pulseed 0.0.1 → 0.1.0

package/dist/goal/goal-tree-manager.d.ts

@@ -0,0 +1,72 @@
+import type { Logger } from "../runtime/logger.js";
+import type { StateManager } from "../state-manager.js";
+import type { ILLMClient } from "../llm/llm-client.js";
+import type { IPromptGateway } from "../prompt/gateway.js";
+import type { EthicsGate } from "../traits/ethics-gate.js";
+import type { GoalDependencyGraph } from "./goal-dependency-graph.js";
+import type { GoalDecompositionConfig, DecompositionResult, GoalTreeState } from "../types/goal-tree.js";
+/**
+ * GoalTreeManager handles recursive goal decomposition and tree state queries.
+ *
+ * Responsibilities:
+ *   - Specificity evaluation (LLM)
+ *   - N-layer recursive decomposition
+ *   - Decomposition validation (coverage + cycle check)
+ *   - Tree state queries
+ */
+export interface GoalTreeManagerOptions {
+    concretenesThreshold?: number;
+    maxDepth?: number;
+    logger?: Logger;
+}
+export declare class GoalTreeManager {
+    private readonly stateManager;
+    private readonly llmClient;
+    private readonly ethicsGate;
+    private readonly goalDependencyGraph;
+    private readonly promptGateway?;
+    private readonly concretenesThreshold;
+    private readonly maxDepth;
+    private readonly logger?;
+    constructor(stateManager: StateManager, llmClient: ILLMClient, ethicsGate: EthicsGate, goalDependencyGraph: GoalDependencyGraph, options?: GoalTreeManagerOptions, promptGateway?: IPromptGateway | undefined);
+    private scoreConcreteness;
+    /**
+     * Evaluates the specificity of a goal using an LLM.
+     * Returns a score between 0 (very abstract) and 1 (very concrete).
+     * Falls back to 0.5 on parse failures.
+     */
+    private evaluateSpecificity;
+    /**
+     * Recursively decomposes a goal into subgoals until each subgoal either:
+     *   (a) has specificity_score >= config.min_specificity -> leaf node
+     *   (b) has decomposition_depth >= config.max_depth -> forced leaf
+     *   (c) concreteness score >= concretenesThreshold (auto-stop)
+     *   (d) current depth >= maxDepth (depth guard)
+     *
+     * Options override instance-level defaults when provided.
+     * Returns a DecompositionResult for the top-level call.
+     *
+     * @deprecated For new goals, use {@link GoalRefiner.refine} instead.
+     * This method is used internally by GoalRefiner and remains callable for backward compatibility.
+     */
+    decomposeGoal(goalId: string, config: GoalDecompositionConfig, options?: {
+        concretenesThreshold?: number;
+        maxDepth?: number;
+    }): Promise<DecompositionResult>;
+    private _decomposeGoalInternal;
+    /**
+     * Validates a decomposition result by checking:
+     *   1. Coverage: subgoals cover all parent dimensions (LLM)
+     *   2. Cycle detection: no circular dependencies introduced
+     *
+     * Returns true only if both checks pass.
+     */
+    private validateDecomposition;
+    /**
+     * Computes the current GoalTreeState for the tree rooted at rootId.
+     * Traverses all descendants recursively.
+     */
+    getTreeState(rootId: string): Promise<GoalTreeState>;
+    private _collectAllDescendantIds;
+}
+//# sourceMappingURL=goal-tree-manager.d.ts.map

package/dist/goal/goal-tree-manager.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"goal-tree-manager.d.ts","sourceRoot":"","sources":["../../src/goal/goal-tree-manager.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,sBAAsB,CAAC;AACnD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACxD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAC3D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,0BAA0B,CAAC;AAC3D,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AAGtE,OAAO,KAAK,EACV,uBAAuB,EACvB,mBAAmB,EACnB,aAAa,EAEd,MAAM,uBAAuB,CAAC;AA2K/B;;;;;;;;GAQG;AACH,MAAM,WAAW,sBAAsB;IACrC,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,eAAe;IAMxB,OAAO,CAAC,QAAQ,CAAC,YAAY;IAC7B,OAAO,CAAC,QAAQ,CAAC,SAAS;IAC1B,OAAO,CAAC,QAAQ,CAAC,UAAU;IAC3B,OAAO,CAAC,QAAQ,CAAC,mBAAmB;IAEpC,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;IAVjC,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAgB;IACrD,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAS;gBAGd,YAAY,EAAE,YAAY,EAC1B,SAAS,EAAE,UAAU,EACrB,UAAU,EAAE,UAAU,EACtB,mBAAmB,EAAE,mBAAmB,EACzD,OAAO,CAAC,EAAE,sBAAsB,EACf,aAAa,CAAC,EAAE,cAAc,YAAA;YAUnC,iBAAiB;IAM/B;;;;OAIG;YACW,mBAAmB;IAoCjC;;;;;;;;;;;;OAYG;IACG,aAAa,CACjB,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,uBAAuB,EAC/B,OAAO,CAAC,EAAE;QAAE,oBAAoB,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;KAAE,GAC7D,OAAO,CAAC,mBAAmB,CAAC;YAmCjB,sBAAsB;IA6PpC;;;;;;OAMG;YACW,qBAAqB;IAuDnC;;;OAGG;IACG,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,aAAa,CAAC;YAqD5C,wBAAwB;CAUvC"}

package/dist/goal/goal-tree-manager.js

@@ -0,0 +1,568 @@
+import { randomUUID } from "node:crypto";
+import { z } from "zod";
+import { GoalSchema } from "../types/goal.js";
+import { scoreConcreteness as _scoreConcreteness } from "./goal-tree-quality.js";
+import { sanitizeThresholdTypes, sanitizeThresholdValues } from "./refiner-prompts.js";
+import { buildThreshold } from "./goal-validation.js";
+// ─── LLM Response Schemas ───
+const SpecificityResponseSchema = z.object({
+    specificity_score: z.number().min(0).max(1),
+    reasoning: z.string(),
+});
+const SubgoalItemSchema = z.object({
+    hypothesis: z.string().default(""),
+    dimensions: z
+        .array(z.object({
+        name: z.string(),
+        label: z.string(),
+        threshold_type: z.enum(["min", "max", "range", "present", "match"]),
+        threshold_value: z.union([z.number(), z.string(), z.boolean(), z.null()]).nullable(),
+        observation_method_hint: z.string().optional().default(""),
+    }))
+        .default([]),
+    constraints: z.array(z.string()).default([]),
+    expected_specificity: z.number().min(0).max(1).optional(),
+});
+const SubgoalsResponseSchema = z.array(SubgoalItemSchema);
+const CoverageResponseSchema = z.object({
+    covers_parent: z.boolean(),
+    missing_dimensions: z.array(z.string()).optional().default([]),
+    reasoning: z.string(),
+});
+// ─── Prompt Builders ───
+function buildSpecificityPrompt(goal) {
+    const dimNames = goal.dimensions.map((d) => d.name).join(", ");
+    const constraintLines = goal.constraints.length > 0
+        ? `\nConstraints: ${goal.constraints.join(", ")}`
+        : "";
+    return `Score how decomposable this goal is (0=high-level, 1=single atomic task).
+Treat 1.0 as a goal that can be executed as one task with no meaningful sub-components.
+Use lower scores when the goal clearly contains distinct aspects that should become separate subgoals.
+Root goals should still be scored for decomposability, not assumed to be leaves.
+
+Title: ${goal.title}
+Description: ${goal.description}
+Dimensions: ${dimNames || "(none)"}${constraintLines}
+Depth: ${goal.decomposition_depth}
+
+Return ONLY: {"specificity_score":<0.0-1.0>,"reasoning":"<brief>"}`;
+}
+function buildSubgoalPrompt(goal, depth, maxDepth, maxChildren) {
+    const constraintLines = goal.constraints.length > 0
+        ? `Constraints: ${goal.constraints.join("; ")}`
+        : "";
+    const dimLines = goal.dimensions.length > 0
+        ? `Dimensions: ${goal.dimensions.map((d) => `${d.name}(${d.label})`).join(", ")}`
+        : "";
+    return `Decompose into <=${maxChildren} concrete subgoals, each covering a distinct aspect.
+
+Parent: ${goal.title}
+Description: ${goal.description}
+${dimLines}
+${constraintLines}
+Depth: ${depth}/${maxDepth}
+
+Return a JSON array. Each item:
+- "hypothesis": string — what this subgoal achieves (1-2 sentences)
+- "dimensions": array of {name,label,threshold_type,threshold_value,observation_method_hint}
+  - threshold_type: "min"|"max"|"range"|"present"|"match" only
+- "constraints": string[]
+- "expected_specificity": 0.0-1.0
+
+Example (1 item):
+[{"hypothesis":"Improve test coverage for auth module","dimensions":[{"name":"coverage_pct","label":"Coverage %","threshold_type":"min","threshold_value":80,"observation_method_hint":"jest --coverage"}],"constraints":[],"expected_specificity":0.85}]
+
+Return ONLY the JSON array.`;
+}
+function buildCoveragePrompt(parent, children) {
+    const parentDims = parent.dimensions.map((d) => d.name).join(", ");
+    const childSummaries = children
+        .map((c, i) => `${i + 1}. "${c.title}": [${c.dimensions.map((d) => d.name).join(",")}]`)
+        .join("\n");
+    return `Do these subgoals cover all parent dimensions?
+
+Parent: ${parent.title}
+Dimensions: ${parentDims || "(none)"}
+Subgoals:
+${childSummaries}
+
+Return ONLY: {"covers_parent":<true|false>,"missing_dimensions":[],"reasoning":"<brief>"}`;
+}
+// ─── Helper: Build a Goal from subgoal spec ───
+function buildGoalFromSubgoalSpec(spec, parentId, parentDepth, now) {
+    const id = randomUUID();
+    const dims = spec.dimensions.map((d) => ({
+        name: d.name,
+        label: d.label,
+        current_value: null,
+        threshold: buildThreshold(d.threshold_type, d.threshold_value),
+        confidence: 0.5,
+        observation_method: {
+            type: "manual",
+            source: "decomposition",
+            schedule: null,
+            endpoint: null,
+            confidence_tier: "self_report",
+        },
+        last_updated: now,
+        history: [],
+        weight: 1.0,
+        uncertainty_weight: null,
+        state_integrity: "ok",
+        dimension_mapping: null,
+    }));
+    return GoalSchema.parse({
+        id,
+        parent_id: parentId,
+        node_type: "subgoal",
+        title: spec.hypothesis.slice(0, 200),
+        description: spec.hypothesis,
+        status: "active",
+        dimensions: dims,
+        gap_aggregation: "max",
+        dimension_mapping: null,
+        constraints: spec.constraints,
+        children_ids: [],
+        target_date: null,
+        origin: "decomposition",
+        pace_snapshot: null,
+        deadline: null,
+        confidence_flag: null,
+        user_override: false,
+        feasibility_note: null,
+        uncertainty_weight: 1.0,
+        decomposition_depth: parentDepth + 1,
+        specificity_score: spec.expected_specificity ?? null,
+        loop_status: "idle",
+        created_at: now,
+        updated_at: now,
+    });
+}
+export class GoalTreeManager {
+    stateManager;
+    llmClient;
+    ethicsGate;
+    goalDependencyGraph;
+    promptGateway;
+    concretenesThreshold;
+    maxDepth;
+    logger;
+    constructor(stateManager, llmClient, ethicsGate, goalDependencyGraph, options, promptGateway) {
+        this.stateManager = stateManager;
+        this.llmClient = llmClient;
+        this.ethicsGate = ethicsGate;
+        this.goalDependencyGraph = goalDependencyGraph;
+        this.promptGateway = promptGateway;
+        // null means concreteness auto-stop is disabled (backward compatible)
+        this.concretenesThreshold = options?.concretenesThreshold ?? null;
+        this.maxDepth = options?.maxDepth ?? 5;
+        this.logger = options?.logger;
+    }
+    // ─── Concreteness Scoring (private) ───
+    async scoreConcreteness(description) {
+        return _scoreConcreteness(description, { llmClient: this.llmClient });
+    }
+    // ─── Specificity Evaluation ───
+    /**
+     * Evaluates the specificity of a goal using an LLM.
+     * Returns a score between 0 (very abstract) and 1 (very concrete).
+     * Falls back to 0.5 on parse failures.
+     */
+    async evaluateSpecificity(goal) {
+        const prompt = buildSpecificityPrompt(goal);
+        try {
+            let parsed;
+            if (this.promptGateway) {
+                parsed = await this.promptGateway.execute({
+                    purpose: "goal_specificity_evaluation",
+                    goalId: goal.id,
+                    responseSchema: SpecificityResponseSchema,
+                    additionalContext: {
+                        prompt,
+                        goalTitle: goal.title,
+                        goalDescription: goal.description,
+                    },
+                });
+            }
+            else {
+                const response = await this.llmClient.sendMessage([{ role: "user", content: prompt }], { temperature: 0 });
+                parsed = this.llmClient.parseJSON(response.content, SpecificityResponseSchema);
+            }
+            return { score: parsed.specificity_score, reasoning: parsed.reasoning };
+        }
+        catch {
+            // Conservative fallback: treat as needing decomposition
+            return { score: 0.5, reasoning: "LLM evaluation failed, defaulting to 0.5" };
+        }
+    }
+    // ─── Core Decomposition ───
+    /**
+     * Recursively decomposes a goal into subgoals until each subgoal either:
+     *   (a) has specificity_score >= config.min_specificity -> leaf node
+     *   (b) has decomposition_depth >= config.max_depth -> forced leaf
+     *   (c) concreteness score >= concretenesThreshold (auto-stop)
+     *   (d) current depth >= maxDepth (depth guard)
+     *
+     * Options override instance-level defaults when provided.
+     * Returns a DecompositionResult for the top-level call.
+     *
+     * @deprecated For new goals, use {@link GoalRefiner.refine} instead.
+     * This method is used internally by GoalRefiner and remains callable for backward compatibility.
+     */
+    async decomposeGoal(goalId, config, options) {
+        const goal = await this.stateManager.loadGoal(goalId);
+        if (!goal) {
+            throw new Error(`GoalTreeManager.decomposeGoal: goal "${goalId}" not found`);
+        }
+        const effectiveConcretenesThreshold = options?.concretenesThreshold ?? this.concretenesThreshold;
+        const effectiveMaxDepth = options?.maxDepth ?? this.maxDepth;
+        // Auto-stop: check concreteness before decomposing (only when threshold is explicitly set)
+        if (effectiveConcretenesThreshold !== null) {
+            const concretenessResult = await this.scoreConcreteness(goal.description);
+            if (concretenessResult.score >= effectiveConcretenesThreshold) {
+                const now = new Date().toISOString();
+                const leafGoal = {
+                    ...goal,
+                    node_type: "leaf",
+                    specificity_score: concretenessResult.score,
+                    updated_at: now,
+                };
+                await this.stateManager.saveGoal(leafGoal);
+                return {
+                    parent_id: goal.id,
+                    children: [],
+                    depth: goal.decomposition_depth,
+                    specificity_scores: { [goal.id]: concretenessResult.score },
+                    reasoning: `Auto-stop: concreteness score ${concretenessResult.score.toFixed(2)} >= threshold ${effectiveConcretenesThreshold}. ${concretenessResult.reason}`,
+                };
+            }
+        }
+        return this._decomposeGoalInternal(goal, config, 0, effectiveMaxDepth);
+    }
+    async _decomposeGoalInternal(goal, config, retryCount, depthLimit) {
+        const now = new Date().toISOString();
+        const effectiveMaxDepth = depthLimit ?? config.max_depth;
+        // Step 1: Evaluate specificity
+        const { score: specificityScore, reasoning } = await this.evaluateSpecificity(goal);
+        // Update goal with specificity score
+        const updatedGoal = {
+            ...goal,
+            specificity_score: specificityScore,
+            updated_at: now,
+        };
+        // Step 2: Determine if this is a leaf node
+        const isRootGoal = goal.decomposition_depth === 0;
+        const isLeaf = (!isRootGoal && specificityScore >= config.min_specificity) ||
+            goal.decomposition_depth >= effectiveMaxDepth;
+        if (isLeaf) {
+            // Mark as leaf node
+            const leafGoal = {
+                ...updatedGoal,
+                node_type: "leaf",
+                updated_at: now,
+            };
+            await this.stateManager.saveGoal(leafGoal);
+            return {
+                parent_id: goal.id,
+                children: [],
+                depth: goal.decomposition_depth,
+                specificity_scores: { [goal.id]: specificityScore },
+                reasoning: specificityScore >= config.min_specificity
+                    ? `Goal is specific enough (score=${specificityScore.toFixed(2)}): ${reasoning}`
+                    : `Max depth ${effectiveMaxDepth} reached, forced leaf`,
+            };
+        }
+        // Step 3: Generate subgoals via LLM
+        const maxChildren = 5;
+        const subgoalPrompt = buildSubgoalPrompt(updatedGoal, goal.decomposition_depth, effectiveMaxDepth, maxChildren);
+        let subgoalSpecs = [];
+        try {
+            if (this.promptGateway) {
+                const parsed = await this.promptGateway.execute({
+                    purpose: "goal_decomposition",
+                    goalId: goal.id,
+                    responseSchema: SubgoalsResponseSchema,
+                    additionalContext: {
+                        prompt: subgoalPrompt,
+                        parentGoalTitle: goal.title,
+                        parentGoalDescription: goal.description,
+                    },
+                });
+                subgoalSpecs = parsed.map((sg) => {
+                    let hypothesis = sg.hypothesis;
+                    if (!hypothesis) {
+                        const firstDimLabel = sg.dimensions?.[0]?.label;
+                        hypothesis = firstDimLabel ? firstDimLabel : goal.title;
+                    }
+                    return {
+                        ...sg,
+                        hypothesis,
+                        dimensions: (sg.dimensions ?? []).map((d) => ({
+                            ...d,
+                            observation_method_hint: d.observation_method_hint ?? "",
+                        })),
+                        constraints: sg.constraints ?? [],
+                    };
+                });
+                subgoalSpecs = subgoalSpecs.slice(0, maxChildren);
+            }
+            else {
+                const subgoalResponse = await this.llmClient.sendMessage([{ role: "user", content: subgoalPrompt }], { temperature: 0 });
+                // Sanitize threshold_type values before schema validation --
+                // LLMs sometimes return "exact", "scale", "qualitative" etc.
+                const rawContent = subgoalResponse.content;
+                let sanitized = sanitizeThresholdValues(sanitizeThresholdTypes(rawContent));
+                // Sanitize hypothesis field: LLMs may use "title", "description", "goal", etc.
+                let preprocessed;
+                try {
+                    preprocessed = JSON.parse(sanitized);
+                }
+                catch {
+                    preprocessed = null;
+                }
+                if (Array.isArray(preprocessed)) {
+                    for (const item of preprocessed) {
+                        if (item && typeof item === "object" && !("hypothesis" in item)) {
+                            this.logger?.warn(`[GoalTreeManager] Subgoal item missing hypothesis. Keys: ${Object.keys(item).join(", ")}`);
+                            const alt = item.title ??
+                                item.description ??
+                                item.goal ??
+                                item.objective ??
+                                item.name ??
+                                item.text ??
+                                item.summary ??
+                                item.label ??
+                                "Unnamed subgoal";
+                            item.hypothesis = String(alt);
+                        }
+                    }
+                    sanitized = JSON.stringify(preprocessed);
+                }
+                const parsed = this.llmClient.parseJSON(sanitized, SubgoalsResponseSchema);
+                subgoalSpecs = parsed.map((sg) => {
+                    // Derive hypothesis from dimensions or parent goal title when the LLM omitted it
+                    let hypothesis = sg.hypothesis;
+                    if (!hypothesis) {
+                        const firstDimLabel = sg.dimensions?.[0]?.label;
+                        hypothesis = firstDimLabel ? firstDimLabel : goal.title;
+                    }
+                    return {
+                        ...sg,
+                        hypothesis,
+                        dimensions: (sg.dimensions ?? []).map((d) => ({
+                            ...d,
+                            observation_method_hint: d.observation_method_hint ?? "",
+                        })),
+                        constraints: sg.constraints ?? [],
+                    };
+                });
+                // Clamp to max_children_per_node
+                subgoalSpecs = subgoalSpecs.slice(0, maxChildren);
+            }
+        }
+        catch (err) {
+            // If subgoal generation fails, treat as leaf -- but log the error for diagnostics
+            this.logger?.error(`[GoalTreeManager] Subgoal generation failed for "${goal.id}": ${err instanceof Error ? err.message : String(err)}`);
+            const leafGoal = {
+                ...updatedGoal,
+                node_type: "leaf",
+                updated_at: now,
+            };
+            await this.stateManager.saveGoal(leafGoal);
+            return {
+                parent_id: goal.id,
+                children: [],
+                depth: goal.decomposition_depth,
+                specificity_scores: { [goal.id]: specificityScore },
+                reasoning: `Subgoal generation failed: ${err instanceof Error ? err.message : String(err)}`,
+            };
+        }
+        // Handle empty decomposition
+        if (subgoalSpecs.length === 0) {
+            const leafGoal = {
+                ...updatedGoal,
+                node_type: "leaf",
+                updated_at: now,
+            };
+            await this.stateManager.saveGoal(leafGoal);
+            return {
+                parent_id: goal.id,
+                children: [],
+                depth: goal.decomposition_depth,
+                specificity_scores: { [goal.id]: specificityScore },
+                reasoning: "LLM returned empty subgoal list, treating as leaf",
+            };
+        }
+        // Step 4: Build child Goal objects
+        const childGoals = subgoalSpecs.map((spec) => buildGoalFromSubgoalSpec(spec, goal.id, goal.decomposition_depth, now));
+        // Step 5: Build the provisional decomposition result for validation
+        const provisionalResult = {
+            parent_id: goal.id,
+            children: childGoals,
+            depth: goal.decomposition_depth,
+            specificity_scores: { [goal.id]: specificityScore },
+            reasoning,
+        };
+        // Step 6: Validate decomposition (retry up to 2 times)
+        const isValid = await this.validateDecomposition(provisionalResult);
+        if (!isValid && retryCount < 2) {
+            // Retry decomposition
+            return this._decomposeGoalInternal(goal, config, retryCount + 1, depthLimit);
+        }
+        // Step 7: Save parent goal (updated specificity_score, node_type stays as-is for non-leaf)
+        await this.stateManager.saveGoal(updatedGoal);
+        // Step 8: Save each child goal and update parent's children_ids
+        const childIds = [];
+        for (const child of childGoals) {
+            await this.stateManager.saveGoal(child);
+            childIds.push(child.id);
+            // Parent-child relationships are tracked via children_ids on the parent goal.
+            // No need to register in GoalDependencyGraph; cycle detection uses detectCycle() directly.
+        }
+        // Update parent goal's children_ids
+        const parentWithChildren = {
+            ...updatedGoal,
+            children_ids: [...updatedGoal.children_ids, ...childIds],
+            updated_at: now,
+        };
+        await this.stateManager.saveGoal(parentWithChildren);
+        // Step 9: Collect specificity scores for result
+        const specificityScores = {
+            [goal.id]: specificityScore,
+        };
+        // Step 10: Recursively decompose each child
+        for (const child of childGoals) {
+            const childResult = await this._decomposeGoalInternal(child, config, 0, depthLimit);
+            // Merge child specificity scores
+            Object.assign(specificityScores, childResult.specificity_scores);
+            // Merge children into child's record
+            if (childResult.children.length > 0) {
+                const reloadedChild = await this.stateManager.loadGoal(child.id);
+                if (reloadedChild) {
+                    // child was saved with updated children_ids from recursive call
+                    void reloadedChild; // already persisted by recursive call
+                }
+            }
+        }
+        return {
+            parent_id: goal.id,
+            children: childGoals,
+            depth: goal.decomposition_depth,
+            specificity_scores: specificityScores,
+            reasoning,
+        };
+    }
+    // ─── Validation ───
+    /**
+     * Validates a decomposition result by checking:
+     *   1. Coverage: subgoals cover all parent dimensions (LLM)
+     *   2. Cycle detection: no circular dependencies introduced
+     *
+     * Returns true only if both checks pass.
+     */
+    async validateDecomposition(result) {
+        const parent = await this.stateManager.loadGoal(result.parent_id);
+        if (!parent)
+            return false;
+        const children = result.children;
+        // Check 1: Coverage validation via LLM
+        if (children.length > 0) {
+            const coveragePrompt = buildCoveragePrompt(parent, children);
+            try {
+                let coverage;
+                if (this.promptGateway) {
+                    const raw = await this.promptGateway.execute({
+                        purpose: "goal_coverage_validation",
+                        goalId: parent.id,
+                        responseSchema: CoverageResponseSchema,
+                        additionalContext: {
+                            prompt: coveragePrompt,
+                            parentGoalTitle: parent.title,
+                            childCount: String(children.length),
+                        },
+                    });
+                    coverage = { covers_parent: raw.covers_parent, missing_dimensions: raw.missing_dimensions ?? [], reasoning: raw.reasoning };
+                }
+                else {
+                    const coverageResponse = await this.llmClient.sendMessage([{ role: "user", content: coveragePrompt }], { temperature: 0 });
+                    const raw = this.llmClient.parseJSON(coverageResponse.content, CoverageResponseSchema);
+                    coverage = { covers_parent: raw.covers_parent, missing_dimensions: raw.missing_dimensions ?? [], reasoning: raw.reasoning };
+                }
+                if (!coverage.covers_parent) {
+                    return false;
+                }
+            }
+            catch {
+                // On parse failure, allow decomposition to proceed
+            }
+        }
+        // Check 2: Cycle detection
+        for (const child of children) {
+            const wouldCycle = this.goalDependencyGraph.detectCycle(result.parent_id, child.id);
+            if (wouldCycle) {
+                return false;
+            }
+        }
+        return true;
+    }
+    // ─── Tree State ───
+    /**
+     * Computes the current GoalTreeState for the tree rooted at rootId.
+     * Traverses all descendants recursively.
+     */
+    async getTreeState(rootId) {
+        const root = await this.stateManager.loadGoal(rootId);
+        if (!root) {
+            return {
+                root_id: rootId,
+                total_nodes: 0,
+                max_depth_reached: 0,
+                active_loops: [],
+                pruned_nodes: [],
+            };
+        }
+        let totalNodes = 0;
+        let maxDepthReached = 0;
+        const activeLoops = [];
+        const prunedNodes = [];
+        const visit = async (goal) => {
+            totalNodes++;
+            if (goal.decomposition_depth > maxDepthReached) {
+                maxDepthReached = goal.decomposition_depth;
+            }
+            if (goal.loop_status === "running") {
+                activeLoops.push(goal.id);
+            }
+            if (goal.status === "cancelled") {
+                prunedNodes.push(goal.id);
+            }
+            for (const childId of goal.children_ids) {
+                const child = await this.stateManager.loadGoal(childId);
+                if (child) {
+                    await visit(child);
+                }
+            }
+        };
+        await visit(root);
+        return {
+            root_id: rootId,
+            total_nodes: totalNodes,
+            max_depth_reached: maxDepthReached,
+            active_loops: activeLoops,
+            pruned_nodes: prunedNodes,
+        };
+    }
+    // ─── Private Helpers ───
+    async _collectAllDescendantIds(goalId) {
+        const goal = await this.stateManager.loadGoal(goalId);
+        if (!goal)
+            return [];
+        const result = [];
+        for (const childId of goal.children_ids) {
+            result.push(childId);
+            result.push(...await this._collectAllDescendantIds(childId));
+        }
+        return result;
+    }
+}
+//# sourceMappingURL=goal-tree-manager.js.map