gentle-pi 2.4.0 → 2.5.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (116) hide show
  1. package/README.md +170 -12
  2. package/assets/agents/gentle-ai-worker.md +9 -0
  3. package/assets/orchestrator-delegation.md +19 -9
  4. package/assets/orchestrator.md +5 -5
  5. package/docs/delegated-verification.md +25 -0
  6. package/docs/telemetry.md +38 -0
  7. package/extensions/ask-user-choice.ts +26 -20
  8. package/extensions/codegraph-tools.ts +94 -5
  9. package/extensions/gentle-agents.ts +588 -0
  10. package/extensions/gentle-ai.ts +898 -79
  11. package/extensions/gentle-shell.ts +547 -0
  12. package/extensions/gentle-todo.ts +199 -0
  13. package/extensions/quiet-tools.ts +1 -1
  14. package/lib/agents-config.ts +318 -0
  15. package/lib/agents-history.ts +80 -0
  16. package/lib/agents-protocol.ts +429 -0
  17. package/lib/agents-runner.ts +490 -0
  18. package/lib/agents-transcript.ts +87 -0
  19. package/lib/agents-view.ts +557 -0
  20. package/lib/agents-widget.ts +222 -0
  21. package/lib/gentle-ai-renderer.ts +142 -26
  22. package/lib/native-choice-list.ts +194 -0
  23. package/lib/native-fullscreen-interaction.ts +47 -0
  24. package/lib/native-pointer-region.ts +164 -0
  25. package/lib/native-review-cli.ts +88 -12
  26. package/lib/review-candidate-view-owner.ts +177 -0
  27. package/lib/review-candidate-view.ts +127 -35
  28. package/lib/review-consent-ui.ts +65 -0
  29. package/lib/review-integration-v2.ts +58 -8
  30. package/lib/review-last-event-controller.ts +1 -0
  31. package/lib/review-relay-contract.ts +11 -0
  32. package/lib/review-repository.ts +2 -2
  33. package/lib/review-risk-assessment.ts +339 -0
  34. package/lib/review-session-standing-permission-ipc.ts +309 -0
  35. package/lib/review-session-standing-permission.ts +219 -0
  36. package/lib/shell-bar.ts +138 -0
  37. package/lib/shell-card.ts +136 -0
  38. package/lib/shell-changes-view.ts +205 -0
  39. package/lib/shell-changes.ts +210 -0
  40. package/lib/shell-gauge.ts +40 -0
  41. package/lib/shell-prompt.ts +119 -0
  42. package/lib/shell-todo.ts +280 -0
  43. package/lib/shell-usage-view.ts +76 -0
  44. package/lib/shell-usage.ts +246 -0
  45. package/lib/telemetry-trigger.ts +151 -0
  46. package/package.json +4 -4
  47. package/runtime/native-review-cli.mjs +87 -11
  48. package/runtime/review-integration-v2.mjs +58 -8
  49. package/runtime/review-relay-contract.mjs +11 -0
  50. package/runtime/review-risk-assessment.mjs +340 -0
  51. package/runtime/telemetry-trigger.mjs +152 -0
  52. package/scripts/build-runtime-modules.mjs +2 -0
  53. package/scripts/gentle-ai-installer.mjs +10 -10
  54. package/scripts/test-packed-runner.mjs +22 -0
  55. package/scripts/verify-package-files.mjs +6 -2
  56. package/skills/_shared/review-ledger-contract.md +3 -1
  57. package/tests/agents-config.test.ts +143 -0
  58. package/tests/agents-fake-child.ts +52 -0
  59. package/tests/agents-history.test.ts +54 -0
  60. package/tests/agents-protocol.test.ts +153 -0
  61. package/tests/agents-runner-process.test.ts +111 -0
  62. package/tests/agents-runner.test.ts +402 -0
  63. package/tests/agents-transcript.test.ts +30 -0
  64. package/tests/agents-view.test.ts +274 -0
  65. package/tests/agents-widget.test.ts +111 -0
  66. package/tests/ask-user-choice.test.ts +157 -3
  67. package/tests/codegraph-tools.test.ts +110 -1
  68. package/tests/devbinary/native-review-parity.devtest.ts +108 -0
  69. package/tests/fixtures/agents-process-child.mjs +23 -0
  70. package/tests/gentle-agents.test.ts +741 -0
  71. package/tests/gentle-ai-binary.test.ts +1 -1
  72. package/tests/gentle-ai-installer.test.ts +47 -47
  73. package/tests/gentle-ai-renderer.test.ts +65 -0
  74. package/tests/gentle-ai.test.ts +28 -12
  75. package/tests/gentle-card-text.ts +35 -0
  76. package/tests/gentle-shell.test.ts +527 -0
  77. package/tests/gentle-todo.test.ts +182 -0
  78. package/tests/native-choice-list.test.ts +202 -0
  79. package/tests/native-fullscreen-interaction.test.ts +125 -0
  80. package/tests/native-pointer-region.test.ts +245 -0
  81. package/tests/native-review-capability-contract.test.ts +16 -1
  82. package/tests/native-review-cli.test.ts +40 -0
  83. package/tests/native-review-consent.test.ts +91 -0
  84. package/tests/native-review-parity-runtime.test.ts +8 -2
  85. package/tests/native-review-parity.test.ts +29 -22
  86. package/tests/orchestrator-budget.test.ts +69 -0
  87. package/tests/orchestrator-rdd-ownership.test.ts +9 -0
  88. package/tests/package-manifest.test.ts +17 -6
  89. package/tests/quiet-tool-rendering.test.ts +96 -37
  90. package/tests/rdd-aware-verification-contract.test.ts +216 -0
  91. package/tests/rdd-status-line.test.ts +286 -0
  92. package/tests/review-candidate-view.test.ts +452 -6
  93. package/tests/review-contract-prompt.test.ts +3 -0
  94. package/tests/review-controller-native-recovery.test.ts +29 -4
  95. package/tests/review-controller-native-routing.test.ts +321 -4
  96. package/tests/review-controller-workspace-root.test.ts +45 -2
  97. package/tests/review-controller.test.ts +25 -0
  98. package/tests/review-host-relay-routing.test.ts +20 -4
  99. package/tests/review-integration-v2.test.ts +112 -0
  100. package/tests/review-last-event-closure.test.ts +7 -2
  101. package/tests/review-relay-contract.test.ts +26 -0
  102. package/tests/review-repository.test.ts +28 -1
  103. package/tests/review-risk-assessment.test.ts +626 -0
  104. package/tests/review-session-standing-permission-controller.test.ts +608 -0
  105. package/tests/review-session-standing-permission-ipc.test.ts +233 -0
  106. package/tests/review-session-standing-permission-runtime.test.ts +212 -0
  107. package/tests/review-session-standing-permission.test.ts +126 -0
  108. package/tests/shell-bar.test.ts +176 -0
  109. package/tests/shell-card.test.ts +118 -0
  110. package/tests/shell-changes-view.test.ts +146 -0
  111. package/tests/shell-changes.test.ts +182 -0
  112. package/tests/shell-prompt.test.ts +118 -0
  113. package/tests/shell-todo.test.ts +170 -0
  114. package/tests/shell-usage-view.test.ts +62 -0
  115. package/tests/shell-usage.test.ts +197 -0
  116. package/tests/telemetry-trigger.test.ts +349 -0
@@ -1,8 +1,11 @@
1
1
  import assert from "node:assert/strict";
2
2
  import { execFileSync } from "node:child_process";
3
3
  import { chmodSync, existsSync, mkdirSync, mkdtempSync, readFileSync, realpathSync, rmSync, writeFileSync } from "node:fs";
4
- import { tmpdir } from "node:os";
5
- import { join } from "node:path";
4
+ import fs from "node:fs";
5
+ import fsp from "node:fs/promises";
6
+ import os, { tmpdir } from "node:os";
7
+ import { syncBuiltinESMExports } from "node:module";
8
+ import { dirname, join, resolve, sep } from "node:path";
6
9
  import test from "node:test";
7
10
  import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
8
11
  import { __testing, createGentleAiExtension, PendingReviewConsentRegistry } from "../extensions/gentle-ai.ts";
@@ -95,6 +98,37 @@ function burnedAcknowledgementStatus(lineageId: string): ReviewStatusV3 {
95
98
  return burned;
96
99
  }
97
100
 
101
+ // gentle-pi#627: gentle-ai reports a stale managed-asset set as a typed stop
102
+ // carrying the exact `gentle-ai sync` invocation that resolves it.
103
+ function managedAssetsOutdatedStatus(lineageId: string): ReviewStatusV3 {
104
+ const stopped = status(lineageId, [], "approved");
105
+ stopped.nextTransition = { kind: "stop", reasonCode: "managed_assets_outdated", continuation: { operation: "sync", command: "gentle-ai sync --agent claude-code", agent: "claude-code", staleAssets: ["orchestration/claude-code.md"] } };
106
+ return stopped;
107
+ }
108
+
109
+ test("STATUS renders the managed_assets_outdated continuation command as the actionable next step", async () => {
110
+ const lineageId = "managed-assets-outdated";
111
+ const native = { targetStatus: async () => managedAssetsOutdatedStatus(lineageId) } as unknown as NativeReviewCli;
112
+ const blocked = await __testing.executeReviewControllerOperation({ operation: "status", lineageId }, process.cwd(), native);
113
+ assert.equal(blocked.status, "blocked");
114
+ assert.equal(blocked.hint, "run gentle-ai sync --agent claude-code");
115
+
116
+ // an unknown stop reason code keeps rendering as a plain blocked result
117
+ const otherStopNative = { targetStatus: async () => burnedAcknowledgementStatus(lineageId) } as unknown as NativeReviewCli;
118
+ const plainBlocked = await __testing.executeReviewControllerOperation({ operation: "status", lineageId }, process.cwd(), otherStopNative);
119
+ assert.equal(plainBlocked.status, "blocked");
120
+ assert.equal(plainBlocked.hint, undefined);
121
+
122
+ // an older gentle-ai may report this same stop with no continuation; the
123
+ // renderer degrades to a plain blocked result instead of losing the status
124
+ const noContinuation = status(lineageId, [], "approved");
125
+ noContinuation.nextTransition = { kind: "stop", reasonCode: "managed_assets_outdated" };
126
+ const noContinuationNative = { targetStatus: async () => noContinuation } as unknown as NativeReviewCli;
127
+ const degraded = await __testing.executeReviewControllerOperation({ operation: "status", lineageId }, process.cwd(), noContinuationNative);
128
+ assert.equal(degraded.status, "blocked");
129
+ assert.equal(degraded.hint, undefined);
130
+ });
131
+
98
132
  test("public acknowledgement relays one current provider vector and never replays after authority burn", async () => {
99
133
  const lineageId = "acknowledge-approved";
100
134
  const requests: Array<Record<string, unknown>> = [];
@@ -423,6 +457,72 @@ test("REPAIR retains frozen committed collect selectors and leaves workspace rou
423
457
  assert.deepEqual(workspaceRequests, [{ cwd, lineageId: workspaceLineage }]);
424
458
  });
425
459
 
460
+ test("selectorless STATUS resumes a retained committed correction lineage", async (t) => {
461
+ const candidateViews = new CandidateViewRegistry();
462
+ t.after(() => candidateViews.cleanupAll());
463
+ const cwd = repository(t);
464
+ const lineageId = "retained-committed-correction";
465
+ const view = candidateViews.create({
466
+ contributorRoot: cwd,
467
+ baseRef: execFileSync("git", ["rev-parse", "HEAD"], { cwd, encoding: "utf8" }).trim(),
468
+ committedOnly: true,
469
+ });
470
+ candidateViews.retain(view.token, lineageId);
471
+ const frozenTarget = candidateViews.resolveProjection(lineageId, cwd);
472
+ const requests: Array<Record<string, unknown>> = [];
473
+ const selections = new Map();
474
+ let captures = 0;
475
+ const native = {
476
+ targetStatus: async (request: Record<string, unknown>) => {
477
+ requests.push(request);
478
+ return status(lineageId, [correctionPlanInput(lineageId)], "correction_required");
479
+ },
480
+ captureCorrectionPlan: async () => {
481
+ captures += 1;
482
+ return { schema: "gentle-ai.review-last-event-closure/v1", operation: "review.capture-correction-plan", lineageId, state: "correction_required", storeRevision: SHA };
483
+ },
484
+ } as unknown as NativeReviewCli;
485
+
486
+ const listed = await __testing.executeReviewControllerOperation({ operation: "status", lineageId }, cwd, native, undefined, candidateViews, undefined, selections);
487
+ const captured = await __testing.executeReviewCaptureOperation({ lineageId, collectBinding: bindingOf(listed), correctionLines: 1 }, cwd, native, undefined, candidateViews, selections, true);
488
+
489
+ assert.equal(captured.outcome, "native-last-event-closure");
490
+ assert.equal(captures, 1);
491
+ assert.deepEqual(requests, Array.from({ length: 2 }, () => ({ cwd, lineageId, agent: "pi", baseRef: frozenTarget.baseCommit, committedOnly: true })));
492
+ });
493
+
494
+ test("selectorless STATUS uses a retained committed selector only at its retained workspace", async (t) => {
495
+ const candidateViews = new CandidateViewRegistry();
496
+ t.after(() => candidateViews.cleanupAll());
497
+ const retainedRoot = repository(t);
498
+ const otherRoot = repository(t);
499
+ const lineageId = "retained-selector-boundary";
500
+ const view = candidateViews.create({
501
+ contributorRoot: retainedRoot,
502
+ baseRef: execFileSync("git", ["rev-parse", "HEAD"], { cwd: retainedRoot, encoding: "utf8" }).trim(),
503
+ committedOnly: true,
504
+ });
505
+ candidateViews.retain(view.token, lineageId);
506
+ const frozenTarget = candidateViews.resolveProjection(lineageId, retainedRoot);
507
+ const requests: Array<Record<string, unknown>> = [];
508
+ const native = {
509
+ targetStatus: async (request: Record<string, unknown>) => {
510
+ requests.push(request);
511
+ return status(String(request.lineageId ?? "unretained"));
512
+ },
513
+ } as unknown as NativeReviewCli;
514
+
515
+ await __testing.executeReviewControllerOperation({ operation: "status", lineageId, input: JSON.stringify({ baseRef: "explicit-base", committedOnly: true }) }, retainedRoot, native, undefined, candidateViews);
516
+ await __testing.executeReviewControllerOperation({ operation: "status", lineageId: "unretained" }, retainedRoot, native, undefined, candidateViews);
517
+ await __testing.executeReviewControllerOperation({ operation: "status", lineageId }, otherRoot, native, undefined, candidateViews);
518
+
519
+ assert.deepEqual(requests, [
520
+ { cwd: retainedRoot, lineageId, agent: "pi", baseRef: "explicit-base", committedOnly: true },
521
+ { cwd: retainedRoot, lineageId: "unretained", agent: "pi" },
522
+ { cwd: retainedRoot, lineageId, agent: "pi", baseRef: frozenTarget.baseCommit, committedOnly: true },
523
+ ]);
524
+ });
525
+
426
526
  test("STATUS preserves retained intended-untracked selection through selectorless same-lineage replacement collection", async () => {
427
527
  const cwd = process.cwd();
428
528
  const lineageId = "selectorless-replacement";
@@ -638,6 +738,94 @@ function repository(t: test.TestContext): string {
638
738
  return cwd;
639
739
  }
640
740
 
741
+ test("candidate lifecycle sweeps startup and cleans every shutdown including reload", async (t) => {
742
+ const cwd = repository(t);
743
+ for (const key of ["HOME", "GENTLE_PI_AGENT_HOME", "PI_CODING_AGENT_DIR", "GENTLE_PI_CONFIG_HOME", "GENTLE_PI_GENTLE_AI_DEV_BINARY"]) {
744
+ const previous = process.env[key];
745
+ process.env[key] = key === "GENTLE_PI_GENTLE_AI_DEV_BINARY" ? "" : join(cwd, key);
746
+ if (process.env[key]) mkdirSync(process.env[key]!, { recursive: true });
747
+ t.after(() => { if (previous === undefined) delete process.env[key]; else process.env[key] = previous; });
748
+ }
749
+ t.mock.method(os, "homedir", () => join(cwd, "HOME"));
750
+ const homeAgent = join(cwd, "HOME", ".agents", "isolation-probe.md");
751
+ mkdirSync(dirname(homeAgent), { recursive: true });
752
+ writeFileSync(homeAgent, "---\nname: isolation-probe\ndescription: Fixture\n---\n");
753
+ writeFileSync(join(cwd, "GENTLE_PI_CONFIG_HOME", "models.json"), JSON.stringify({ "isolation-probe": { model: "fixture/model" } }));
754
+ t.diagnostic(`startup routing: cwd=${cwd}; HOME, GENTLE_PI_AGENT_HOME, PI_CODING_AGENT_DIR, GENTLE_PI_CONFIG_HOME are same-named children; dev-binary override disabled`);
755
+ const calls: string[] = [];
756
+ const destinations: string[] = [];
757
+ const violations: string[] = [];
758
+ const discovery: string[] = [];
759
+ const assets = resolve("assets");
760
+ const inside = (root: string, path: string) => path === root || path.startsWith(`${root}${sep}`);
761
+ const originalExists = fs.existsSync;
762
+ const originalRealpath = fs.realpathSync;
763
+ // Install guards BEFORE startup, including RED: a swallowed startup error
764
+ // must never hide an attempted external write or read real-home contents.
765
+ for (const [module, names] of [
766
+ [fs, ["existsSync", "lstatSync", "readdirSync", "readFileSync", "mkdirSync", "writeFileSync", "rmSync"]],
767
+ [fsp, ["access", "readdir", "readFile", "mkdir", "writeFile"]],
768
+ ] as const) {
769
+ for (const name of names) {
770
+ const original = (module as any)[name];
771
+ t.mock.method(module as any, name, (value: string, ...args: unknown[]) => {
772
+ const path = resolve(value);
773
+ const writes = /^(mkdir|writeFile|rm)/.test(name);
774
+ const allowed = inside(cwd, path) || (!writes && inside(assets, path));
775
+ if (!allowed && name === "existsSync") return false;
776
+ if (!allowed && ["access", "readdir"].includes(name)) return Promise.reject(Object.assign(new Error("fixture discovery boundary"), { code: "ENOENT" }));
777
+ if (!allowed) { violations.push(`${name}:${path}`); throw new Error("fixture filesystem boundary"); }
778
+ if (writes) {
779
+ let ancestor = path;
780
+ while (!originalExists(ancestor)) ancestor = dirname(ancestor);
781
+ assert.ok(inside(cwd, originalRealpath(ancestor)), "write ancestor escaped fixture");
782
+ destinations.push(path);
783
+ }
784
+ if (name.startsWith("readdir")) discovery.push(path);
785
+ return original(value, ...args);
786
+ });
787
+ }
788
+ }
789
+ syncBuiltinESMExports();
790
+ // Prove the guard refuses an escape without performing the attempted write.
791
+ const escape = join(cwd, "..", "startup-isolation-escape");
792
+ assert.throws(() => fs.writeFileSync(escape, "blocked"), /fixture filesystem boundary/);
793
+ assert.deepEqual(violations, [`writeFileSync:${escape}`]);
794
+ violations.length = 0;
795
+ const hooks = new Map<string, Array<(event: unknown, ctx: ExtensionContext) => unknown>>();
796
+ const registry = new CandidateViewRegistry();
797
+ t.mock.method(registry, "sweepOrphans", (root: string) => { assert.equal(root, cwd); calls.push("sweep"); });
798
+ t.mock.method(registry, "cleanupAll", () => { calls.push("cleanup"); });
799
+ createGentleAiExtension({ nativeReviewCli: null, candidateViews: registry, processEnv: {} })({
800
+ on(name: string, handler: (event: unknown, ctx: ExtensionContext) => unknown) {
801
+ hooks.set(name, [...(hooks.get(name) ?? []), handler]);
802
+ },
803
+ registerTool() {}, registerCommand() {},
804
+ } as unknown as ExtensionAPI);
805
+ try {
806
+ assert.equal(hooks.get("session_start")?.length, 1);
807
+ await hooks.get("session_start")![0]!({ reason: "startup" }, reviewContext(cwd));
808
+ assert.deepEqual(violations, []);
809
+ assert.ok(destinations.length > 0, "startup must actually write fixture assets");
810
+ assert.ok(discovery.includes(join(cwd, "HOME", ".agents")), "home agent discovery must use the fixture");
811
+ assert.equal(existsSync(join(cwd, "GENTLE_PI_AGENT_HOME", "gentle-ai", "managed-assets.json")), true);
812
+ assert.match(readFileSync(homeAgent, "utf8"), /model: fixture\/model/);
813
+ assert.ok(destinations.includes(homeAgent), "the discovered fixture agent must actually receive routing");
814
+ assert.ok(destinations.includes(join(cwd, "GENTLE_PI_AGENT_HOME", "subagents.json")));
815
+ t.diagnostic(`confined startup: ${destinations.length} filesystem mutations; fixture home-agent routing and managed-assets manifest verified; external violations=0`);
816
+ assert.deepEqual(calls, ["sweep"]);
817
+ } finally {
818
+ // Restore only mocks before fixture teardown; never clean external paths.
819
+ t.mock.restoreAll();
820
+ syncBuiltinESMExports();
821
+ }
822
+ t.mock.method(registry, "cleanupAll", () => { calls.push("cleanup"); });
823
+ for (const reason of ["quit", "reload", "new", "resume", "fork"]) {
824
+ for (const hook of hooks.get("session_shutdown")!) await hook({ reason }, reviewContext(cwd));
825
+ }
826
+ assert.deepEqual(calls, ["sweep", ...Array(5).fill("cleanup")]);
827
+ });
828
+
641
829
  interface RegisteredControllerTool {
642
830
  execute: (toolCallId: string, params: unknown, signal: AbortSignal | undefined, onUpdate: undefined, ctx: ExtensionContext) => Promise<{ details?: unknown }>;
643
831
  }
@@ -722,7 +910,7 @@ test("ordinary START binds the native workspace candidate and returns the native
722
910
  assert.equal(startCalls, 1);
723
911
  });
724
912
 
725
- test("pre-lineage intended-untracked selection revalidates and starts at an explicit workspace root", async (t) => {
913
+ for (const statusSchema of ["gentle-ai.review-integration.status/v6", "gentle-ai.review-integration.status/v7"]) test(`pre-lineage intended-untracked selection revalidates and starts at an explicit workspace root (${statusSchema})`, async (t) => {
726
914
  const cwd = realpathSync(repository(t)), sessionCwd = repository(t), eligible = "selected.md";
727
915
  writeFileSync(join(cwd, eligible), "selected\n");
728
916
  const initialTarget = startStatus(cwd), target = startStatus(cwd, undefined, [eligible]);
@@ -735,7 +923,7 @@ test("pre-lineage intended-untracked selection revalidates and starts at an expl
735
923
  ],
736
924
  submission: { operationToken: "status", argumentTokens: ["--contract=gentle-ai.review-integration/v2", "--next-transition=true", "--agent=pi", "--projection=workspace", "--intended-untracked-selection={{value}}"], values: [{ slot: "intended_untracked_selection", domain: "schema_bound_json", schema: "gentle-ai.review-intended-untracked-selection/v1", substitutionLocation: 4 }] },
737
925
  };
738
- const initial = { ...initialTarget, nextTransition: { kind: "collect", reasonCode: "intended_untracked_selection_required", collect: { inputs: [selection] } }, raw: { schema: "gentle-ai.review-integration.status/v6" } } as ReviewStatusV3;
926
+ const initial = { ...initialTarget, nextTransition: { kind: "collect", reasonCode: "intended_untracked_selection_required", collect: { inputs: [selection] } }, raw: { schema: statusSchema } } as ReviewStatusV3;
739
927
  const requests: Array<Record<string, unknown>> = [], starts: Array<Record<string, unknown>> = [], retained = new Map();
740
928
  const native = {
741
929
  reviewMode: async () => ({ operation: "status", scope: "clone", status: { global: "on", cloneLocal: "on", effective: "on", source: "clone_local" } }),
@@ -778,6 +966,22 @@ test("ordinary START relays native consent without authoring or advancing it", a
778
966
  assert.equal(result.mutation_outcome, "none");
779
967
  });
780
968
 
969
+ test("shutdown preserves failed consent candidates without interrupting session teardown", async (t) => {
970
+ const cwd = repository(t);
971
+ const consent = decodeReviewConsentV3(JSON.parse(readFileSync(join(process.cwd(), "tests", "fixtures", "devbinary", "consent-v3.captured.json"), "utf8")));
972
+ const native = {
973
+ targetStatus: async () => startStatus(cwd),
974
+ start: async () => { throw new NativeReviewConsentRequiredError(consent); },
975
+ } as unknown as NativeReviewCli;
976
+ const registry = new CandidateViewRegistry();
977
+ const runtime = reviewRuntime(native, registry);
978
+ const ctx = reviewContext(cwd);
979
+ const result = await runtime.controller.execute("pending", { operation: "start", input: JSON.stringify({ mode: "ordinary" }) }, undefined, undefined, ctx);
980
+ assert.equal((result.details as { outcome: string }).outcome, "native-review-consent-required");
981
+ t.mock.method(registry, "cleanup", () => { throw new Error("fixture unsafe ownership"); });
982
+ assert.doesNotThrow(() => runtime.sessionShutdown({ reason: "reload" }, ctx));
983
+ });
984
+
781
985
  test("ordinary START obeys the review-mode kill switch before target STATUS", async () => {
782
986
  let targetCalls = 0;
783
987
  const native = {
@@ -1220,6 +1424,119 @@ test("ordinary START refuses a target projection that no longer matches the froz
1220
1424
  assert.equal(startCalls, 0);
1221
1425
  });
1222
1426
 
1427
+ // gentle-pi#455: a consent binding one active Pi session's START created must
1428
+ // be answerable from another active session presenting the same opaque
1429
+ // binding id -- consent bindings are not partitioned by which session's
1430
+ // START created them, only by the repository and candidate they are scoped to.
1431
+ test("answer-consent resolves a valid binding presented by a different active Pi session", async (t) => {
1432
+ const cwd = repository(t);
1433
+ const consent = decodeReviewConsentV3(JSON.parse(readFileSync(join(process.cwd(), "tests", "fixtures", "devbinary", "consent-v3.captured.json"), "utf8")));
1434
+ const registry = new PendingReviewConsentRegistry();
1435
+ const sessionA = Symbol("session-a"), sessionB = Symbol("session-b");
1436
+ let answerCalls = 0;
1437
+ const native = {
1438
+ targetStatus: async () => startStatus(cwd),
1439
+ start: async () => { throw new NativeReviewConsentRequiredError(consent); },
1440
+ answerConsent: async () => {
1441
+ answerCalls += 1;
1442
+ return { kind: "granted", start: { lineageId: "cross-session", state: "reviewing", riskLevel: "low", selectedLenses: [], changedFiles: 1, changedLines: 1, correctionBudget: 1, action: "created", lensesRequired: false, riskReasons: [] } };
1443
+ },
1444
+ } as unknown as NativeReviewCli;
1445
+ const started = await __testing.executeReviewControllerOperation({ operation: "start", input: JSON.stringify({ mode: "ordinary" }) }, cwd, native, undefined, undefined, undefined, undefined, registry, sessionA);
1446
+ assert.equal(typeof started.consent_binding, "string", JSON.stringify(started));
1447
+ const answered = await __testing.executeReviewControllerOperation({ operation: "answer-consent", input: JSON.stringify({ consentBinding: started.consent_binding, answer: "granted" }) }, cwd, native, undefined, undefined, undefined, undefined, registry, sessionB);
1448
+ assert.equal(answerCalls, 1, JSON.stringify(answered));
1449
+ assert.equal((answered.result as { lineage_id?: string } | undefined)?.lineage_id, "cross-session", JSON.stringify(answered));
1450
+ assert.notEqual(answered.status, "blocked", JSON.stringify(answered));
1451
+ });
1452
+
1453
+ // gentle-pi#455: cross-session resolution must still enforce single use --
1454
+ // once a binding is answered by any session, no session (including the one
1455
+ // whose START created it) may answer it again.
1456
+ test("a consumed consent binding cannot be answered a second time from any session", async (t) => {
1457
+ const cwd = repository(t);
1458
+ const consent = decodeReviewConsentV3(JSON.parse(readFileSync(join(process.cwd(), "tests", "fixtures", "devbinary", "consent-v3.captured.json"), "utf8")));
1459
+ const registry = new PendingReviewConsentRegistry();
1460
+ const sessionA = Symbol("session-a"), sessionB = Symbol("session-b");
1461
+ let answerCalls = 0;
1462
+ const native = {
1463
+ targetStatus: async () => startStatus(cwd),
1464
+ start: async () => { throw new NativeReviewConsentRequiredError(consent); },
1465
+ answerConsent: async () => {
1466
+ answerCalls += 1;
1467
+ return { kind: "granted", start: { lineageId: "single-use", state: "reviewing", riskLevel: "low", selectedLenses: [], changedFiles: 1, changedLines: 1, correctionBudget: 1, action: "created", lensesRequired: false, riskReasons: [] } };
1468
+ },
1469
+ } as unknown as NativeReviewCli;
1470
+ const started = await __testing.executeReviewControllerOperation({ operation: "start", input: JSON.stringify({ mode: "ordinary" }) }, cwd, native, undefined, undefined, undefined, undefined, registry, sessionA);
1471
+ const binding = started.consent_binding as string;
1472
+ const first = await __testing.executeReviewControllerOperation({ operation: "answer-consent", input: JSON.stringify({ consentBinding: binding, answer: "granted" }) }, cwd, native, undefined, undefined, undefined, undefined, registry, sessionB);
1473
+ assert.equal(answerCalls, 1, JSON.stringify(first));
1474
+ const replay = await __testing.executeReviewControllerOperation({ operation: "answer-consent", input: JSON.stringify({ consentBinding: binding, answer: "granted" }) }, cwd, native, undefined, undefined, undefined, undefined, registry, sessionA);
1475
+ assert.equal(answerCalls, 1, JSON.stringify(replay));
1476
+ assert.equal(replay.status, "blocked", JSON.stringify(replay));
1477
+ assert.equal(replay.outcome, "consent-binding-stale", JSON.stringify(replay));
1478
+ assert.equal((replay.diagnostics as { code?: string } | undefined)?.code, "consent-binding-already-consumed", JSON.stringify(replay));
1479
+ });
1480
+
1481
+ // gentle-pi#455 correction: cross-session resolution looks a binding up by
1482
+ // its opaque id alone, so it must independently refuse an answer presented
1483
+ // from a different repository than the one its owning START minted it for,
1484
+ // without ever running the mode gate or native answerConsent, and without
1485
+ // consuming the binding -- leaving it answerable from the right repository.
1486
+ test("answer-consent refuses a binding presented from a different repository than the one its START minted, and leaves it answerable from the right one", async (t) => {
1487
+ const cwdA = repository(t), cwdB = repository(t);
1488
+ const consent = decodeReviewConsentV3(JSON.parse(readFileSync(join(process.cwd(), "tests", "fixtures", "devbinary", "consent-v3.captured.json"), "utf8")));
1489
+ const registry = new PendingReviewConsentRegistry();
1490
+ const sessionA = Symbol("session-a"), sessionB = Symbol("session-b");
1491
+ let answerCalls = 0;
1492
+ const native = {
1493
+ targetStatus: async () => startStatus(cwdA),
1494
+ start: async () => { throw new NativeReviewConsentRequiredError(consent); },
1495
+ answerConsent: async () => {
1496
+ answerCalls += 1;
1497
+ return { kind: "granted", start: { lineageId: "right-repo", state: "reviewing", riskLevel: "low", selectedLenses: [], changedFiles: 1, changedLines: 1, correctionBudget: 1, action: "created", lensesRequired: false, riskReasons: [] } };
1498
+ },
1499
+ } as unknown as NativeReviewCli;
1500
+ const started = await __testing.executeReviewControllerOperation({ operation: "start", input: JSON.stringify({ mode: "ordinary" }) }, cwdA, native, undefined, undefined, undefined, undefined, registry, sessionA);
1501
+ const binding = started.consent_binding as string;
1502
+ const wrongRepo = await __testing.executeReviewControllerOperation({ operation: "answer-consent", input: JSON.stringify({ consentBinding: binding, answer: "granted" }) }, cwdB, native, undefined, undefined, undefined, undefined, registry, sessionB);
1503
+ assert.equal(answerCalls, 0, JSON.stringify(wrongRepo));
1504
+ assert.equal(wrongRepo.status, "blocked", JSON.stringify(wrongRepo));
1505
+ assert.equal(wrongRepo.outcome, "consent-binding-repository-mismatch", JSON.stringify(wrongRepo));
1506
+ assert.equal((wrongRepo.diagnostics as { code?: string } | undefined)?.code, "consent-binding-repository-mismatch", JSON.stringify(wrongRepo));
1507
+ assert.equal(wrongRepo.mutation_performed, false, JSON.stringify(wrongRepo));
1508
+ const rightRepo = await __testing.executeReviewControllerOperation({ operation: "answer-consent", input: JSON.stringify({ consentBinding: binding, answer: "granted" }) }, cwdA, native, undefined, undefined, undefined, undefined, registry, sessionA);
1509
+ assert.equal(answerCalls, 1, JSON.stringify(rightRepo));
1510
+ assert.equal((rightRepo.result as { lineage_id?: string } | undefined)?.lineage_id, "right-repo", JSON.stringify(rightRepo));
1511
+ });
1512
+
1513
+ // gentle-pi#323: within the live consent TTL window, a content-independent
1514
+ // replay key let a second ordinary START reuse the first START's still-live
1515
+ // (never lineage-bound) frozen candidate view even though the live candidate
1516
+ // content changed underneath it, and then dead-ended at
1517
+ // candidate-target-projection-drift instead of minting a fresh view and a
1518
+ // fresh consent envelope. An intended-untracked selection is retained across
1519
+ // both calls so the pre-existing empty-untracked drift-recovery branch does
1520
+ // not mask the general content-change gap this covers.
1521
+ test("ordinary START mints a fresh candidate view when candidate content changes within the live consent TTL window", async (t) => {
1522
+ const candidateViews = new CandidateViewRegistry();
1523
+ t.after(() => candidateViews.cleanupAll());
1524
+ const cwd = repository(t);
1525
+ writeFileSync(join(cwd, "extra.md"), "kept\n");
1526
+ const consent = decodeReviewConsentV3(JSON.parse(readFileSync(join(process.cwd(), "tests", "fixtures", "devbinary", "consent-v3.captured.json"), "utf8")));
1527
+ const native = {
1528
+ targetStatus: async () => startStatus(cwd, undefined, ["extra.md"]),
1529
+ start: async () => { throw new NativeReviewConsentRequiredError(consent); },
1530
+ } as unknown as NativeReviewCli;
1531
+ const first = await __testing.executeReviewControllerOperation({ operation: "start", input: JSON.stringify({ mode: "ordinary" }) }, cwd, native, undefined, candidateViews);
1532
+ assert.equal(first.outcome, "native-review-consent-required", JSON.stringify(first));
1533
+ writeFileSync(join(cwd, "tracked.txt"), "candidate two\n");
1534
+ const second = await __testing.executeReviewControllerOperation({ operation: "start", input: JSON.stringify({ mode: "ordinary" }) }, cwd, native, undefined, candidateViews);
1535
+ assert.notEqual(second.outcome, "native-operation-failed", JSON.stringify(second));
1536
+ assert.equal(second.outcome, "native-review-consent-required", JSON.stringify(second));
1537
+ assert.notEqual(second.consent_binding, first.consent_binding, JSON.stringify(second));
1538
+ });
1539
+
1223
1540
  test("controller-owned dispatch confines single and parallel graph actors to the current candidate view", async (t) => {
1224
1541
  const cwd = repository(t);
1225
1542
  const candidateViews = new CandidateViewRegistry();
@@ -6,9 +6,9 @@ import { join } from "node:path";
6
6
  import test from "node:test";
7
7
  import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
8
8
  import { __testing, createGentleAiExtension } from "../extensions/gentle-ai.ts";
9
- import type { NativeReviewCli } from "../lib/native-review-cli.ts";
9
+ import { NativeReviewIntegrationError, type NativeReviewCli } from "../lib/native-review-cli.ts";
10
10
  import { CandidateViewRegistry } from "../lib/review-candidate-view.ts";
11
- import type { AuthorityRepairAssessmentV1, ReviewStatusV3 } from "../lib/review-integration-v2.ts";
11
+ import { decodeReviewFailureV2, type AuthorityRepairAssessmentV1, type ReviewStatusV3 } from "../lib/review-integration-v2.ts";
12
12
 
13
13
  interface RegisteredTool {
14
14
  execute: (
@@ -240,6 +240,49 @@ test("INSPECT and STATUS operate on the explicit workspace root while the sessio
240
240
  assert.deepEqual(observedCwds, [realpathSync(worktree), realpathSync(worktree), sessionCwd]);
241
241
  });
242
242
 
243
+ // gentle-pi#599: inspect without an explicit workspaceRoot against a parent
244
+ // repository that wraps an untracked nested Git repository fails at the
245
+ // native preflight with a decoded, actionable `invalid_request` envelope. The
246
+ // facade used to discard that envelope's cause, code, retry_safe, and
247
+ // next_action in favor of a generic `native-status-unavailable` outcome
248
+ // indistinguishable from authority corruption. It must preserve the native
249
+ // envelope instead.
250
+ test("inspect without workspaceRoot preserves the native invalid_request cause for a nested foreign Git repository", async (t) => {
251
+ const sessionCwd = repository(t);
252
+ const { controller } = runtime(fakeNative({
253
+ targetStatus: async () => {
254
+ throw new NativeReviewIntegrationError(decodeReviewFailureV2({
255
+ schema: "gentle-ai.review-integration.failure/v2",
256
+ contract: "gentle-ai.review-integration/v2",
257
+ operation: "review.status",
258
+ phase: "preflight",
259
+ code: "invalid_request",
260
+ message: "The negotiated review request is invalid.",
261
+ mutation_outcome: "not_started",
262
+ authority_applicability: "not_evaluated",
263
+ retry_safe: true,
264
+ replayability: "not_replayable",
265
+ required_inputs: [],
266
+ next_action: "correct_request",
267
+ cause: "untracked directory \"engram\" holds another Git repository that is not a linked worktree of this one, so it cannot enter the review candidate: add it to .gitignore, move it outside this repository, or register it as a linked worktree",
268
+ }));
269
+ },
270
+ }));
271
+ const result = await controller.execute("inspect-nested-foreign-repo", { operation: "inspect" }, undefined, undefined, context(sessionCwd));
272
+ const details = result.details as {
273
+ status?: string;
274
+ outcome?: string;
275
+ next_action?: string;
276
+ native_failure?: { code?: string; cause?: string; retry_safe?: boolean };
277
+ };
278
+ assert.equal(details.status, "blocked");
279
+ assert.equal(details.outcome, "native-status-unavailable");
280
+ assert.equal(details.native_failure?.code, "invalid_request");
281
+ assert.equal(details.native_failure?.retry_safe, true);
282
+ assert.match(details.native_failure?.cause ?? "", /engram/);
283
+ assert.equal(details.next_action, "correct_request");
284
+ });
285
+
243
286
  test("START freezes the candidate from the explicit workspace root and returns the actor binding envelope", async (t) => {
244
287
  const sessionCwd = repository(t);
245
288
  writeFileSync(join(sessionCwd, "unrelated.ts"), "export const unrelated = true;\n");
@@ -363,6 +363,7 @@ test("general STATUS returns the typed native-status-unsupported boundary withou
363
363
  mutation_performed: false,
364
364
  inventory_complete: false,
365
365
  next_action: "require-upstream-read-only-native-status-inventory",
366
+ remediation_command: "gentle-ai review status --cwd <repo> --contract gentle-ai.review-integration/v2 --agent pi --next-transition",
366
367
  evidence: {
367
368
  native_contract: "gentle-ai/2.1.4",
368
369
  general_status: "unsupported",
@@ -371,6 +372,30 @@ test("general STATUS returns the typed native-status-unsupported boundary withou
371
372
  });
372
373
  });
373
374
 
375
+ test("gentle-pi#185: general STATUS on a non-negotiated native CLI names the exact status command to run", async (t) => {
376
+ // The legacy `correctionForecast` restoration guard this issue originally
377
+ // reported (extensions/gentle-ai.ts, then around line 5329) was scoped to
378
+ // `targetStatus !== undefined` and skipped restoring a candidate view when
379
+ // the native CLI lacked negotiated STATUS support, reproducing the #176
380
+ // empty-registry failure. That entire manual FINALIZE lifecycle (and the
381
+ // `correctionForecast` guard with it) has since been replaced by the
382
+ // STATUS/capture-driven flow: every operation now checks negotiated STATUS
383
+ // support up front, so no candidate-view restoration is ever attempted on
384
+ // a non-negotiated CLI. This test proves the current equivalent boundary
385
+ // (`nativeStatusUnsupported`) fails closed with an actionable envelope
386
+ // instead of a bare machine token, so a non-negotiated CLI can never
387
+ // surface as a silent empty-registry failure.
388
+ const fixture = createRepository(t);
389
+ const { controller } = registerRuntime();
390
+ const result = await controllerCall(controller, extensionContext(fixture.repository), { operation: "status" });
391
+ assert.equal(result.status, "blocked");
392
+ assert.equal(result.outcome, "native-status-unsupported");
393
+ assert.equal(
394
+ result.remediation_command,
395
+ "gentle-ai review status --cwd <repo> --contract gentle-ai.review-integration/v2 --agent pi --next-transition",
396
+ );
397
+ });
398
+
374
399
  test("failed START gives exact mode and serialization guidance and creates no lineage", async (t) => {
375
400
  const fixture = createRepository(t);
376
401
  const { controller } = registerRuntime();
@@ -142,9 +142,22 @@ function finalizeStatus(lineageId: string, inputs?: readonly ReviewCollectInputV
142
142
  } as unknown as ReviewStatusV3;
143
143
  }
144
144
 
145
+ function providerRefuterRequiredStatus(lineageId: string): ReviewStatusV3 {
146
+ const refuter: ReviewCollectInputV3 = {
147
+ name: "provider_refuter",
148
+ schema: "https://gentle-ai.dev/schema/review/refuter/v1",
149
+ captureOperation: "review.capture-refuter",
150
+ arguments: bindingArguments(lineageId, "review-risk", 0),
151
+ };
152
+ return {
153
+ ...finalizeStatus(lineageId),
154
+ nextTransition: { kind: "collect", reasonCode: "provider_refuter_required", collect: { inputs: [refuter] } },
155
+ };
156
+ }
157
+
145
158
  interface RoutingHarness {
146
159
  statusQueue: ReviewStatusV3[];
147
- statusCalls: Array<{ cwd: string; lineageId?: string }>;
160
+ statusCalls: Array<{ cwd: string; lineageId?: string; agent?: "pi" }>;
148
161
  native: NativeReviewCli;
149
162
  }
150
163
 
@@ -156,7 +169,7 @@ function nativeHarness(statuses: readonly ReviewStatusV3[]): RoutingHarness {
156
169
  };
157
170
  harness.native = {
158
171
  targetStatus: async (request) => {
159
- harness.statusCalls.push({ cwd: request.cwd, ...(request.lineageId === undefined ? {} : { lineageId: request.lineageId }) });
172
+ harness.statusCalls.push({ cwd: request.cwd, ...(request.lineageId === undefined ? {} : { lineageId: request.lineageId }), ...(request.agent === undefined ? {} : { agent: request.agent }) });
160
173
  const next = harness.statusQueue.shift();
161
174
  if (next === undefined) throw new Error("status queue exhausted");
162
175
  return next;
@@ -226,7 +239,7 @@ function prepared(request: ReviewHostRelayRequest) { return { request, promptByt
226
239
  test("grouped capture forecasts once, reaches a four-reviewer barrier, and reconciles unclosed submissions", async (t) => {
227
240
  t.after(() => __testing.setReviewHostRelayGroupRunnersForTesting());
228
241
  const cwd = repository(t), lineageId = "relay-group", inputs = groupInputs(lineageId), lenses = inputs.map((input) => input.arguments.find((argument) => argument.name === "lens")!.value);
229
- const harness = nativeHarness([finalizeStatus(lineageId, inputs), finalizeStatus(lineageId, inputs), ...inputs.map((_input, index) => finalizeStatus(lineageId, index === 1 ? [...inputs.slice(index), { name: "unrelated", schema: "example", captureOperation: "external.example", arguments: [] }] : inputs.slice(index))), finalizeStatus(lineageId)]);
242
+ const harness = nativeHarness([finalizeStatus(lineageId, inputs), finalizeStatus(lineageId, inputs), ...inputs.map((_input, index) => finalizeStatus(lineageId, index === 1 ? [...inputs.slice(index), { name: "unrelated", schema: "example", captureOperation: "external.example", arguments: [] }] : inputs.slice(index))), providerRefuterRequiredStatus(lineageId)]);
230
243
  let ready!: () => void;
231
244
  const allStarted = new Promise<void>((resolve) => { ready = resolve; });
232
245
  const releases = new Map<string, () => void>(), completed: string[] = [], submitted: string[] = [];
@@ -249,7 +262,9 @@ test("grouped capture forecasts once, reaches a four-reviewer barrier, and recon
249
262
  assert.deepEqual(completed, [...lenses].reverse(), "reviewer completion order is not submission order");
250
263
  assert.deepEqual(submitted, lenses);
251
264
  assert.deepEqual({ outcome: result.outcome, statusCalls: harness.statusCalls.length, providerAction: result.provider_action }, { outcome: "native-reviewer-group-status-reconciled", statusCalls: 7, providerAction: "stop" });
252
- assert.equal(result.next_transition, undefined);
265
+ assert.equal(harness.statusCalls.at(-1)?.agent, "pi", "post-last-capture reconciliation preserves the Pi host runtime");
266
+ assert.equal((result.next_transition as { kind?: string; reasonCode?: string } | undefined)?.kind, "collect");
267
+ assert.equal((result.next_transition as { kind?: string; reasonCode?: string } | undefined)?.reasonCode, "provider_refuter_required");
253
268
  });
254
269
 
255
270
  test("captured v5 STATUS accepts capture-phase Pn when authority has advanced to Rn at the forecast boundary", async (t) => {
@@ -465,6 +480,7 @@ test("a submission whose outcome is genuinely indeterminate still reconciles thr
465
480
  assert.deepEqual(result.failure, { kind: "submission-refused", stage: "submit", exit_code: null, timed_out: true, elapsed_ms: 120_004, timeout_ms: 120_000 });
466
481
  assert.match(String(result.reason), /exceeded its 120000ms bound/);
467
482
  assert.equal(harness.statusCalls.length, 2, "an indeterminate submission reconciles exactly once through STATUS");
483
+ assert.equal(harness.statusCalls.at(-1)?.agent, "pi", "ordinary host-relay reconciliation preserves the Pi host runtime");
468
484
  });
469
485
 
470
486
  test("a relay timeout reports its one-slot measurements and no auto-follow", async (t) => {