@ferris1225/pi-subagents 4.0.1 → 4.1.2

package/src/fixloop.ts

@@ -1,16 +1,15 @@
 /**
- * Auto-fix loop: when a reviewer returns REVIEW_FAIL, the extension dispatches a
- * worker (briefed with the review's concrete findings) and then a reviewer
- * re-review, repeating up to maxFixRounds times before waking the main agent with
- * the full chain. The reviewer stays read-only and in its own context; the loop
- * is orchestrated by the extension layer, not by the reviewer itself, so the
- * independence guarantee (no self-confirmation bias) is preserved.
+ * Managed workflow policy and handoff formatting.
  *
- * The main agent is never woken mid-loop: the reviewer's FAIL result is intercepted
- * before delivery, the chain runs in the background, and only the final group
- * (initial review → worker fixes → re-reviews) is delivered at the end.
+ * Successful top-level writers can continue through documentation sync and an
+ * independent final review. A direct passing reviewer is also forced through
+ * documentation sync plus a fresh review when documenter is enabled. Any final
+ * gate failure may then use the established worker → optional documenter →
+ * reviewer fix rounds. Internal steps are launched by dispatch directly, so
+ * they never re-enter this top-level policy or wake the main agent mid-chain.
  */
 
+import { isWriteCapableAgent, type AgentConfig } from "./agents.ts";
 import { getResultOutput, isFailedResult, reviewVerdict, type SingleResult } from "./spawn.ts";
 import { extractKeyFragments, formatUsageCompact, sumUsage } from "./monitor.ts";
 import type { SubagentsConfig } from "./config.ts";
@@ -31,12 +30,92 @@ export function shouldTriggerFixLoop(result: SingleResult, config: SubagentsConf
 	// child happened to emit, which could end in a stray `VERDICT: REVIEW_FAIL`.
 	// Guard explicitly in addition to isFailedResult so the intent is clear and
 	// a future change to isFailedResult can never let a crashed reviewer start a
-	// phantom auto-fix chain (and re-add a run controller id that was deleted in
-	// the catch path).
+	// phantom auto-fix chain.
 	if (result.dispatchFailed) return false;
 	return reviewVerdict(getResultOutput(result)) === "fail";
 }
 
+export interface WorkflowAgentAvailability {
+	worker: boolean;
+	cleaner: boolean;
+	documenter: boolean;
+	reviewer: boolean;
+	writer: boolean;
+}
+
+export function workflowAgentAvailability(
+	agents: readonly Pick<AgentConfig, "name" | "tools">[],
+): WorkflowAgentAvailability {
+	const names = new Set(agents.map((agent) => agent.name));
+	return {
+		worker: names.has("worker"),
+		cleaner: names.has("cleaner"),
+		documenter: names.has("documenter"),
+		reviewer: names.has("reviewer"),
+		writer: agents.some(isWriteCapableAgent),
+	};
+}
+
+export type ManagedWorkflowKind = "auto-fix" | "post-writer" | "review-pass-sync";
+
+export interface ManagedWorkflowPlan {
+	kind: ManagedWorkflowKind;
+	initialRelation: string;
+}
+
+/** Conservative pre-run check used to reserve one shared-repository lane
+ * around a complete writer workflow or a reviewer that needs a stable diff.
+ * The actual result is classified again by getManagedWorkflowPlan before a
+ * downstream child starts. */
+export function canStartManagedWorkflow(
+	agent: Pick<AgentConfig, "name" | "tools">,
+	availability: WorkflowAgentAvailability,
+): boolean {
+	// Every shared write-capable role—including custom agents—owns the repository
+	// lane even when no downstream role is enabled. Otherwise its edits can race
+	// a managed writer's documentation snapshot.
+	if (isWriteCapableAgent(agent)) return true;
+	if (agent.name === "reviewer") {
+		// Hold a stable diff snapshot against every discoverable writer even when
+		// this review is advisory or maxFixRounds=0. Classification happens only
+		// after the read-only child returns, too late to acquire the lane safely.
+		return availability.writer;
+	}
+	return false;
+}
+
+/** Classify only healthy top-level results. In particular, a reviewer without a
+ * machine verdict is advisory and cannot start any write-capable child. */
+export function getManagedWorkflowPlan(
+	result: SingleResult,
+	config: SubagentsConfig,
+	availability: WorkflowAgentAvailability,
+): ManagedWorkflowPlan | undefined {
+	if (result.parked || result.dispatchFailed || isFailedResult(result)) return undefined;
+	if (result.agent === "worker" || result.agent === "cleaner") {
+		if (!availability.documenter && !availability.reviewer) return undefined;
+		return {
+			kind: "post-writer",
+			initialRelation: result.agent === "cleaner" ? "initial cleanup" : "initial implementation",
+		};
+	}
+	if (result.agent === "documenter") {
+		return availability.reviewer
+			? { kind: "post-writer", initialRelation: "documentation pass" }
+			: undefined;
+	}
+	if (result.agent !== "reviewer") return undefined;
+
+	const verdict = reviewVerdict(getResultOutput(result));
+	if (verdict === "pass" && availability.documenter && availability.reviewer) {
+		return { kind: "review-pass-sync", initialRelation: "pre-documentation review" };
+	}
+	if (verdict === "fail" && availability.worker && shouldTriggerFixLoop(result, config)) {
+		return { kind: "auto-fix", initialRelation: "initial review" };
+	}
+	return undefined;
+}
+
 /**
  * Build the worker task brief for one fix round from a reviewer's findings.
  * The worker gets the full review text so it can address concrete file:line
@@ -56,20 +135,83 @@ export function buildFixTaskBrief(reviewerResult: SingleResult, round: number, m
 		`Fix EVERY finding in the reviewer's findings list — there is no severity triage; all of them get fixed.`,
 		`If a finding is factually wrong or clearly out of scope, say so explicitly instead of fixing it.`,
 		`Do NOT refactor unrelated code beyond what the findings require.`,
+		`Do NOT commit, push, publish, tag, or release; do not bump versions. The parent chain still owns documentation sync and final review.`,
 		`After editing, run the project's format/build/tests when they exist and report`,
 		`exactly what you changed (paths + short rationale) so a reviewer can verify.`,
 		remaining > 0
 			? `A reviewer will re-review your changes automatically after you finish.`
-			: `This is the last auto-fix round; the main agent will be woken with the full chain.`,
+			: `This is the last auto-fix round; optional documentation sync and a fresh reviewer still run before final delivery.`,
 	].join("\n");
 }
 
+interface DocumentationBriefOptions {
+	title: string;
+	reports: Array<{ label: string; result: SingleResult }>;
+	closing: string;
+}
+
+function buildDocumentationBrief(options: DocumentationBriefOptions): string {
+	const reportSections = options.reports.flatMap(({ label, result }) => [
+		`${label}:`,
+		`---`,
+		getResultOutput(result),
+		`---`,
+		``,
+	]);
+	return [
+		options.title,
+		``,
+		...reportSections,
+		`Inspect the actual git diff (the complete pending diff) and relevant implementation; the report is only a lead.`,
+		`Synchronize stale README/docs, examples, API comments, docstrings, and explanatory comments with the behavior that will be committed.`,
+		`Change documentation surfaces only; never alter runtime behavior or tests to make prose true.`,
+		`Make zero edits when the diff creates no documentation drift.`,
+		`Do NOT commit, push, publish, tag, or release; do not bump versions. ${options.closing}`,
+		`Report exact documentation/comment paths changed, or state explicitly that no sync was needed.`,
+	].join("\n");
+}
+
+/** Build the pre-commit documentation handoff after one auto-fix worker. */
+export function buildDocumenterTaskBrief(
+	workerResult: SingleResult,
+	round: number,
+	reviewerResult?: SingleResult,
+): string {
+	return buildDocumentationBrief({
+		title: `Documentation sync after auto-fix round ${round}.`,
+		reports: [
+			...(reviewerResult ? [{ label: "The triggering reviewer reported", result: reviewerResult }] : []),
+			{ label: "The worker reported", result: workerResult },
+		],
+		closing: "a fresh reviewer gate runs after you.",
+	});
+}
+
+/** Build the automatic documentation stage after a successful top-level writer. */
+export function buildPostWriterDocumenterBrief(writerResult: SingleResult): string {
+	return buildDocumentationBrief({
+		title: `Documentation sync after successful top-level ${writerResult.agent}.`,
+		reports: [{ label: `The ${writerResult.agent} reported`, result: writerResult }],
+		closing: "the managed workflow owns any final reviewer and delivery.",
+	});
+}
+
+/** A direct passing review cannot be the final gate while documenter is enabled:
+ * the preliminary report is context, but the actual pending diff is authoritative. */
+export function buildReviewPassDocumenterBrief(reviewerResult: SingleResult): string {
+	return buildDocumentationBrief({
+		title: "Documentation sync required before accepting a direct passing review.",
+		reports: [{ label: "The preliminary reviewer reported", result: reviewerResult }],
+		closing: "the preliminary pass is not final and a fresh reviewer gate runs after you.",
+	});
+}
+
 /**
  * One step of an auto-fix chain as delivered: the run id (so the condensed
  * summary can point at per-run detail via subagent_status), the result, and
  * the human-readable role within the chain ("initial review", "fix round 1",
- * "re-review round 2"). runId is undefined for steps that never spawned a run
- * (e.g. an unknown agent).
+ * "re-review round 2"). runId is optional only for synthetic steps that never
+ * spawned a child.
  */
 export interface ChainStep {
 	runId?: number;
@@ -77,6 +219,11 @@ export interface ChainStep {
 	relation: string;
 }
 
+export interface ManagedWorkflowOutcome {
+	kind: ManagedWorkflowKind;
+	steps: ChainStep[];
+}
+
 /** Max distinguishing fragments kept in a one-line chain summary. */
 export const CHAIN_SUMMARY_FRAGMENTS_MAX = 3;
 
@@ -87,58 +234,124 @@ export function chainKeyFragments(result: SingleResult): string[] {
 	return extractKeyFragments(getResultOutput(result)).slice(0, CHAIN_SUMMARY_FRAGMENTS_MAX);
 }
 
-/**
- * Condensed, readable summary of a completed auto-fix chain: one line per step
- * (run id, role, verdict / what changed) plus aggregate usage. Full per-step
- * reports stay addressable via `subagent_status <id>`; the caller appends the
- * final step's full block only when its detail is actionable (FAIL verdict or a
- * crash), so the delivered message stays short instead of stacking every
- * round's raw output.
- */
-export function formatChainSummary(steps: readonly ChainStep[]): string {
-	const rounds = steps.filter((step) => step.relation.startsWith("fix round")).length;
-	const last = steps[steps.length - 1];
-	const stepStatus = (step: ChainStep): string => {
-		const { result } = step;
-		if (result.agent === "reviewer") {
-			const verdict = reviewVerdict(getResultOutput(result));
-			if (verdict) return verdict.toUpperCase();
-		}
-		return isFailedResult(result) ? "failed" : "completed";
-	};
-	const lines = [
-		`## Auto-fix chain: ${Math.max(1, rounds)} round${rounds === 1 ? "" : "s"} — final ${stepStatus(last)}`,
-		"",
-	];
-	for (const step of steps) {
-		const fragments = chainKeyFragments(step.result);
-		const suffix =
-			fragments.length > 0
-				? step.result.agent === "worker"
-					? ` — changed: ${fragments.join(" · ")}`
-					: ` — ${fragments.join(" · ")}`
-				: "";
-		const id = step.runId !== undefined ? `#${step.runId} ` : "";
-		lines.push(`- ${id}${step.result.agent} · ${step.relation} · ${stepStatus(step)}${suffix}`);
+function workflowResultStatus(result: SingleResult): string {
+	if (isFailedResult(result)) return "failed";
+	if (result.agent === "reviewer") {
+		const verdict = reviewVerdict(getResultOutput(result));
+		return verdict ? verdict.toUpperCase() : "NO_VERDICT";
 	}
+	return "completed";
+}
+
+function workflowStepLine(step: ChainStep): string {
+	const fragments = chainKeyFragments(step.result);
+	const writer = step.result.agent === "worker" || step.result.agent === "cleaner" || step.result.agent === "documenter";
+	const suffix = fragments.length > 0
+		? writer
+			? ` — changed: ${fragments.join(" · ")}`
+			: ` — ${fragments.join(" · ")}`
+		: "";
+	const id = step.runId !== undefined ? `#${step.runId} ` : "";
+	return `- ${id}${step.result.agent} · ${step.relation} · ${workflowResultStatus(step.result)}${suffix}`;
+}
+
+function appendWorkflowFooter(lines: string[], steps: readonly ChainStep[]): void {
 	const total = sumUsage(steps.map((step) => step.result.usage));
 	const usage = formatUsageCompact(total);
 	lines.push("", `Totals: ${steps.length} run${steps.length === 1 ? "" : "s"}${usage ? ` · ${usage}` : ""}`);
 	const ids = steps.filter((step) => step.runId !== undefined).map((step) => `#${step.runId}`);
 	lines.push(`Full per-run reports (output, usage, failed tools): subagent_status ${ids.join(" ")}`);
+}
+
+/** Condensed compatibility summary for a direct REVIEW_FAIL auto-fix chain. */
+export function formatChainSummary(
+	steps: readonly ChainStep[],
+	terminalResult: SingleResult = steps[steps.length - 1]!.result,
+): string {
+	const rounds = steps.filter((step) => step.relation.startsWith("fix round")).length;
+	const lines = [
+		`## Auto-fix chain: ${Math.max(1, rounds)} round${rounds === 1 ? "" : "s"} — final ${workflowResultStatus(terminalResult)}`,
+		"",
+		...steps.map(workflowStepLine),
+	];
+	appendWorkflowFooter(lines, steps);
+	return lines.join("\n");
+}
+
+/** One clear final delivery for all newly managed writer/documenter workflows. */
+export function formatManagedWorkflowSummary(
+	steps: readonly ChainStep[],
+	terminalResult: SingleResult = steps[steps.length - 1]!.result,
+): string {
+	const route = steps.map((step) => step.result.agent).join(" → ");
+	const fixRounds = steps.filter((step) => step.relation.startsWith("fix round")).length;
+	const roundNote = fixRounds > 0 ? ` · ${fixRounds} fix round${fixRounds === 1 ? "" : "s"}` : "";
+	const lines = [
+		`## Managed workflow: ${route}${roundNote} — final ${workflowResultStatus(terminalResult)}`,
+		"",
+		...steps.map(workflowStepLine),
+	];
+	appendWorkflowFooter(lines, steps);
 	return lines.join("\n");
 }
 
+/** Build the first independent final gate after a top-level writer or required
+ * post-pass documentation sync. Reports carry intent; the actual pending diff
+ * remains authoritative. */
+export function buildFinalReviewBrief(
+	initialResult: SingleResult,
+	documenterResult?: SingleResult,
+): string {
+	const documenterSection = documenterResult
+		? [
+			``,
+			`The documenter's full sync report:`,
+			`---`,
+			getResultOutput(documenterResult),
+			`---`,
+		]
+		: [];
+	return [
+		`Fresh final gate for a managed ${initialResult.agent} workflow.`,
+		``,
+		`The top-level ${initialResult.agent}'s full report:`,
+		`---`,
+		getResultOutput(initialResult),
+		`---`,
+		...documenterSection,
+		``,
+		`Run \`git status\` and \`git diff\` and inspect the actual pending code and documentation; reports are context, not proof.`,
+		`Remain read-only. Verify correctness, regressions, tests, documentation drift, and that documenter was the last writer when it ran.`,
+		`This is an acceptance gate, not an advisory audit. End with exactly one standalone machine verdict line:`,
+		`VERDICT: REVIEW_PASS when no finding remains, otherwise VERDICT: REVIEW_FAIL.`,
+	].join("\n");
+}
+
 /**
  * The re-review brief handed to the reviewer after a worker fix round. Includes
- * the prior review AND the worker's report so the reviewer can adjudicate
- * rejections instead of restating findings. The convergence contract keeps
- * rounds from ping-ponging: rule on the open findings once, add only defects
- * this round's edits introduced, never re-open a verified resolution.
+ * the prior review, worker report, and optional documenter report so the
+ * reviewer can adjudicate rejections instead of restating findings. The
+ * convergence contract keeps rounds from ping-ponging: rule on the open
+ * findings once, add only defects this round's edits introduced, never re-open
+ * a verified resolution.
  */
-export function buildReReviewBrief(reviewerResult: SingleResult, round: number, workerResult: SingleResult): string {
+export function buildReReviewBrief(
+	reviewerResult: SingleResult,
+	round: number,
+	workerResult: SingleResult,
+	documenterResult?: SingleResult,
+): string {
 	const review = getResultOutput(reviewerResult);
 	const workerReport = getResultOutput(workerResult);
+	const documenterSection = documenterResult
+		? [
+			``,
+			`The documenter's pre-commit sync report:`,
+			`---`,
+			getResultOutput(documenterResult),
+			`---`,
+		]
+		: [];
 	return [
 		`Re-review after auto-fix round ${round}.`,
 		``,
@@ -151,6 +364,7 @@ export function buildReReviewBrief(reviewerResult: SingleResult, round: number,
 		`---`,
 		workerReport,
 		`---`,
+		...documenterSection,
 		``,
 		`Rule on EVERY previous finding: resolved, or still open. A finding the worker rejected must be`,
 		`adjudicated ONCE — accept the rejection unless you can concretely refute the worker's reasoning;`,

package/src/index.ts

@@ -1,93 +1,93 @@
-/**
- * pi-subagents — focused sub-agent delegation for pi.
- *
- * Assembly point: builds the shared runtime and registers everything.
- * The heavy lifting lives in focused modules:
- *   - dispatch.ts         — the `subagent` tool contract and auto-fix chain
- *   - thread-lifecycle.ts — queued generations, resume/fork, isolation settlement
- *   - tools.ts            — subagent_control / subagent_wait / status / stop
- *   - announcements.ts — session-start recovery, notices, and widget install
- *   - widget.ts        — active-only TUI run status
- *   - runtime.ts       — shared per-session state
- *
- * Also registers the `/subagents-setup` command and a `before_agent_start` hook
- * that injects a delegation directive into the parent system prompt so the main
- * model uses the tool proactively.
- *
- * The tool is not registered inside child sub-agent processes, which prevents
- * runaway recursion and keeps child context windows clean.
- */
-
-import { getAgentDir, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
-import { Text } from "@earendil-works/pi-tui";
-import { discoverAgents } from "./agents.ts";
-import { registerAnnouncements } from "./announcements.ts";
-import { getConfigPath, loadConfig } from "./config.ts";
-import { registerSubagentTool } from "./dispatch.ts";
-import { matchRunIds } from "./format.ts";
-import { buildDelegationDirective } from "./prompt.ts";
-import { createRuntime } from "./runtime.ts";
-import { runSetup } from "./setup.ts";
-import { currentSubagentDepth } from "./spawn.ts";
-import { registerLookupTools } from "./tools.ts";
-import { clearActiveRunsWidget } from "./widget.ts";
-
-export { matchRunIds };
-
-export default function (pi: ExtensionAPI): void {
-	const configPath = getConfigPath(getAgentDir());
-	const runtime = createRuntime(pi, configPath);
-
-	// Recursion guard: sub-agent children are leaf processes. The `subagent` tool is
-	// excluded from their toolset at spawn (--exclude-tools); this check is defense
-	// in depth so a child can never expose the tool back to its model, even if
-	// another extension ignores the depth marker.
-	if (currentSubagentDepth() >= 1) {
-		pi.registerCommand("subagents-setup", {
-			description: "Configure pi-subagents (unavailable in nested sub-agent processes)",
-			handler: async (_args, ctx) => {
-				ctx.ui.notify("pi-subagents setup is unavailable in nested sub-agent processes.", "warning");
-			},
-		});
-		return;
-	}
-
-	pi.registerMessageRenderer("subagent-result", (message, _options, theme) =>
-		new Text(
-			`${theme.fg("toolTitle", theme.bold("subagent result"))}\n${message.content}`,
-			0,
-			0,
-		),
-	);
-
-	pi.on("session_shutdown", async (_event, ctx) => {
-		clearActiveRunsWidget(ctx);
-		await runtime.shutdown();
-	});
-
-	registerSubagentTool(pi, runtime);
-	registerLookupTools(pi, runtime);
-
-	pi.registerCommand("subagents-setup", {
-		description: "Configure pi-subagents: agents, selected models, capability-aware thinking, and runtime settings",
-		handler: async (_args, ctx) => {
-			await runSetup(ctx, configPath);
-		},
-	});
-
-	registerAnnouncements(pi, runtime);
-
-	// Proactive dispatch: inject the delegation directive into the parent system prompt.
-	pi.on("before_agent_start", async (event, ctx) => {
-		const config = await loadConfig(configPath);
-		if (!config.proactiveInjection) return undefined;
-		const { agents } = discoverAgents(ctx.cwd, {
-			scope: config.agentScope,
-			enabledNames: config.enabledAgents,
-			projectTrusted: ctx.isProjectTrusted?.() === true,
-		});
-		const directive = buildDelegationDirective(agents);
-		if (!directive) return undefined;
-		return { systemPrompt: `${event.systemPrompt}\n${directive}` };
-	});
-}
+/**
+ * pi-subagents — focused sub-agent delegation for pi.
+ *
+ * Assembly point: builds the shared runtime and registers everything.
+ * The heavy lifting lives in focused modules:
+ *   - dispatch.ts         — tool contract, managed role policy, internal steps
+ *   - thread-lifecycle.ts — stable generations, controls, final integration/delivery
+ *   - tools.ts            — subagent_control / subagent_wait / status / stop
+ *   - announcements.ts — session-start recovery, notices, and widget install
+ *   - widget.ts        — active-only TUI run status
+ *   - runtime.ts       — shared per-session state
+ *
+ * Also registers the `/subagents-setup` command and a `before_agent_start` hook
+ * that injects a delegation directive into the parent system prompt so the main
+ * model uses the tool proactively.
+ *
+ * The tool is not registered inside child sub-agent processes, which prevents
+ * runaway recursion and keeps child context windows clean.
+ */
+
+import { getAgentDir, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { Text } from "@earendil-works/pi-tui";
+import { discoverAgents } from "./agents.ts";
+import { registerAnnouncements } from "./announcements.ts";
+import { getConfigPath, loadConfig } from "./config.ts";
+import { registerSubagentTool } from "./dispatch.ts";
+import { matchRunIds } from "./format.ts";
+import { buildDelegationDirective } from "./prompt.ts";
+import { createRuntime } from "./runtime.ts";
+import { runSetup } from "./setup.ts";
+import { currentSubagentDepth } from "./spawn.ts";
+import { registerLookupTools } from "./tools.ts";
+import { clearActiveRunsWidget } from "./widget.ts";
+
+export { matchRunIds };
+
+export default function (pi: ExtensionAPI): void {
+	const configPath = getConfigPath(getAgentDir());
+	const runtime = createRuntime(pi, configPath);
+
+	// Recursion guard: sub-agent children are leaf processes. The `subagent` tool is
+	// excluded from their toolset at spawn (--exclude-tools); this check is defense
+	// in depth so a child can never expose the tool back to its model, even if
+	// another extension ignores the depth marker.
+	if (currentSubagentDepth() >= 1) {
+		pi.registerCommand("subagents-setup", {
+			description: "Configure pi-subagents (unavailable in nested sub-agent processes)",
+			handler: async (_args, ctx) => {
+				ctx.ui.notify("pi-subagents setup is unavailable in nested sub-agent processes.", "warning");
+			},
+		});
+		return;
+	}
+
+	pi.registerMessageRenderer("subagent-result", (message, _options, theme) =>
+		new Text(
+			`${theme.fg("toolTitle", theme.bold("subagent result"))}\n${message.content}`,
+			0,
+			0,
+		),
+	);
+
+	pi.on("session_shutdown", async (_event, ctx) => {
+		clearActiveRunsWidget(ctx);
+		await runtime.shutdown();
+	});
+
+	registerSubagentTool(pi, runtime);
+	registerLookupTools(pi, runtime);
+
+	pi.registerCommand("subagents-setup", {
+		description: "Configure pi-subagents: agents, selected models, capability-aware thinking, and runtime settings",
+		handler: async (_args, ctx) => {
+			await runSetup(ctx, configPath);
+		},
+	});
+
+	registerAnnouncements(pi, runtime);
+
+	// Proactive dispatch: inject the delegation directive into the parent system prompt.
+	pi.on("before_agent_start", async (event, ctx) => {
+		const config = await loadConfig(configPath);
+		if (!config.proactiveInjection) return undefined;
+		const { agents } = discoverAgents(ctx.cwd, {
+			scope: config.agentScope,
+			enabledNames: config.enabledAgents,
+			projectTrusted: ctx.isProjectTrusted?.() === true,
+		});
+		const directive = buildDelegationDirective(agents, { maxFixRounds: config.maxFixRounds });
+		if (!directive) return undefined;
+		return { systemPrompt: `${event.systemPrompt}\n${directive}` };
+	});
+}