsolute } from "node:path";
+import { determineUnityTestOutcome, type NormalizedUnityTestResult } from "./unity-tests";
+
+const outcomes = ["passed", "passed_with_flakes", "tests_failed", "empty_selection", "run_error", "timed_out", "cancelled", "uncertain"];
+const record = (value: unknown): value is Record<string, unknown> => !!value && typeof value === "object" && !Array.isArray(value);
+const nonnegative = (value: unknown): value is number => typeof value === "number" && Number.isFinite(value) && value >= 0;
+const count = (value: unknown): value is number => nonnegative(value) && Number.isSafeInteger(value);
+const strings = (value: unknown): value is string[] => Array.isArray(value) && value.every(item => typeof item === "string" && !!item.trim() && !/[\0\r\n;]/.test(item));
+const relativeId = (value: unknown): value is string => typeof value === "string" && !!value.trim()
+  && !isAbsolute(value) && !/^(?:[A-Za-z]:|[\\/])/.test(value) && !value.split(/[\\/]/).includes("..") && !/\0/.test(value);
+
+/** Read the durable schema, not a transport response. Missing optional counts remain unknown.
+ * Test records may be bounded or absent: never require tests.length === summary.total.
+ */
+export function validateNormalizedUnityTestArtifact(value: unknown): NormalizedUnityTestResult {
+  const invalid = (reason: string): never => { throw new Error(`Invalid normalized Unity test artifact: ${reason}.`); };
+  if (!record(value) || value.schemaVersion !== 1) invalid("expected schemaVersion 1 object");
+  const result = value as Record<string, unknown>;
+  if (!["pipeline", "unity-cli", "editor-executable"].includes(result.source as string)) invalid("unsupported source");
+  if (!["EditMode", "PlayMode"].includes(result.platform as string)) invalid("unsupported platform");
+  if (!outcomes.includes(result.outcome as string)) invalid("unsupported outcome");
+  if (!record(result.selection) || !strings(result.selection.testFilters) || !strings(result.selection.testCategories)) invalid("selection must contain testFilters and testCategories arrays");
+  if (!record(result.summary)) invalid("summary must be an object");
+  const summary = result.summary as Record<string, unknown>;
+  for (const key of ["total", "passed", "failed", "skipped", "inconclusive"]) {
+    if (summary[key] !== undefined && !count(summary[key])) invalid(`summary.${key} must be a non-negative integer`);
+  }
+  if (count(summary.total)) {
+    const accounted = [summary.passed, summary.failed, summary.skipped, summary.inconclusive].reduce<number>((sum, item) => sum + (count(item) ? item : 0), 0);
+    if (accounted > summary.total) invalid("summary counts exceed total");
+  }
+  if (!Array.isArray(result.tests)) invalid("tests must be an array");
+  const tests = result.tests as unknown[];
+  for (const test of tests) {
+    if (!record(test) || typeof test.name !== "string" || !test.name.trim() || typeof test.status !== "string" || !test.status.trim()) invalid("test records require name and status");
+    const item = test as Record<string, unknown>;
+    for (const key of ["message", "stackTrace"]) if (item[key] !== undefined && typeof item[key] !== "string") invalid(`test ${key} must be a string`);
+    if (item.durationSeconds !== undefined && !nonnegative(item.durationSeconds)) invalid("test durationSeconds must be non-negative");
+    if (item.attempts !== undefined && (!count(item.attempts) || item.attempts < 1)) invalid("test attempts must be positive");
+  }
+  if (count(summary.total) && tests.length > summary.total) invalid("test records exceed total");
+  const typed = result as unknown as NormalizedUnityTestResult;
+  for (const [status, key] of [["passed", "passed"], ["failed", "failed"], ["skipped", "skipped"], ["inconclusive", "inconclusive"]] as const) {
+    const observed = typed.tests.filter(test => test.status.toLowerCase() === status).length;
+    if (count(summary[key]) && observed > summary[key]) invalid(`test records conflict with summary.${key}`);
+  }
+  if (result.projectRelativeId !== undefined && !relativeId(result.projectRelativeId)) invalid("projectRelativeId must be project-relative");
+  if (result.backendArtifacts !== undefined && (!record(result.backendArtifacts) || !Object.values(result.backendArtifacts).every(relativeId))) invalid("backendArtifacts must contain project-relative paths");
+  for (const key of ["startedAt", "completedAt"]) if (result[key] !== undefined && (typeof result[key] !== "string" || !Number.isFinite(Date.parse(result[key] as string)))) invalid(`${key} must be a timestamp`);
+  if (typed.startedAt && typed.completedAt && Date.parse(typed.completedAt) < Date.parse(typed.startedAt)) invalid("completion precedes start");
+  if (result.durationSeconds !== undefined && !nonnegative(result.durationSeconds)) invalid("durationSeconds must be non-negative");
+  if (result.flakyTests !== undefined && (!Array.isArray(result.flakyTests) || !result.flakyTests.every(item => record(item) && typeof item.name === "string" && !!item.name.trim() && count(item.attempts) && item.attempts > 0))) invalid("invalid flakyTests");
+  if (result.outcome === "passed" || result.outcome === "passed_with_flakes") {
+    if (determineUnityTestOutcome(typed.summary) !== "passed" || typed.tests.some(test => !["passed", "success"].includes(test.status.toLowerCase()))) invalid("passing outcome lacks consistent positive passing counts/records");
+    if (result.outcome === "passed_with_flakes" && !typed.flakyTests?.length) invalid("passed_with_flakes requires flakyTests evidence");
+  }
+  if (result.outcome === "empty_selection" && ((typed.summary.total ?? 0) > 0 || tests.length > 0)) invalid("empty selection conflicts with executed tests");
+  return typed;
+}

package/src/unity-batchmode.ts

@@ -29,6 +29,8 @@ export type UnityParsedTestResults = {
   failedTests: UnityFailedTest[];
   /** Complete bounded per-test evidence for normalized artifacts, never routine tool output. */
   tests: UnityParsedTestCase[];
+  /** Observed XML record lower bounds, counted before record output is truncated. */
+  testRecordCounts?: { total: number; passed: number; failed: number; skipped: number; inconclusive: number; other: number };
 };
 
 export type UnityBatchmodeArtifacts = {
@@ -82,10 +84,10 @@ function decodeXmlText(value: string | undefined): string | undefined {
 
 function parseAttributes(tagSource: string): Record<string, string> {
   const attributes: Record<string, string> = {};
-  const attributeRegex = /(\w[\w:-]*)\s*=\s*"([^"]*)"/g;
+  const attributeRegex = /(\w[\w:-]*)\s*=\s*(?:"([^"]*)"|'([^']*)')/g;
   for (const match of tagSource.matchAll(attributeRegex)) {
     const key = match[1];
-    const value = match[2] ?? "";
+    const value = match[2] ?? match[3] ?? "";
     attributes[key] = value;
   }
   return attributes;
@@ -102,6 +104,12 @@ function parseOptionalNumber(value: string | undefined): number | undefined {
   return Number.isFinite(parsed) ? parsed : undefined;
 }
 
+function parseTestCount(value: string | undefined): number | undefined {
+  // Only omission is unknown. Retain invalid supplied counters as NaN so the
+  // inspection validator rejects them instead of silently treating them as absent.
+  return value === undefined ? undefined : value.trim() ? Number(value) : Number.NaN;
+}
+
 export function parseUnityTestResultsXml(xml: string): UnityParsedTestResults | null {
   const testRunMatch = xml.match(/<test-run\b([^>]*)>/i);
   const testRunCloseIndex = xml.search(/<\/test-run\s*>/i);
@@ -112,33 +120,43 @@ export function parseUnityTestResultsXml(xml: string): UnityParsedTestResults |
   const rootAttributes = parseAttributes(testRunMatch[1] ?? "");
   const failedTests: UnityFailedTest[] = [];
   const tests: UnityParsedTestCase[] = [];
+  const testRecordCounts = { total: 0, passed: 0, failed: 0, skipped: 0, inconclusive: 0, other: 0 };
 
-  const testCaseRegex = /<test-case\b([^>]*)>([\s\S]*?)<\/test-case>/gi;
+  // Match the self-closing alternative first so it cannot consume the body of
+  // the next paired record. Both forms carry authoritative failure evidence.
+  const testCaseRegex = /<test-case\b([^>]*?)(?:\/\s*>|>([\s\S]*?)<\/test-case\s*>)/gi;
   for (const match of xml.matchAll(testCaseRegex)) {
     const attributes = parseAttributes(match[1] ?? "");
     const body = match[2] ?? "";
     const result = String(attributes.result ?? attributes.label ?? "").toLowerCase();
     const success = String(attributes.success ?? "").toLowerCase();
     const isFailure = result === "failed" || success === "false";
+    const status = isFailure ? "Failed" : attributes.result ?? attributes.label ?? "Unknown";
+    const statusKey = isFailure ? "failed" : result === "passed" || result === "success" ? "passed" : result === "skipped" ? "skipped" : result === "inconclusive" ? "inconclusive" : "other";
+    testRecordCounts.total++;
+    testRecordCounts[statusKey]++;
     const failureMessage = body.match(/<message[^>]*>([\s\S]*?)<\/message>/i);
     const stackTrace = body.match(/<stack-trace[^>]*>([\s\S]*?)<\/stack-trace>/i);
     const name = truncateEvidence(attributes.fullname ?? attributes.name ?? "(unknown test)", 1_000) ?? "(unknown test)";
-    if (tests.length < 2_000) tests.push({ name, status: attributes.result ?? attributes.label ?? "Unknown", ...(parseOptionalNumber(attributes.duration) === undefined ? {} : { durationSeconds: parseOptionalNumber(attributes.duration) }), ...(truncateEvidence(decodeXmlText(failureMessage?.[1]), 4_000) ? { message: truncateEvidence(decodeXmlText(failureMessage?.[1]), 4_000) } : {}), ...(truncateEvidence(decodeXmlText(stackTrace?.[1]), 8_000) ? { stackTrace: truncateEvidence(decodeXmlText(stackTrace?.[1]), 8_000) } : {}) });
+    if (tests.length < 2_000) tests.push({ name, status, ...(parseOptionalNumber(attributes.duration) === undefined ? {} : { durationSeconds: parseOptionalNumber(attributes.duration) }), ...(truncateEvidence(decodeXmlText(failureMessage?.[1]), 4_000) ? { message: truncateEvidence(decodeXmlText(failureMessage?.[1]), 4_000) } : {}), ...(truncateEvidence(decodeXmlText(stackTrace?.[1]), 8_000) ? { stackTrace: truncateEvidence(decodeXmlText(stackTrace?.[1]), 8_000) } : {}) });
     if (!isFailure) continue;
     if (failedTests.length < 50) failedTests.push({ name, message: truncateEvidence(decodeXmlText(failureMessage?.[1]), 1_000), stackTrace: truncateEvidence(decodeXmlText(stackTrace?.[1]), 4_000) });
   }
 
-  const skipped = parseOptionalNumber(rootAttributes.skipped) ?? parseOptionalNumber(rootAttributes.inconclusive);
+  // These are separate counters. Synthesizing skipped from inconclusive makes
+  // combined-count validation double-count one observed category.
+  const skipped = parseTestCount(rootAttributes.skipped);
 
   const parsed: UnityParsedTestResults = {
-    total: parseOptionalNumber(rootAttributes.total) ?? parseOptionalNumber(rootAttributes.testcasecount),
-    passed: parseOptionalNumber(rootAttributes.passed),
-    failed: parseOptionalNumber(rootAttributes.failed),
+    total: parseTestCount(rootAttributes.total) ?? parseTestCount(rootAttributes.testcasecount),
+    passed: parseTestCount(rootAttributes.passed),
+    failed: parseTestCount(rootAttributes.failed),
     skipped,
-    inconclusive: parseOptionalNumber(rootAttributes.inconclusive),
+    inconclusive: parseTestCount(rootAttributes.inconclusive),
     durationSeconds: parseOptionalNumber(rootAttributes.duration),
     failedTests,
     tests,
+    testRecordCounts,
   };
   if (parsed.total === undefined && parsed.passed === undefined && parsed.failed === undefined && parsed.failedTests.length === 0) {
     return null;
@@ -146,6 +164,23 @@ export function parseUnityTestResultsXml(xml: string): UnityParsedTestResults |
   return parsed;
 }
 
+/** Missing counters and omitted/bounded records are unknown, not zero executions.
+ * Supplied counters and observed record lower bounds must nevertheless agree.
+ * An optional linked summary can fill missing XML counters, not replace them.
+ */
+export function hasConflictingUnityXmlTestEvidence(
+  results: UnityParsedTestResults,
+  linkedSummary?: Pick<UnityParsedTestResults, "total" | "passed" | "failed" | "skipped" | "inconclusive">,
+): boolean {
+  const keys = ["total", "passed", "failed", "skipped", "inconclusive"] as const;
+  const counts = Object.fromEntries(keys.map(key => [key, results[key] ?? linkedSummary?.[key]])) as Pick<UnityParsedTestResults, typeof keys[number]>;
+  if (keys.some(key => counts[key] !== undefined && (!Number.isSafeInteger(counts[key]) || counts[key]! < 0))) return true;
+  const observed = results.testRecordCounts;
+  if (keys.some(key => counts[key] !== undefined && (observed?.[key] ?? 0) > counts[key]!)) return true;
+  const accounted = keys.slice(1).reduce((sum, key) => sum + Math.max(counts[key] ?? 0, observed?.[key] ?? 0), 0);
+  return counts.total !== undefined && accounted > counts.total;
+}
+
 function buildArtifactCandidates(cwd: string, projectRoot: string, rawPath: string): string[] {
   if (path.isAbsolute(rawPath)) {
     return [path.normalize(rawPath)];

package/src/unity-cli.ts

@@ -1,6 +1,6 @@
 import { execFile } from "node:child_process";
-import { readFile, realpath } from "node:fs/promises";
-import { join } from "node:path";
+import { readFile, realpath, stat } from "node:fs/promises";
+import { isAbsolute, join, relative } from "node:path";
 import { applyDefaultUnityBatchmodeArgs, buildUnityBatchmodeArgs, projectPathsMatch } from "./unity-core";
 import type { RunningUnityProcess } from "./unity-processes";
 
@@ -265,10 +265,10 @@ export function parseUnityCliStatusOutput(output: string, projectRoot: string):
 
 export async function listRunningUnityCliEditorsForProject(
   projectRoot: string,
-  options: { cliCommand?: string; timeout?: number } = {},
+  options: { cliCommand?: string; timeout?: number; execute?: UnityCliExecutor } = {},
 ): Promise<{ processes: RunningUnityProcess[]; warning?: string }> {
   const command = resolveUnityCliCommand(options);
-  const result = await execFileCollect(command, ["--format", "json", "--no-banner", "--non-interactive", "status", "--project", projectRoot], {
+  const result = await (options.execute ?? execFileCollect)(command, ["--format", "json", "--no-banner", "--non-interactive", "status", "--project", projectRoot], {
     timeout: options.timeout ?? 5000,
   });
 
@@ -277,11 +277,14 @@ export async function listRunningUnityCliEditorsForProject(
   }
 
   const processes = parseUnityCliStatusOutput(result.stdout, projectRoot);
-  if (processes.length > 0) {
-    return { processes };
+  const payload = parseJsonObject(result.stdout);
+  const statusWarnings = envelopeMessages(payload, "warnings");
+  if (statusWarnings.length > 0) {
+    return { processes, warning: `Unity CLI status response is incomplete; process absence is uncertain: ${statusWarnings.join("; ")}` };
   }
+  if (processes.length > 0) return { processes };
+
 
-  const payload = parseJsonObject(result.stdout);
   const errors = Array.isArray(payload?.errors) ? payload.errors : [];
   const onlyNoInstances = errors.some((entry) => getRecord(entry)?.code === "STATUS_NO_INSTANCES");
   if (result.error && !onlyNoInstances) {
@@ -430,14 +433,27 @@ export function isUnityCliTimeout(result: Pick<UnityCliExecResult, "error">): bo
   return error?.code === "ETIMEDOUT" || error?.killed === true || error?.signal === "SIGTERM";
 }
 
+const UNITY_CLI_MAX_DIAGNOSTICS = 8;
+function envelopeMessages(payload: Record<string, unknown> | null, fieldName: "errors" | "warnings" | "info"): string[] {
+  const entries = Array.isArray(payload?.[fieldName]) ? payload[fieldName] : [];
+  return entries.slice(0, UNITY_CLI_MAX_DIAGNOSTICS).flatMap((entry): string[] => {
+    const item = getRecord(entry);
+    const message = optionalString(item?.message, item?.detail, typeof entry === "string" ? entry : undefined);
+    return message ? [redactUnityPlanningOutput(summarizeUnityCliText(message, 1_000, 10))] : [];
+  });
+}
+
+/** Warnings make discovery incomplete; informational descriptor/envelope notes do not. */
+function cliEnvelopeDiagnostics(payload: Record<string, unknown> | null): string[] {
+  return envelopeMessages(payload, "warnings").map(message => `warning: ${message}`);
+}
+function cliEnvelopeInfo(payload: Record<string, unknown> | null): string[] {
+  return envelopeMessages(payload, "info").map(message => `info: ${message}`);
+}
+
 function cliFailureMessage(result: UnityCliExecResult): string | undefined {
   const payload = parseJsonObject(result.stdout);
-  const errors = Array.isArray(payload?.errors) ? payload.errors : [];
-  const messages = errors.flatMap((entry): string[] => {
-    const message = optionalString(getRecord(entry)?.message);
-    return message ? [message] : [];
-  });
-  const message = messages[0] ?? (result.stderr.trim() || result.error?.message);
+  const message = envelopeMessages(payload, "errors")[0] ?? (result.stderr.trim() || result.error?.message);
   return message ? summarizeUnityCliText(message) : undefined;
 }
 
@@ -467,7 +483,9 @@ export async function inspectUnityCliProjectCapabilities(
   const command = resolveUnityCliCommand(options);
   const versionTimeout = options.timeout ?? UNITY_CLI_VERSION_TIMEOUT_MS;
   const discoveryTimeout = options.timeout ?? UNITY_CLI_DISCOVERY_TIMEOUT_MS;
-  const versionResult = await execute(command, ["--version"], { timeout: versionTimeout, signal: options.signal });
+  // Explicitly select the supported human formatter: inherited UNITY_FORMAT must not turn
+  // --version into JSON (or another representation) and corrupt capability reporting.
+  const versionResult = await execute(command, ["--format", "human", "--version"], { timeout: versionTimeout, signal: options.signal });
   if (versionResult.error && (versionResult.error as NodeJS.ErrnoException).code === "ENOENT") return result;
   if (versionResult.error) {
     result.warnings.push(`Unity CLI version probe ${isUnityCliTimeout(versionResult) ? "timed out" : "failed"}: ${cliFailureMessage(versionResult) ?? "unknown error"}`);
@@ -479,9 +497,15 @@ export async function inspectUnityCliProjectCapabilities(
   const pipelineResult = await execute(command, ["--format", "json", "--no-banner", "--non-interactive", "pipeline", "list"], { timeout: discoveryTimeout, signal: options.signal });
   const pipelinePayload = parseJsonObject(pipelineResult.stdout);
   const pipelineData = getRecord(pipelinePayload?.data);
-  if (pipelineResult.error || pipelinePayload?.success !== true || !Array.isArray(pipelineData?.instances)) {
+  const pipelineDiagnostics = cliEnvelopeDiagnostics(pipelinePayload);
+  result.warnings.push(...cliEnvelopeInfo(pipelinePayload));
+  if (pipelineResult.error || pipelinePayload?.success !== true || !Array.isArray(pipelineData?.instances) || pipelineDiagnostics.length > 0) {
     result.pipelineDiscovery = isUnityCliTimeout(pipelineResult) ? "timeout" : "unavailable";
-    result.warnings.push(`Unity Pipeline instance discovery ${result.pipelineDiscovery === "timeout" ? "timed out; Pipeline startup state is uncertain" : "failed"}: ${cliFailureMessage(pipelineResult) ?? "malformed or unsupported JSON response"}`);
+    const diagnostic = pipelineDiagnostics.join("; ") || cliFailureMessage(pipelineResult) || "malformed or unsupported JSON response";
+    result.warnings.push(`Unity Pipeline instance discovery ${result.pipelineDiscovery === "timeout" ? "timed out; Pipeline startup state is uncertain" : "is incomplete or failed; Pipeline startup state is uncertain"}: ${diagnostic}`);
+    // Retain any known positive exact-copy instance descriptors, but never use this
+    // incomplete response as a safe launch or connected-dispatch signal.
+    result.matchingInstances = parseUnityCliPipelineListOutput(pipelineResult.stdout, projectRoot).instances;
     return result;
   }
   result.pipelineDiscovery = "available";
@@ -500,9 +524,17 @@ export async function inspectUnityCliProjectCapabilities(
   result.commandDiscoveryAttempted = true;
   const listResult = await execute(command, ["--format", "json", "--no-banner", "--non-interactive", "list", "--project-path", projectRoot], { timeout: discoveryTimeout, signal: options.signal });
   const catalog = parseUnityCliCommandCatalog(listResult.stdout);
-  if (listResult.error || !catalog.valid) {
+  const listPayload = parseJsonObject(listResult.stdout);
+  const commandDiagnostics = cliEnvelopeDiagnostics(listPayload);
+  result.warnings.push(...cliEnvelopeInfo(listPayload));
+  if (listResult.error || !catalog.valid || commandDiagnostics.length > 0) {
     result.commandDiscovery = isUnityCliTimeout(listResult) ? "timeout" : "unavailable";
-    result.warnings.push(`Unity Pipeline command discovery for the exact project copy ${result.commandDiscovery === "timeout" ? "timed out; command availability is uncertain" : "failed"}: ${cliFailureMessage(listResult) ?? "malformed or unsupported JSON response"}`);
+    result.warnings.push(`Unity Pipeline command discovery for the exact project copy ${result.commandDiscovery === "timeout" ? "timed out; command availability is uncertain" : "is incomplete or failed; command availability is uncertain"}: ${commandDiagnostics.join("; ") || cliFailureMessage(listResult) || "malformed or unsupported JSON response"}`);
+    // Commands in a warning-bearing catalog are informational only, not advertised
+    // capability evidence. Keep descriptors for status visibility without enabling dispatch.
+    result.advertisedCommands = catalog.commands;
+    result.advertisedCommandCount = catalog.total;
+    result.advertisedCommandsTruncated = catalog.truncated;
     return result;
   }
   result.commandDiscovery = "available";
@@ -522,6 +554,7 @@ export const UNITY_PLANNING_READ_COMMANDS = Object.freeze([
   "get_authoring_root",
   "get_build_settings",
   "get_player_settings",
+  "get_runtime_pipeline_settings",
   "get_scene_hierarchy",
   "editor_status",
   "list_open_scenes",
@@ -542,6 +575,21 @@ export type UnityPlanningInspectionResult =
   | { outcome: "dispatched"; command: string; output: string; truncated: boolean }
   | { outcome: "rejected"; code: string; message: string };
 
+export type UnityPipelineRunScriptRequest = {
+  projectRoot: string;
+  unityVersion: string;
+  file: string;
+  entry?: string;
+  args?: unknown[];
+  dryRun?: boolean;
+};
+
+/** Attach bounded native discovery guidance without changing the readiness decision. */
+export function unityCapabilityDiagnosticSuffix(capabilities: UnityCliProjectCapabilities): string {
+  const message = summarizeUnityCliText(redactUnityPlanningOutput(capabilities.warnings.slice(0, 8).join("; ")), 1_000, 10);
+  return message ? ` ${message}` : "";
+}
+
 function planningInspectionReadiness(capabilities: UnityCliProjectCapabilities): string | undefined {
   if (!capabilities.cliAvailable) return "unity_cli_unavailable";
   if (capabilities.pipelineDiscovery !== "available") return `pipeline_${capabilities.pipelineDiscovery}`;
@@ -556,12 +604,32 @@ function caseInsensitiveField(record: Record<string, unknown>, name: string): un
   return entry?.[1];
 }
 
+function runScriptCommandFailure(output: string): "malformed" | "failure" | undefined {
+  const envelope = parseJsonObject(output);
+  if (!envelope || caseInsensitiveField(envelope, "success") !== true) return envelope ? "failure" : "malformed";
+  const wrapper = getRecord(caseInsensitiveField(envelope, "data")) ?? envelope;
+  if (caseInsensitiveField(wrapper, "success") === false) return "failure";
+  let response: unknown = caseInsensitiveField(wrapper, "result");
+  // Legacy wrapper puts the documented RunScriptResponse in data.result; compact
+  // transport puts it in result. A bare success envelope has no command evidence.
+  if (typeof response === "string") { try { response = JSON.parse(response); } catch { return "malformed"; } }
+  const result = getRecord(response);
+  if (!result || !Object.prototype.hasOwnProperty.call(result, "diagnostics") || !Array.isArray(caseInsensitiveField(result, "diagnostics"))) return "malformed";
+  if (caseInsensitiveField(result, "success") === false || caseInsensitiveField(result, "failed") === true) return "failure";
+  const diagnostics = caseInsensitiveField(result, "diagnostics") as unknown[];
+  if (diagnostics.some(item => {
+    const diagnostic = getRecord(item);
+    return String(caseInsensitiveField(diagnostic ?? {}, "severity") ?? "").toLowerCase() === "error";
+  })) return "failure";
+  return undefined;
+}
+
 function connectedCommandFailure(output: string, isEval: boolean): "malformed" | "failure" | undefined {
   const envelope = parseJsonObject(output);
   if (!envelope) return "malformed";
   if (caseInsensitiveField(envelope, "success") !== true) return "failure";
-  const data = getRecord(caseInsensitiveField(envelope, "data"));
-  if (!data) return "malformed";
+  // Pipeline 0.6 compact responses omit the CLI wrapper's data field.
+  const data = getRecord(caseInsensitiveField(envelope, "data")) ?? envelope;
   if (caseInsensitiveField(data, "success") === false) return "failure";
   if (!isEval) return undefined;
 
@@ -610,7 +678,7 @@ export async function dispatchUnityPlanningInspection(
   }));
   const initial = await inspect(projectRoot, request.unityVersion);
   const initialFailure = planningInspectionReadiness(initial);
-  if (initialFailure) return { outcome: "rejected", code: initialFailure, message: "Exact-copy Pipeline planning inspection is not established." };
+  if (initialFailure) return { outcome: "rejected", code: initialFailure, message: `Exact-copy Pipeline planning inspection is not established.${unityCapabilityDiagnosticSuffix(initial)}` };
 
   const isEval = request.command === "eval";
   const hasBoundedArgs = (request.args?.length ?? 0) <= 12
@@ -633,7 +701,7 @@ export async function dispatchUnityPlanningInspection(
   const refreshed = await inspect(projectRoot, request.unityVersion);
   const refreshedFailure = planningInspectionReadiness(refreshed);
   if (refreshedFailure || !haveSameKnownProcessIds(initial.matchingInstances, refreshed.matchingInstances)) {
-    return { outcome: "rejected", code: "unity_project_identity_changed", message: "Pipeline identity changed or disconnected immediately before planning dispatch." };
+    return { outcome: "rejected", code: "unity_project_identity_changed", message: `Pipeline identity changed or disconnected immediately before planning dispatch.${unityCapabilityDiagnosticSuffix(refreshed)}` };
   }
   if (!refreshed.advertisedCommands.includes(request.command)) {
     return { outcome: "rejected", code: "planning_command_unadvertised", message: "The refreshed exact Pipeline copy did not advertise the requested command." };
@@ -647,11 +715,11 @@ export async function dispatchUnityPlanningInspection(
     ...(isEval ? [request.evalSnippet!.trim()] : request.args ?? []),
   ];
   const execution = await options.execute(command, args, { timeout: options.timeout ?? UNITY_CLI_DISCOVERY_TIMEOUT_MS, signal: options.signal });
+  const raw = [execution.stdout, execution.stderr].filter(Boolean).join("\n");
+  const output = summarizeUnityCliText(redactUnityPlanningOutput(raw), 4_000, 40);
   if (execution.error) {
-    return { outcome: "rejected", code: isUnityCliTimeout(execution) ? "planning_command_timeout" : "planning_command_failed", message: "Connected command did not complete successfully; its effect may be uncertain." };
+    return { outcome: "rejected", code: isUnityCliTimeout(execution) ? "planning_command_timeout" : "planning_command_failed", message: `Connected command did not complete successfully; its effect may be uncertain.${output ? ` ${output}` : ""}` };
   }
-  const raw = [execution.stdout, execution.stderr].filter(Boolean).join("\n");
-  const output = redactUnityPlanningOutput(summarizeUnityCliText(raw, 4_000, 40));
   const reportedFailure = connectedCommandFailure(execution.stdout, isEval);
   if (reportedFailure) {
     return {
@@ -663,6 +731,37 @@ export async function dispatchUnityPlanningInspection(
   return { outcome: "dispatched", command: request.command, output, truncated: output.length < raw.trim().length };
 }
 
+/** Dispatch the documented Pipeline 0.6 ephemeral run_script form; hotpatch is deliberately not exposed. */
+export async function dispatchUnityPipelineRunScript(
+  request: UnityPipelineRunScriptRequest,
+  options: { cliCommand?: string; timeout?: number; signal?: AbortSignal; execute: UnityCliExecutor; inspect?: (projectRoot: string, unityVersion: string) => Promise<UnityCliProjectCapabilities> },
+): Promise<UnityPlanningInspectionResult> {
+  let projectRoot: string; let file: string;
+  try { projectRoot = await realpath(request.projectRoot); file = await realpath(request.file); } catch {
+    return { outcome: "rejected", code: "run_script_path_unavailable", message: "The project root or existing script file could not be canonicalized." };
+  }
+  const relativeFile = relative(projectRoot, file);
+  let fileStats: Awaited<ReturnType<typeof stat>>;
+  try { fileStats = await stat(file); } catch { return { outcome: "rejected", code: "run_script_file_invalid", message: "run_script requires one readable existing C# file." }; }
+  if (!relativeFile || isAbsolute(relativeFile) || relativeFile === ".." || relativeFile.startsWith(`..${process.platform === "win32" ? "\\" : "/"}`) || !fileStats.isFile() || !/\.cs$/i.test(relativeFile)) return { outcome: "rejected", code: "run_script_file_invalid", message: "run_script requires one existing .cs file inside the exact project root." };
+  let serializedArgs: string;
+  try { serializedArgs = JSON.stringify(request.args ?? []); } catch { return { outcome: "rejected", code: "run_script_args_invalid", message: "run_script arguments must be JSON-serializable." }; }
+  if (serializedArgs === undefined || serializedArgs.length > 4_000 || (request.entry?.length ?? 0) > 500 || /[\u0000-\u001f\u007f]/.test(request.entry ?? "")) return { outcome: "rejected", code: "run_script_args_invalid", message: "run_script entry or JSON arguments exceed bounded request limits." };
+  const inspect = options.inspect ?? ((root, version) => inspectUnityCliProjectCapabilities(root, version, { cliCommand: options.cliCommand, timeout: options.timeout, signal: options.signal, execute: options.execute }));
+  const initial = await inspect(projectRoot, request.unityVersion);
+  if (planningInspectionReadiness(initial) || !initial.advertisedCommands.includes("run_script")) return { outcome: "rejected", code: "run_script_unavailable", message: `The exact reachable Pipeline copy does not establish advertised run_script support.${unityCapabilityDiagnosticSuffix(initial)}` };
+  const refreshed = await inspect(projectRoot, request.unityVersion);
+  if (planningInspectionReadiness(refreshed) || !haveSameKnownProcessIds(initial.matchingInstances, refreshed.matchingInstances) || !refreshed.advertisedCommands.includes("run_script")) return { outcome: "rejected", code: "unity_project_identity_changed", message: `Pipeline identity or run_script availability changed immediately before dispatch.${unityCapabilityDiagnosticSuffix(refreshed)}` };
+  const timeout = Math.max(1, Math.min(options.timeout ?? 30_000, 86_400_000));
+  const args = ["--format", "json", "--no-banner", "--non-interactive", "command", "--project-path", projectRoot, "--timeout", String(Math.ceil(timeout / 1000)), "run_script", "--file", relativeFile, "--mode", "ephemeral", "--args", serializedArgs, "--timeout_ms", String(timeout), ...(request.entry?.trim() ? ["--entry", request.entry.trim()] : []), ...(request.dryRun ? ["--dry_run", "true"] : [])];
+  const execution = await options.execute(resolveUnityCliCommand({ cliCommand: options.cliCommand }), args, { timeout, signal: options.signal });
+  const raw = [execution.stdout, execution.stderr].filter(Boolean).join("\n"); const output = summarizeUnityCliText(redactUnityPlanningOutput(raw), 4_000, 40);
+  if (execution.error) return { outcome: "rejected", code: isUnityCliTimeout(execution) ? "run_script_timeout" : "run_script_failed", message: `run_script did not complete successfully; its effect may be uncertain.${output ? ` ${output}` : ""}` };
+  const failure = runScriptCommandFailure(execution.stdout);
+  if (failure) return { outcome: "rejected", code: failure === "malformed" ? "run_script_malformed" : "run_script_reported_failure", message: `${failure === "malformed" ? "run_script returned malformed JSON evidence" : "run_script reported failure"}.${output ? ` ${output}` : ""}` };
+  return { outcome: "dispatched", command: "run_script", output, truncated: output.length < raw.trim().length };
+}
+
 /** Keep connected inspection output useful without returning common credential forms verbatim. */
 export function redactUnityPlanningOutput(value: string): string {
   return value

package/src/unity-core.ts

@@ -104,8 +104,8 @@ export function normalizeForCommandSearch(value: string, platform: SupportedPlat
   return platform === "win32" ? normalized.toLowerCase() : normalized;
 }
 
-function realpathMatchesOnDarwin(candidatePath: string, projectRoot: string, platform: SupportedPlatform): boolean | null {
-  if (platform !== "darwin" || process.platform !== "darwin") {
+function realpathMatchesOnNativePlatform(candidatePath: string, projectRoot: string, platform: SupportedPlatform): boolean | null {
+  if ((platform !== "darwin" && platform !== "win32") || process.platform !== platform) {
     return null;
   }
 
@@ -113,7 +113,7 @@ function realpathMatchesOnDarwin(candidatePath: string, projectRoot: string, pla
     return realpathSync.native(candidatePath) === realpathSync.native(projectRoot);
   } catch {
     // Only use filesystem identity when both paths can be resolved. Falling back
-    // to the case-sensitive textual comparison preserves case-sensitive APFS.
+    // to the platform-specific textual comparison preserves existing behavior.
     return null;
   }
 }
@@ -130,7 +130,7 @@ export function projectPathsMatch(candidatePath: string, projectRoot: string, pl
     return false;
   }
 
-  const realpathMatch = realpathMatchesOnDarwin(trimmedCandidatePath, trimmedProjectRoot, comparisonPlatform);
+  const realpathMatch = realpathMatchesOnNativePlatform(trimmedCandidatePath, trimmedProjectRoot, comparisonPlatform);
   if (realpathMatch !== null) {
     return realpathMatch;
   }

package/src/unity-pipeline.ts

@@ -1,6 +1,6 @@
 import { realpath } from "node:fs/promises";
 import { projectPathsMatch } from "./unity-core";
-import { resolveUnityCliCommand, type UnityCliExecResult, type UnityCliExecutor, type UnityCliProjectCapabilities } from "./unity-cli";
+import { redactUnityPlanningOutput, resolveUnityCliCommand, summarizeUnityCliText, unityCapabilityDiagnosticSuffix, type UnityCliExecResult, type UnityCliExecutor, type UnityCliProjectCapabilities } from "./unity-cli";
 
 /** Public limits are deliberately small enough that connected work cannot create an unbounded agent wait loop. */
 export const UNITY_PIPELINE_COMPILE_TIMEOUT_SECONDS = 180;
@@ -31,6 +31,8 @@ export type UnityPipelineOperationDetails = {
   testPlatform?: "EditMode" | "PlayMode";
   testFilter?: string;
   counts?: { total: number; passed?: number; failed: number; inconclusive?: number };
+  /** Bounded nonfatal Pipeline envelope guidance; never compiler-error evidence. */
+  warnings?: string[];
 };
 export type UnityPipelineTestRecord = { name: string; status: string; durationSeconds?: number; message?: string; stackTrace?: string };
 /** testRecords are terminal evidence for the caller's durable artifact only; do not expose them in tool details. */
@@ -72,6 +74,39 @@ function bounded(value: string, limit = UNITY_PIPELINE_MAX_STACK_CHARS): string
   const oneLine = value.replace(/[\r\n\t]+/g, " ").replace(/\s+/g, " ").trim();
   return oneLine.length > limit ? `${oneLine.slice(0, limit - 1)}…` : oneLine;
 }
+function pipelineEnvelopeWarnings(output: string): string[] {
+  const outer = (() => { try { return record(JSON.parse(output)); } catch { return undefined; } })();
+  const data = record(outer?.data);
+  const entries = [outer?.warnings, data?.warnings].flatMap(value => Array.isArray(value) ? value.slice(0, UNITY_PIPELINE_MAX_DIAGNOSTICS) : []);
+  return entries.slice(0, UNITY_PIPELINE_MAX_DIAGNOSTICS).flatMap(entry => {
+    const item = record(entry); const message = string(field(item ?? {}, "message", "detail", "warning")) ?? string(entry);
+    return message ? [bounded(redactUnityPlanningOutput(message), 1_000)] : [];
+  });
+}
+function retainWarnings(warnings: string[], output: string): void {
+  for (const warning of pipelineEnvelopeWarnings(output)) {
+    if (warnings.length >= UNITY_PIPELINE_MAX_DIAGNOSTICS) break;
+    if (!warnings.includes(warning)) warnings.push(warning);
+  }
+}
+function warningText(warnings: string[]): string {
+  return warnings.length ? `\nPipeline warnings: ${warnings.join("; ")}` : "";
+}
+function pipelineFailureDiagnostic(response: UnityCliExecResult): string {
+  const raw = [response.stdout, response.stderr, response.error?.message].filter(Boolean).join("\n");
+  let message: string | undefined;
+  try {
+    const outer = record(JSON.parse(response.stdout)); const data = record(outer?.data);
+    const errors = Array.isArray(outer?.errors) ? outer?.errors : [];
+    const first = record(errors[0]);
+    const messages = [
+      string(field(first ?? {}, "code")), string(field(first ?? {}, "message", "detail")),
+      ...[data, outer].flatMap(item => ["error", "errordetails", "message"].map(key => string(field(item ?? {}, key)))),
+    ].filter((value): value is string => Boolean(value));
+    message = [...new Set(messages)].join("; ") || undefined;
+  } catch { /* retain bounded native raw fallback */ }
+  return summarizeUnityCliText(redactUnityPlanningOutput(message ?? raw), 1_000, 10);
+}
 function throwIfAborted(signal?: AbortSignal): void {
   if (signal?.aborted) throw new Error("Unity Pipeline operation aborted; its Editor operation may still be running.");
 }
@@ -102,8 +137,10 @@ export function parseUnityPipelineEnvelope(output: string): ParsedEnvelope {
   let outer: RecordValue | undefined;
   try { outer = record(JSON.parse(output)); } catch { return { result: {}, outerSuccess: false, malformed: "Unity Pipeline returned malformed JSON." }; }
   if (!outer) return { result: {}, outerSuccess: false, malformed: "Unity Pipeline returned a non-object JSON envelope." };
+  // Pipeline 0.6 may return the compact exec envelope { success, result, warnings }
+  // rather than the older CLI wrapper { success, data: { result } }.
   const data = record(outer.data);
-  const rawResult = data?.result ?? data;
+  const rawResult = data?.result ?? outer.result ?? data;
   if (typeof rawResult === "string") {
     try {
       const parsed = record(JSON.parse(rawResult));
@@ -148,15 +185,14 @@ function diagnostics(result: RecordValue): string[] {
   });
   return [...new Set(values)].slice(0, UNITY_PIPELINE_MAX_DIAGNOSTICS);
 }
-/** True only for Pipeline 0.5's explicit rejected outer envelope, before a main-thread command is dispatched. */
-export function isUnityPipelineInitialSettlingBusy(output: string): boolean {
-  let outer: RecordValue | undefined;
-  try { outer = record(JSON.parse(output)); } catch { return false; }
-  const data = record(outer?.data);
-  return outer?.success === false
-    && string(field(data ?? {}, "error"))?.toLowerCase() === "server busy"
-    && statusOf(data ?? {}) === "busy"
-    && field(data ?? {}, "retryable") === true;
+/**
+ * Do not retry generic busy envelopes. Pipeline 0.6 uses busy for modal dialogs as
+ * well as startup settling, and neither is distinguishable in the legacy CLI-shaped
+ * response. Retrying would blindly repeat a command blocked by a user-visible modal.
+ * A future retry must be wired only to a source-verified pre-dispatch discriminator.
+ */
+export function isUnityPipelineInitialSettlingBusy(_output: string): boolean {
+  return false;
 }
 
 export function normalizeUnityPipelineCompile(output: string): NormalizedCompile {
@@ -363,7 +399,7 @@ async function dispatchMainThreadCommand(deps: PipelineDependencies, projectRoot
 
 async function requirePreflight(deps: PipelineDependencies, projectRoot: string, unityVersion: string, commands: string[], operation: "recompile" | "tests", signal: AbortSignal | undefined, deadline: number, now: () => number, allowAutonomousExitPlayMode = true): Promise<{ capabilities: UnityCliProjectCapabilities; exitedPlayMode: boolean; playModeHandling: UnityPipelinePlayModeHandling; scriptChangesWhilePlaying?: UnityScriptChangesWhilePlayingPolicy }> {
   const capabilities = await inspectWithDeadline(deps, projectRoot, unityVersion, signal, deadline, now, "preflight");
-  const error = capabilityError(capabilities, commands); if (error) throw new Error(error);
+  const error = capabilityError(capabilities, commands); if (error) throw new Error(error + unityCapabilityDiagnosticSuffix(capabilities));
   let editor = await executeCommand(deps, projectRoot, "editor_status", [], signal, deadline, now);
   if (editor.error) throw new Error("Unity Pipeline editor_status failed; operation not started.");
   let status = editorStatus(editor.stdout);
@@ -450,11 +486,12 @@ export async function runUnityPipelineRecompile(request: UnityPipelineCompileReq
   ensureBeforeDeadline(deadline, now, "recompile before dispatch");
   throwIfAborted(signal);
   const dispatched = await dispatchMainThreadCommand(deps, projectRoot, "recompile", [], "recompile", signal, deadline, now, sleep);
-  if (dispatched.error) throw new Error("Unity Pipeline recompile dispatch failed; operation may not have started.");
+  if (dispatched.error) throw new Error(`Unity Pipeline recompile dispatch failed; operation may not have started.${pipelineFailureDiagnostic(dispatched) ? ` ${pipelineFailureDiagnostic(dispatched)}` : ""}`);
+  const dispatchWarnings = pipelineEnvelopeWarnings(dispatched.stdout);
   let state = normalizeUnityPipelineCompile(dispatched.stdout);
   if (state.state === "uncertain") throw new Error("Unity Pipeline recompile dispatch returned malformed or uncertain evidence; operation may have started.");
   if (state.state === "failed") throw new Error(`Unity recompile failed: ${state.diagnostics.join("; ") || "compiler failure reported"}`);
-  if (state.state === "up_to_date") return { text: `${lifecyclePrefix}Unity scripts are up to date for ${projectRoot}; no compilation was triggered.`,  details: { projectRoot, operation: "recompile", terminalState: "up_to_date", elapsedSeconds: elapsed(start, now), compilationTriggered: false, ...playModeDetails(preflight) } };
+  if (state.state === "up_to_date") return { text: `${lifecyclePrefix}Unity scripts are up to date for ${projectRoot}; no compilation was triggered.${warningText(dispatchWarnings)}`,   details: { projectRoot, operation: "recompile", terminalState: "up_to_date", elapsedSeconds: elapsed(start, now), compilationTriggered: false, ...(dispatchWarnings.length ? { warnings: dispatchWarnings } : {}), ...playModeDetails(preflight) } };
   for (let poll = 0; now() < deadline; poll += 1) {
     options.onUpdate?.(`Unity recompile ${state.state}; ${elapsed(start, now).toFixed(1)}s elapsed.`);
     const delay = Math.min(UNITY_PIPELINE_BACKOFF_SECONDS[Math.min(poll, UNITY_PIPELINE_BACKOFF_SECONDS.length - 1)]! * 1000, deadline - now());
@@ -467,9 +504,10 @@ export async function runUnityPipelineRecompile(request: UnityPipelineCompileReq
     const response = await executeCommand(deps, projectRoot, "recompile_status", [], signal, deadline, now);
     if (response.error) continue; // Domain reload can briefly disconnect the same exact copy.
     if (isUnityPipelineInitialSettlingBusy(response.stdout)) continue;
+    retainWarnings(dispatchWarnings, response.stdout);
     state = normalizeUnityPipelineCompile(response.stdout);
     if (state.state === "failed") throw new Error(`Unity recompile failed: ${state.diagnostics.join("; ") || "compiler failure reported"}`);
-    if (state.state === "completed" || state.state === "up_to_date") return { text: `${lifecyclePrefix}Unity recompile completed for ${projectRoot} in ${elapsed(start, now).toFixed(1)}s; 0 compiler errors.`,  details: { projectRoot, operation: "recompile", terminalState: state.state, elapsedSeconds: elapsed(start, now), compilationTriggered: true, ...playModeDetails(preflight) } };
+    if (state.state === "completed" || state.state === "up_to_date") return { text: `${lifecyclePrefix}Unity recompile completed for ${projectRoot} in ${elapsed(start, now).toFixed(1)}s; 0 compiler errors.${warningText(dispatchWarnings)}`,  details: { projectRoot, operation: "recompile", terminalState: state.state, elapsedSeconds: elapsed(start, now), compilationTriggered: true, ...(dispatchWarnings.length ? { warnings: dispatchWarnings } : {}), ...playModeDetails(preflight) } };
     if (state.state === "uncertain") throw new Error("Unity Pipeline recompile status is malformed or uncertain; operation may still be running.");
   }
   throw timeoutMessage("recompile");
@@ -498,7 +536,8 @@ export async function runUnityPipelineTests(request: UnityPipelineTestRequest, d
   const args = ["--mode", request.testPlatform === "EditMode" ? "editor" : "playmode", ...selectorArgs, "--async_tests", "true"];
   ensureBeforeDeadline(deadline, now, "tests before dispatch"); throwIfAborted(signal);
   const dispatched = await dispatchMainThreadCommand(deps, projectRoot, "run_tests", args, "tests", signal, deadline, now, sleep);
-  if (dispatched.error) throw new Error("Unity Pipeline test dispatch failed; test run may not have started.");
+  if (dispatched.error) throw new Error(`Unity Pipeline test dispatch failed; test run may not have started.${pipelineFailureDiagnostic(dispatched) ? ` ${pipelineFailureDiagnostic(dispatched)}` : ""}`);
+  const dispatchWarnings = pipelineEnvelopeWarnings(dispatched.stdout);
   let state = normalizeUnityPipelineTest(dispatched.stdout);
   if (state.state === "uncertain" || state.state === "inactive") throw new Error("Unity Pipeline test dispatch returned inactive, malformed, or uncertain evidence; test run may not have started.");
   if (state.state === "failed" || state.state === "cancelled") throw new Error(`Unity ${request.testPlatform} tests failed: ${state.failures.join("; ") || state.state}.`);
@@ -509,7 +548,7 @@ export async function runUnityPipelineTests(request: UnityPipelineTestRequest, d
   if (state.state === "completed") {
     const counts = passingCounts(state);
     if (!counts) throw new Error("Unity test result is terminal but lacks passing evidence (consistent positive total, passed count, and reported zero failures).");
-    return { text: `${lifecyclePrefix}Unity ${request.testPlatform} tests passed for ${projectRoot}: ${counts.total} executed, ${counts.passed} passed, 0 failed in ${elapsed(start, now).toFixed(2)}s.`, details: { projectRoot, operation: "tests", terminalState: "completed", elapsedSeconds: elapsed(start, now), ...playModeDetails(preflight), testPlatform: request.testPlatform, testFilter: request.testFilter ?? request.testCategory, counts }, testRecords: state.testRecords };
+    return { text: `${lifecyclePrefix}Unity ${request.testPlatform} tests passed for ${projectRoot}: ${counts.total} executed, ${counts.passed} passed, 0 failed in ${elapsed(start, now).toFixed(2)}s.${warningText(dispatchWarnings)}`, details: { projectRoot, operation: "tests", terminalState: "completed", elapsedSeconds: elapsed(start, now), ...playModeDetails(preflight), testPlatform: request.testPlatform, testFilter: request.testFilter ?? request.testCategory, counts, ...(dispatchWarnings.length ? { warnings: dispatchWarnings } : {}) }, testRecords: state.testRecords };
   }
   for (let poll = 0; now() < deadline; poll += 1) {
     options.onUpdate?.(`Unity ${request.testPlatform} tests ${state.state}; ${elapsed(start, now).toFixed(1)}s elapsed.`);
@@ -523,6 +562,7 @@ export async function runUnityPipelineTests(request: UnityPipelineTestRequest, d
     const response = await executeCommand(deps, projectRoot, "test_status", [], signal, deadline, now);
     if (response.error) continue;
     if (isUnityPipelineInitialSettlingBusy(response.stdout)) continue;
+    retainWarnings(dispatchWarnings, response.stdout);
     state = normalizeUnityPipelineTest(response.stdout);
     if (!checkCorrelation(expected, state.correlation)) throw new Error("Unity Pipeline test status was displaced by a different run; operation state is uncertain.");
     if (state.state === "failed" || state.state === "cancelled") throw new Error(`Unity ${request.testPlatform} tests failed: ${state.failures.join("; ") || state.state}.`);
@@ -531,7 +571,7 @@ export async function runUnityPipelineTests(request: UnityPipelineTestRequest, d
     if (state.state !== "completed") continue;
     const counts = passingCounts(state);
     if (!counts) throw new Error("Unity test result is terminal but lacks passing evidence (consistent positive total, passed count, and reported zero failures).");
-    return { text: `${lifecyclePrefix}Unity ${request.testPlatform} tests passed for ${projectRoot}: ${counts.total} executed, ${counts.passed} passed, 0 failed in ${elapsed(start, now).toFixed(2)}s.`, details: { projectRoot, operation: "tests", terminalState: "completed", elapsedSeconds: elapsed(start, now), ...playModeDetails(preflight), testPlatform: request.testPlatform, testFilter: request.testFilter ?? request.testCategory, counts }, testRecords: state.testRecords };
+    return { text: `${lifecyclePrefix}Unity ${request.testPlatform} tests passed for ${projectRoot}: ${counts.total} executed, ${counts.passed} passed, 0 failed in ${elapsed(start, now).toFixed(2)}s.${warningText(dispatchWarnings)}`, details: { projectRoot, operation: "tests", terminalState: "completed", elapsedSeconds: elapsed(start, now), ...playModeDetails(preflight), testPlatform: request.testPlatform, testFilter: request.testFilter ?? request.testCategory, counts, ...(dispatchWarnings.length ? { warnings: dispatchWarnings } : {}) }, testRecords: state.testRecords };
   }
   throw timeoutMessage("tests");
 }