gentle-pi 3.2.0 → 3.3.0

package/extensions/gentle-ai.ts

@@ -1,5 +1,6 @@
 import { consumeReviewMutation, pendingReviewMutation, recordReviewMutation } from "../lib/review-reminder-receipt.ts";
 import { resolveSessionWorktree } from "../lib/session-worktree-registry.ts";
+import { OddRuntimeDelegationGate } from "../lib/odd-runtime-delegation-gate.ts";
 import { resolveResearchCapabilities, renderResearchCapabilities } from "../lib/sdd-research-capabilities.ts";
 import { declareReviewRelayHandshake } from "../lib/review-relay-contract.ts";
 import { execFileSync } from "node:child_process";
@@ -116,6 +117,7 @@ import {
 	readOrchestratorSettings,
 	restoreOrchestratorSettings,
 	type OrchestratorSettingsReadResult,
+	parseOrchestratorModelRef,
 } from "../lib/profiles-orchestrator.ts";
 import { measureAgentsViewLayout, type AgentsViewLayout } from "../lib/agents-view-layout.ts";
 import { NativeChoiceList } from "../lib/native-choice-list.ts";
@@ -132,6 +134,7 @@ import {
 	REVIEW_HOST_RELAY_SUBMISSION_MISSING_MESSAGE,
 	REVIEW_HOST_RELAY_UNAVAILABLE_MESSAGE,
 	ReviewHostRelayError,
+	reviewHostMediatedRoleSlots,
 	reviewHostRelaySlots,
 	reviewHostRelayUnachievableDetail,
 	reviewHostRelayUnachievableReason,
@@ -146,6 +149,7 @@ import {
 	type ReviewHostRelaySlot,
 	type ReviewProviderRoleVectorSlot,
 } from "../lib/review-host-relay.ts";
+import type { InProcessReviewerRegistry } from "../lib/inprocess-reviewer.ts";
 import {
 	JOURNAL_STATUS,
 	REVIEW_OPERATION,
@@ -1274,7 +1278,7 @@ Organic Driven Development (ODD) is the predefined workflow of this orchestrator
 3. **Resolve uncertainty.** Recommend optional research only for a named uncertainty; ask one focused user question only for a real unresolved product decision, then stop and wait; use at most one scoped read-only assumption challenge for a high-consequence unproven premise.
 4. **Classify.** The work is substantial when exploration yields two or more meaningful implementation steps, or progress worth recovering after an interruption. Small, understood work stays small and creates no durable task artifacts.
 5. **Track before the first write.** For substantial authorized implementation, create \`odd/tasks/<feature-name>.md\` and its Engram mirror \`odd/<feature-name>/tasks\` automatically, then create or rebuild the visible \`todo\` list from the reconciled feature tasks, all before the first source write and without asking permission for tasks or storage. Tell the user in one line which feature document was created and how many tasks it holds.
-6. **Implement task by task.** Route each task through the orchestrator's Work Routing Ladder, with the configured TDD mode and applicable checks. Check an item off only after its outcome and checks were observed; update the file, mirror, and visible \`todo\` projection after every task transition and material plan change. Every task closes with at least one work-unit commit on the feature branch, branch first when on the default branch, with tests and docs alongside the behavior, using a Conventional Commit message; record the commit identity in the feature document as evidence. Work-unit commits on the feature branch are part of authorized substantial ODD implementation; push, pull request creation, and merge remain the user's decisions.
+6. **Implement task by task.** Route each task through the orchestrator's Work Routing Ladder, honoring its mandatory delegation triggers, with the configured TDD mode and applicable checks. These triggers are mandatory, not advisory: executing past a fired trigger inline is a routing defect even if the work succeeds. Check an item off only after its outcome and checks were observed; update the file, mirror, and visible \`todo\` projection after every task transition and material plan change. Every task closes with at least one work-unit commit on the feature branch, branch first when on the default branch, with tests and docs alongside the behavior, using a Conventional Commit message; record the commit identity in the feature document as evidence. Work-unit commits on the feature branch are part of authorized substantial ODD implementation; push, pull request creation, and merge remain the user's decisions.
 7. **Close.** Report the verified outcome, every failed, skipped, or pending check, and the next step. The native review candidate is a work-unit commit or a PR slice, never a TODO checkbox and never the accumulated feature branch; native review runs only under the user-owned RDD switch.
 Resume an interrupted feature with \`mem_context\`, then project- and feature-scoped \`mem_search\`, then \`mem_get_observation\` for the full document, then the task file itself; reconcile before continuing the next unfinished task. Detail for steps 3–7: \`orchestrator-delegation.md\` and \`orchestrator-memory.md\`.
 
@@ -4136,6 +4140,39 @@ function reportProfilesDrops(ctx: ExtensionContext, path: string, drops: Profile
 	}
 }
 
+/** Pi's own live-session controls: the ExtensionAPI's setModel/setThinkingLevel. */
+type LiveSession = Pick<ExtensionAPI, "setModel" | "setThinkingLevel">;
+
+/**
+ * Switch the running session to the profile's orchestrator. `settings.json`
+ * is the default for new sessions only; Pi's `setModel`/`setThinkingLevel`
+ * are what move the live one. Failures never undo the persisted default: the
+ * profile is applied for the next session either way, and the note says what
+ * this session did. Nothing here may throw — settings.json is already written
+ * and the apply must finish reporting.
+ */
+async function switchLiveOrchestrator(ctx: ExtensionContext, live: LiveSession, entry: AgentRoutingEntry): Promise<string> {
+	const reference = parseOrchestratorModelRef(entry.model);
+	if (reference === undefined) return "";
+	const label = `${reference.provider}/${reference.model}`;
+	const model = ctx.modelRegistry.find(reference.provider, reference.model);
+	if (model === undefined) return `\n${label} is not in the model catalog; this session keeps its current model.`;
+	let switched = false;
+	try {
+		switched = await live.setModel(model);
+	} catch (error) {
+		return `\nThis session could not switch to ${label}: ${sanitizeTerminalText(error instanceof Error ? error.message : String(error))}.`;
+	}
+	if (!switched) return `\nno authentication is configured for ${reference.provider}; this session keeps its current model.`;
+	if (entry.thinking === undefined) return `\nThis session now runs on ${label}.`;
+	try {
+		live.setThinkingLevel(entry.thinking);
+	} catch (error) {
+		return `\nThis session now runs on ${label}, but its thinking level could not be set to ${entry.thinking}: ${sanitizeTerminalText(error instanceof Error ? error.message : String(error))}.`;
+	}
+	return `\nThis session now runs on ${label} · ${entry.thinking}.`;
+}
+
 function profileSnapshotFrom(
 	current: AgentModelConfig,
 	settings: OrchestratorSettingsReadResult,
@@ -4149,6 +4186,7 @@ function profileSnapshotFrom(
 
 async function runProfilesPanelAction(
 	ctx: ExtensionContext,
+	live: LiveSession,
 	path: string,
 	file: AgentProfilesFile,
 	result: Exclude<ProfilesPanelResult, { type: "close" }>,
@@ -4290,6 +4328,10 @@ async function runProfilesPanelAction(
 					orchestratorRollback = () => restoreOrchestratorSettings(settingsPath, previous);
 					orchestratorNote = `\nOrchestrator set to ${formatOrchestratorSelection(orchestratorEntry)} in ${sanitizeTerminalText(settingsPath)}.`;
 				}
+				// settings.json only governs future sessions. The session the user is
+				// sitting in keeps its model until told otherwise, which made a profile
+				// look applied while the orchestrator kept answering with the old one.
+				orchestratorNote += await switchLiveOrchestrator(ctx, live, orchestratorEntry);
 			}
 			// A pin that does not resolve changes nothing at launch, so the global apply
 			// above is what governs this repository. Saying so keeps a broken pin from
@@ -4536,7 +4578,7 @@ async function runProfilesPanelAction(
 	}
 }
 
-async function handleProfilesCommand(ctx: ExtensionContext): Promise<void> {
+async function handleProfilesCommand(ctx: ExtensionContext, live: LiveSession): Promise<void> {
 	const path = profilesFilePath(gentleAiConfigHome());
 	const read = readProfilesFileResult(path);
 	if (read.status === "invalid") {
@@ -4586,7 +4628,7 @@ async function handleProfilesCommand(ctx: ExtensionContext): Promise<void> {
 	);
 	while (result.type !== "close") {
 		selectedName = "name" in result ? result.name : undefined;
-		file = await runProfilesPanelAction(ctx, path, file, result);
+		file = await runProfilesPanelAction(ctx, live, path, file, result);
 		result = await showProfilesPanel(
 			ctx,
 			file,
@@ -5770,6 +5812,19 @@ function nativeStatusInputRejection(reason: string, field?: string): Record<stri
 	};
 }
 
+function nativeInspectInputRejection(reason: string, field?: string): Record<string, unknown> {
+	return {
+		operation: REVIEW_CONTROLLER_OPERATION.INSPECT,
+		status: "blocked",
+		outcome: "native-inspect-input-invalid",
+		reason,
+		...(field === undefined ? {} : { field }),
+		...baseRefRejectionHint(reason),
+		mutation_performed: false,
+		mutation_outcome: "none",
+	};
+}
+
 const PENDING_REVIEW_CONSENT_TTL_MS = 10 * 60 * 1000;
 const REVIEW_SESSION_PERMISSION_STATUS_KEY = "gentle-review-session-permission";
 const REVIEW_SESSION_PERMISSION_STATUS_TEXT = "reviews allowed for this session";
@@ -6575,6 +6630,27 @@ function reviewHostRelayFailureReport(error: ReviewHostRelayError): Record<strin
 		// refusal reason lives (gentle-pi#524); dropping it hid every admission
 		// refusal behind "submission-refused".
 		...(error.stderr.length === 0 ? {} : { stderr: error.stderr }),
+		// gentle-shell#1156: what the reviewer child's own event stream revealed.
+		...(error.reviewerEvidence === undefined ? {} : { reviewer: error.reviewerEvidence }),
+	};
+}
+
+// gentle-pi#311 P2 (superseding gentle-shell#1136 / #1158): the lens's
+// completion selection comes from its entry in the agent model routing
+// config, keyed by its routing key (`review-<lens>`). There is no extension
+// allowlist and no ambient default model: the in-process completion resolves
+// its provider through the live model registry the caller supplies, or it is
+// refused before materialize ever runs (validateReviewerSelectionConfiguration
+// in lib/review-host-relay.ts).
+function reviewHostRelaySelection(lens: string | undefined, config: AgentModelConfig): { selection?: string; thinking?: string; routingKey: string } {
+	const routingKey = lens === undefined || lens.length === 0 ? "review capture" : lens.startsWith("review-") ? lens : `review-${lens}`;
+	const entry = config[routingKey];
+	const model = typeof entry === "object" && entry !== null && typeof (entry as AgentRoutingEntry).model === "string" && (entry as AgentRoutingEntry).model!.length > 0 ? (entry as AgentRoutingEntry).model : undefined;
+	const thinking = typeof entry === "object" && entry !== null && typeof (entry as AgentRoutingEntry).thinking === "string" ? (entry as AgentRoutingEntry).thinking : undefined;
+	return {
+		...(model === undefined ? {} : { selection: model }),
+		...(thinking === undefined ? {} : { thinking }),
+		routingKey,
 	};
 }
 
@@ -6677,6 +6753,7 @@ async function executeReviewHostRelayCapture(
 	selections: Map<string, RetainedNativeStatusSelection>,
 	route: RetainedNativeCaptureRoute | undefined,
 	signal?: AbortSignal,
+	modelRegistry?: InProcessReviewerRegistry,
 ): Promise<Record<string, unknown>> {
 	try {
 		if (slot.submission === undefined) {
@@ -6686,12 +6763,23 @@ async function executeReviewHostRelayCapture(
 				REVIEW_HOST_RELAY_SUBMISSION_MISSING_MESSAGE,
 			);
 		}
-		const result = await activeReviewHostRelayRunner({
-			captureArgumentTokens: slot.captureArgumentTokens,
-			targetCwd: cwd,
-			submission: slot.submission,
-			...(signal === undefined ? {} : { signal }),
-		});
+		const result = await activeReviewHostRelayRunner((() => {
+			// gentle-pi#311 P2 / P3: the lens's (or, for a v9 host-mediated role
+			// slot, the fixed review-refuter/review-validator routing key's)
+			// user-owned completion selection rides the request alongside the
+			// live model registry; the relay validates and refuses a missing
+			// registry or a routing entry with no configured model typed before
+			// anything launches, never a fallback to a child.
+			const launch = reviewHostRelaySelection(slot.routingKey ?? slot.lens, readModelConfig(cwd));
+			return {
+				captureArgumentTokens: slot.captureArgumentTokens,
+				targetCwd: cwd,
+				submission: slot.submission,
+				...launch,
+				...(modelRegistry === undefined ? {} : { reviewerRegistry: modelRegistry }),
+				...(signal === undefined ? {} : { signal }),
+			};
+		})());
 		const closure = decodeRelayLastEventClosure(result.submission);
 		if (closure !== undefined) return mapAndClearLastEventClosure(closure, binding, selections, cwd);
 		return {
@@ -6704,6 +6792,7 @@ async function executeReviewHostRelayCapture(
 				...(slot.lens === undefined ? {} : { lens: slot.lens }),
 				...(slot.order === undefined ? {} : { order: slot.order }),
 				...(slot.subjectHash === undefined ? {} : { subject_hash: slot.subjectHash }),
+				...(slot.routingKey === undefined ? {} : { role: slot.name }),
 				prompt_bytes: result.promptByteLength,
 				result_bytes: result.resultByteLength,
 				submission: result.submission,
@@ -7285,6 +7374,11 @@ async function executeReviewCaptureOperation(
 	candidateViews: CandidateViewRegistry | null = new CandidateViewRegistry(),
 	retainedUntrackedSelections: Map<string, RetainedNativeStatusSelection> = new Map(),
 	requireRegisteredRoute = false,
+	// gentle-pi#311 P2: the live model registry a lens materialize capture
+	// resolves its in-process completion through. Appended last (rather than
+	// inserted) so every existing positional call site — none of which pass an
+	// eighth argument — keeps compiling unchanged.
+	modelRegistry?: InProcessReviewerRegistry,
 ): Promise<Record<string, unknown>> {
 	const parameters = parseReviewCaptureParameters(parametersValue);
 	if (nativeReviewCli === null || nativeReviewCli.targetStatus === undefined) {
@@ -7344,7 +7438,33 @@ async function executeReviewCaptureOperation(
 				mutation_outcome: "none",
 			};
 		}
-		return withCorrectionTarget(await executeReviewHostRelayCapture(hostRelaySlots[0]!, nativeReviewCli, cwd, selected.binding, retainedUntrackedSelections, route, signal));
+		return withCorrectionTarget(await executeReviewHostRelayCapture(hostRelaySlots[0]!, nativeReviewCli, cwd, selected.binding, retainedUntrackedSelections, route, signal, modelRegistry));
+	}
+
+	// gentle-pi#311 P3: gentle-ai's v9 contract renders the refuter and
+	// targeted-validator role captures host-mediated, exactly like the lens
+	// slot above — same relay machinery, only its fixed review-refuter /
+	// review-validator routing key (carried on the slot) and its own schema
+	// differ. An older gentle-ai's self-contained --execute=true vector still
+	// falls through to reviewProviderRoleVectorSlots below unchanged.
+	const hostMediatedRoleSlots = reviewHostMediatedRoleSlots([selected.input]);
+	if (hostMediatedRoleSlots.length === 1) {
+		if (parameters.correctionLines !== undefined) return captureBindingRejected("correctionLines is valid only for a correction-plan capture");
+		if (parameters.reviewerRunAcknowledged !== true) {
+			return {
+				tool: "gentle_review_capture",
+				status: "blocked",
+				outcome: "reviewer-model-run-forecast",
+				cost_forecast: {
+					transport: "pi_host_relay",
+					model_runs: 1,
+					roles: [hostMediatedRoleSlots[0]!.routingKey],
+				},
+				mutation_performed: false,
+				mutation_outcome: "none",
+			};
+		}
+		return withCorrectionTarget(await executeReviewHostRelayCapture(hostMediatedRoleSlots[0]!, nativeReviewCli, cwd, selected.binding, retainedUntrackedSelections, route, signal, modelRegistry));
 	}
 
 	if (selected.input.captureOperation === "review.capture-correction-plan") {
@@ -7423,6 +7543,8 @@ async function executeReviewCaptureGroupOperation(
 	candidateViews: CandidateViewRegistry | null = new CandidateViewRegistry(),
 	retainedUntrackedSelections: Map<string, RetainedNativeStatusSelection> = new Map(),
 	requireRegisteredRoute = false,
+	// gentle-pi#311 P2: see executeReviewCaptureOperation's matching parameter.
+	modelRegistry?: InProcessReviewerRegistry,
 ): Promise<Record<string, unknown>> {
 	const parameters = parseReviewCaptureGroupParameters(parametersValue);
 	if (nativeReviewCli === null || nativeReviewCli.targetStatus === undefined) return { ...captureGroupRejected("native target STATUS is unavailable"), outcome: "native-status-unsupported" };
@@ -7463,6 +7585,8 @@ async function executeReviewCaptureGroupOperation(
 		captureArgumentTokens: slot.captureArgumentTokens,
 		targetCwd: cwd,
 		submission: slot.submission!,
+		...reviewHostRelaySelection(slot.lens, readModelConfig(cwd)),
+		...(modelRegistry === undefined ? {} : { reviewerRegistry: modelRegistry }),
 		...(signal === undefined ? {} : { signal }),
 	}));
 	let prepared: readonly ReviewHostRelayPreparedResult[];
@@ -7585,6 +7709,30 @@ async function executeReviewControllerOperation(
 		parameters.operation === REVIEW_CONTROLLER_OPERATION.INSPECT &&
 		nativeReviewCli !== null
 	) {
+		const rawInspect = parameters.input === undefined
+			? undefined
+			: parseControllerJson(parameters.input, REVIEW_CONTROLLER_OPERATION.INSPECT);
+		const unknownField = rawInspect === undefined
+			? undefined
+			: Object.keys(rawInspect).find((field) => !["baseRef", "committedOnly"].includes(field));
+		if (unknownField !== undefined) return nativeInspectInputRejection("unknown-field", unknownField);
+		const baseRef = rawInspect?.baseRef;
+		if (baseRef !== undefined && !isCanonicalProcessString(baseRef)) return nativeInspectInputRejection("base-ref-invalid");
+		if (baseRef !== undefined && rawInspect?.committedOnly !== true) return nativeInspectInputRejection("committed-only-required");
+		if (rawInspect !== undefined && baseRef === undefined) return nativeInspectInputRejection("committed-only-invalid");
+		let canonicalBaseRef: string | undefined;
+		if (typeof baseRef === "string") {
+			try {
+				canonicalBaseRef = resolveCanonicalCandidateBase(defaultCwd, baseRef).commit;
+			} catch (error) {
+				if (error instanceof CandidateViewError && error.diagnostics !== undefined) return nativeOperationFailure(parameters.operation, Object.assign(error, { candidateViewPreNative: true }));
+				if (error instanceof CandidateViewError && (error.reason === "base-ref-ambiguous" || error.reason === "base-ref-unresolvable" || error.reason === "base-ref-moved")) return nativeInspectInputRejection(error.reason);
+				return nativeInspectInputRejection("base-ref-unresolvable");
+			}
+		}
+		const inspectSelector = canonicalBaseRef === undefined
+			? {}
+			: { baseRef: canonicalBaseRef, committedOnly: true as const };
 		// A new inspect supersedes every pre-lineage selection before its first
 		// STATUS attempt. A failed or changed-candidate inspect cannot leave an
 		// older selection available for a later START.
@@ -7595,6 +7743,7 @@ async function executeReviewControllerOperation(
 					nativeReviewCli,
 					{
 						cwd: defaultCwd,
+						...inspectSelector,
 						...(signal === undefined ? {} : { signal }),
 					},
 					retainedUntrackedSelections,
@@ -7670,6 +7819,7 @@ async function executeReviewControllerOperation(
 					nativeReviewCli,
 					{
 						cwd: defaultCwd,
+						...inspectSelector,
 						untrackedScope: parameters.untrackedScope,
 						expectedUntrackedInventory: inventory,
 						intendedUntracked: selected.intendedUntracked,
@@ -8583,6 +8733,11 @@ function createGentleAiExtensionForTesting(
 	const candidateViews = dependencies.candidateViews === undefined ? new CandidateViewRegistry() : dependencies.candidateViews;
 	const herdrLifecycle = createHerdrConfirmationLifecycle(pi.events);
 	const permissionEnvironment = dependencies.processEnv ?? process.env;
+	const oddDelegationGate = new OddRuntimeDelegationGate();
+	const oddSessionId = (ctx: ExtensionContext): string => {
+		try { return ctx.sessionManager.getSessionId(); }
+		catch { return ""; }
+	};
 
 	const setReviewSessionPermissionStatus = (context: ExtensionContext, active: boolean): void => {
 		try {
@@ -8704,6 +8859,7 @@ function createGentleAiExtensionForTesting(
 				candidateViews,
 				((sessionKey: PendingReviewConsentSessionKey) => processRetainedNativeStatusSelections.get(sessionKey) ?? processRetainedNativeStatusSelections.set(sessionKey, new Map()).get(sessionKey)!)(pendingReviewConsentSessionKey(ctx, pendingReviewConsentFallbackKey)),
 				true,
+				ctx.modelRegistry,
 			);
 			return { content: [{ type: "text", text: JSON.stringify(details) }], details };
 		},
@@ -8717,7 +8873,7 @@ function createGentleAiExtensionForTesting(
 		promptSnippet: "Use one exact current STATUS collectBinding for one ordinary native capture; call fresh STATUS before every additional capture.",
 		promptGuidelines: [
 			"Pass only lineageId, the JSON-serialized exact collectBinding from current STATUS, and the route-specific optional acknowledgement or correctionLines value. Never compose provider argument tokens, prompts, results, verdicts, or lens arrays.",
-			"A materialize reviewer slot first forecasts one model run; re-submit that same exact binding with reviewerRunAcknowledged: true to authorize one host relay. Correction-plan slots require correctionLines inside the provider-issued bounds, counted in diff lines (one replaced source line is one deletion plus one addition) — a different unit from the frozen logical correction budget. Refuter and validation vectors execute exactly once as provider-rendered.",
+			"A materialize reviewer slot first forecasts one model run; re-submit that same exact binding with reviewerRunAcknowledged: true to authorize one host relay. Correction-plan slots require correctionLines inside the provider-issued bounds, counted in diff lines (one replaced source line is one deletion plus one addition) — a different unit from the frozen logical correction budget. A refuter or targeted-validator slot forecasts and runs the same way when the provider renders it host-mediated; an older provider's self-contained refuter/validation vector still executes exactly once as provider-rendered.",
 			"A native terminal closure or nonterminal capture returns directly. Do not expect automatic STATUS, FINALIZE, receipt, delivery, or another capture; call fresh STATUS before any next capture.",
 		],
 		parameters: REVIEW_CAPTURE_PARAMETERS,
@@ -8742,6 +8898,7 @@ function createGentleAiExtensionForTesting(
 				candidateViews,
 				((sessionKey: PendingReviewConsentSessionKey) => processRetainedNativeStatusSelections.get(sessionKey) ?? processRetainedNativeStatusSelections.set(sessionKey, new Map()).get(sessionKey)!)(pendingReviewConsentSessionKey(ctx, pendingReviewConsentFallbackKey)),
 				true,
+				ctx.modelRegistry,
 			);
 			return {
 				content: [{ type: "text", text: JSON.stringify(details) }],
@@ -8978,6 +9135,10 @@ function createGentleAiExtensionForTesting(
 		const retiredSync = readAgentStartNames(event).includes("sdd-sync") || /\bSDD sync executor\b/i.test(event.systemPrompt ?? "");
 		const isSddAgent = retiredSync || isSddAgentStartEvent(event);
 		const isNamedAgent = isNamedAgentStartEvent(event);
+		oddDelegationGate.start(
+			oddSessionId(ctx),
+			!isNamedAgent && !isSddAgent && permissionEnvironment.GENTLE_PI_AGENTS_CHILD !== "1",
+		);
 		const subagentDepthKey = pendingReviewConsentSessionKey(ctx, pendingReviewConsentFallbackKey);
 		if (isSddAgent || isNamedAgent) {
 			processAgentEndSubagentDepth.set(subagentDepthKey, (processAgentEndSubagentDepth.get(subagentDepthKey) ?? 0) + 1);
@@ -9087,6 +9248,7 @@ function createGentleAiExtensionForTesting(
 	// consent, or chooses a partial candidate. Durable own-mutation receipts
 	// gate STATUS and consume only the generation captured before that await.
 	pi.on("agent_end", async (_event, ctx) => {
+		oddDelegationGate.endChild(oddSessionId(ctx));
 		if (nativeReviewCli?.reviewMode === undefined || nativeReviewCli.targetStatus === undefined) return;
 		if (ctx.hasUI !== true || !reminderSessionActive) return;
 		const sessionKey = pendingReviewConsentSessionKey(ctx, pendingReviewConsentFallbackKey);
@@ -9122,6 +9284,7 @@ function createGentleAiExtensionForTesting(
 	pi.on("tool_result", (event, ctx) => {
 		if (!reminderSessionActive || event.isError !== false || (event.toolName !== "write" && event.toolName !== "edit")) return;
 		if (!isRecord(event.input) || typeof event.input.path !== "string" || !event.input.path.trim()) return;
+		oddDelegationGate.recordSuccess(oddSessionId(ctx), event.toolName, event.input, ctx.cwd);
 		try {
 			const root = resolveSessionWorktree(event.input.path, ctx.cwd)?.root;
 			if (root) recordReviewMutation(pi, ctx.sessionManager, root, { source: "direct", toolName: event.toolName, toolCallId: event.toolCallId });
@@ -9135,6 +9298,10 @@ function createGentleAiExtensionForTesting(
 			event.input,
 		);
 		if (sensitivePathDenied) return sensitivePathDenied;
+		const oddDelegationDenied = oddDelegationGate.beforeTool(
+			oddSessionId(ctx), event.toolName, event.input, ctx.cwd, readActiveToolNames(pi),
+		);
+		if (oddDelegationDenied) return oddDelegationDenied;
 		if (event.toolName === "subagent_run") {
 			const sddAgent = sddDispatchAgentName(event.input);
 			if (sddAgent === "invalid") {
@@ -9300,7 +9467,7 @@ function createGentleAiExtensionForTesting(
 	pi.registerCommand("gentle:profiles", {
 		description: "Create, switch, and manage global agent-model profiles for el Gentleman.",
 		handler: async (_args, ctx) => {
-			await handleProfilesCommand(ctx);
+			await handleProfilesCommand(ctx, pi);
 		},
 	});