loopforge 1.8.0 → 1.15.1

package/dist/engine.js

@@ -9,10 +9,78 @@ import { randomUUID } from "node:crypto";
 import { getPolicy } from "./policy.js";
 import { FSBackend, readLineageMd } from "./backends/fs.js";
 import { AgentStatus, Mode, makeAnalysis, makeExecutionEvidence, makeExecutionFeedback, makeLoopCompileRequest, makeLoopObjective, makeLoopRoundResult, makeSelfEvaluation, makeSessionState, makeTaskId, makeVaultConfig, SELF_EVAL_REGEX, } from "./protocol.js";
-import { scoreQuality } from "./builder.js";
 import { compileLoop } from "./loop-compiler.js";
 import { logEvent } from "./observability.js";
 // ═══════════════════════════════════════════════════════════════════════════
+// Shared helpers
+// ═══════════════════════════════════════════════════════════════════════════
+/** Parse ExecutionEvidence from a raw JSON object. Shared by buildSelfEvaluation
+ *  and invokeLoopCompile — both parse the same execution_evidence shape. */
+export function parseExecutionEvidence(raw) {
+    if (!raw || typeof raw !== "object")
+        return undefined;
+    const testResults = raw.test_results;
+    return makeExecutionEvidence({
+        files_changed: Array.isArray(raw.files_changed)
+            ? raw.files_changed.filter((v) => typeof v === "string")
+            : [],
+        test_results: testResults && typeof testResults.passed === "number"
+            ? {
+                passed: testResults.passed,
+                failed: testResults.failed ?? 0,
+                skipped: testResults.skipped ?? 0,
+            }
+            : null,
+        success_criteria_met: Array.isArray(raw.success_criteria_met)
+            ? raw.success_criteria_met.filter((v) => typeof v === "string")
+            : [],
+        success_criteria_remaining: Array.isArray(raw.success_criteria_remaining)
+            ? raw.success_criteria_remaining.filter((v) => typeof v === "string")
+            : [],
+        progress_estimate: typeof raw.progress_estimate === "number"
+            ? Math.max(0, Math.min(1, raw.progress_estimate))
+            : 0.0,
+    });
+}
+/** Parse CriterionRevision[] from a raw JSON array. Shared by buildSelfEvaluation
+ *  and invokeLoopCompile — both parse the same revised_success_criteria shape. */
+export function parseCriterionRevisions(raw) {
+    if (!Array.isArray(raw))
+        return [];
+    return raw
+        .filter((v) => typeof v === "object" && v !== null &&
+        typeof v.old === "string" &&
+        typeof v.new === "string")
+        .map((v) => {
+        const r = v;
+        return { old: r.old, new: r.new };
+    });
+}
+/** Parse WorkerResult[] from a raw JSON array. Shared by buildSelfEvaluation
+ *  and invokeLoopCompile — both parse the same worker_results shape. */
+export function parseWorkerResults(raw) {
+    if (!Array.isArray(raw))
+        return [];
+    return raw
+        .filter((v) => typeof v === "object" && v !== null &&
+        typeof v.agentId === "string" &&
+        typeof v.subTask === "string" &&
+        typeof v.resultSummary === "string")
+        .map((v) => {
+        const w = v;
+        return {
+            agentId: w.agentId,
+            subAgentType: typeof w.subAgentType === "string" ? w.subAgentType : "general-purpose",
+            subTask: w.subTask,
+            resultSummary: w.resultSummary,
+            success: typeof w.success === "boolean" ? w.success : false,
+            discoveredConstraints: Array.isArray(w.discoveredConstraints)
+                ? w.discoveredConstraints.filter((c) => typeof c === "string")
+                : [],
+        };
+    });
+}
+// ═══════════════════════════════════════════════════════════════════════════
 // Self-Evaluation extraction (v1.1 — autonomous loop feedback)
 // ═══════════════════════════════════════════════════════════════════════════
 /** Extract a structured SelfEvaluation from agent output text.
@@ -45,70 +113,17 @@ export function extractSelfEvaluation(text) {
  *  (structured evaluation parameter path). */
 export function buildSelfEvaluation(raw) {
     // P4: Parse execution evidence from raw JSON
-    let executionEvidence;
-    const rawEvidence = raw.execution_evidence;
-    if (rawEvidence && typeof rawEvidence === "object") {
-        const testResults = rawEvidence.test_results;
-        executionEvidence = makeExecutionEvidence({
-            files_changed: Array.isArray(rawEvidence.files_changed)
-                ? rawEvidence.files_changed.filter((v) => typeof v === "string")
-                : [],
-            test_results: testResults && typeof testResults.passed === "number"
-                ? {
-                    passed: testResults.passed,
-                    failed: testResults.failed ?? 0,
-                    skipped: testResults.skipped ?? 0,
-                }
-                : null,
-            success_criteria_met: Array.isArray(rawEvidence.success_criteria_met)
-                ? rawEvidence.success_criteria_met.filter((v) => typeof v === "string")
-                : [],
-            success_criteria_remaining: Array.isArray(rawEvidence.success_criteria_remaining)
-                ? rawEvidence.success_criteria_remaining.filter((v) => typeof v === "string")
-                : [],
-            progress_estimate: typeof rawEvidence.progress_estimate === "number"
-                ? Math.max(0, Math.min(1, rawEvidence.progress_estimate))
-                : 0.0,
-        });
-    }
+    const executionEvidence = parseExecutionEvidence(raw.execution_evidence);
     // P5: Parse corrections
     const retractedConstraints = Array.isArray(raw.retracted_constraints)
         ? raw.retracted_constraints.filter((v) => typeof v === "string")
         : [];
-    const revisedCriteria = Array.isArray(raw.revised_success_criteria)
-        ? raw.revised_success_criteria
-            .filter((v) => typeof v === "object" && v !== null &&
-            typeof v.old === "string" &&
-            typeof v.new === "string")
-            .map((v) => {
-            const r = v;
-            return { old: r.old, new: r.new };
-        })
-        : [];
+    const revisedCriteria = parseCriterionRevisions(raw.revised_success_criteria);
     const wrongAssumptions = Array.isArray(raw.wrong_assumptions)
         ? raw.wrong_assumptions.filter((v) => typeof v === "string")
         : [];
     // Multi-agent: Parse worker delegation results
-    const workerResults = Array.isArray(raw.worker_results)
-        ? raw.worker_results
-            .filter((v) => typeof v === "object" && v !== null &&
-            typeof v.agentId === "string" &&
-            typeof v.subTask === "string" &&
-            typeof v.resultSummary === "string")
-            .map((v) => {
-            const w = v;
-            return {
-                agentId: w.agentId,
-                subAgentType: typeof w.subAgentType === "string" ? w.subAgentType : "general-purpose",
-                subTask: w.subTask,
-                resultSummary: w.resultSummary,
-                success: typeof w.success === "boolean" ? w.success : false,
-                discoveredConstraints: Array.isArray(w.discoveredConstraints)
-                    ? w.discoveredConstraints.filter((c) => typeof c === "string")
-                    : [],
-            };
-        })
-        : [];
+    const workerResults = parseWorkerResults(raw.worker_results);
     return makeSelfEvaluation({
         success: typeof raw.success === "boolean" ? raw.success : false,
         output_summary: typeof raw.output_summary === "string" ? raw.output_summary : "",
@@ -134,6 +149,9 @@ export function buildSelfEvaluation(raw) {
         wrong_assumptions: wrongAssumptions,
         // Multi-agent: Worker delegation results
         worker_results: workerResults,
+        // v1.10: Checkpoint compression
+        compression_checkpoint: typeof raw.compression_checkpoint === "boolean" ? raw.compression_checkpoint : false,
+        checkpoint_label: typeof raw.checkpoint_label === "string" ? raw.checkpoint_label : "",
     });
 }
 /** Fallback heuristic when structured self-eval extraction fails.
@@ -173,15 +191,14 @@ function makeEngineMetrics() {
 // LoopForgeEngine
 // ═══════════════════════════════════════════════════════════════════════════
 export class LoopForgeEngine {
-    skillsDir;
     state = null;
     backend = null;
     metrics = null;
     feedbackWriteBuffer = [];
     lastTask = null;
-    seenConstraints = new Set();
-    constructor(skillsDir = "skills", backend) {
-        this.skillsDir = skillsDir;
+    // skillsDir parameter retained for API backward compatibility (v1.15).
+    // Previously stored as this.skillsDir but never read — now accepted & dropped.
+    constructor(_skillsDir = "skills", backend) {
         if (backend)
             this.backend = backend;
     }
@@ -244,12 +261,12 @@ export class LoopForgeEngine {
                     is_active: true,
                     timestamp: now,
                     user_intent: String(signal.task_type ?? "").slice(0, 200),
-                    quality_score: signal.quality_score ?? 0,
+                    success: signal.success ?? false,
                     execution_feedback: JSON.stringify({
-                        status: (signal.quality_score ?? 0) >= 3
+                        success: signal.success ?? false,
+                        status: signal.success
                             ? "success"
                             : "partial",
-                        quality_score: signal.quality_score ?? 0,
                         constraint_compliance: {
                             all_hard_constraints_met: !Array.isArray(signal.violations) || signal.violations.length === 0,
                             violations: signal.violations ?? [],
@@ -299,7 +316,6 @@ export class LoopForgeEngine {
             goal_id: response.goal_id,
             goal_text_hash: response.goal_text_hash,
             recompile_level: response.recompile_level,
-            quality_score: 0,
             constraints_active: response.constraints_active,
             task: request.task,
             success: true,
@@ -319,7 +335,6 @@ export class LoopForgeEngine {
             timestamp: new Date().toISOString().replace(/\.\d+Z$/, ""),
             user_intent: `loop_compile round ${response.round} — ${response.goal_id}`,
             task_type: "loop_lineage",
-            quality_score: 0,
             skill_used: response.technique_used,
             technique_used: response.technique_used,
             loop_id: response.loop_id,
@@ -352,7 +367,6 @@ export class LoopForgeEngine {
                 technique_used: response.technique_used,
                 loop_objective: loopObjDict,
                 success: true,
-                quality_score: 0,
                 output_summary: lastOutputSummary,
                 constraint_violations: lastViolations,
             });
@@ -410,30 +424,29 @@ export class LoopForgeEngine {
     hydrateLoopContext(loopId) {
         const prefix = `loop:${loopId}:r`;
         const results = this.resolveBackend().queryEntries({ prefix });
-        // Merge feedback quality scores into lineage entries
+        // Merge feedback success flags into lineage entries
         const fbEntries = this.resolveBackend().queryEntries({
             prefix,
             feedbackOnly: true,
         });
-        const fbQuality = new Map();
+        const fbSuccess = new Map();
         for (const fe of fbEntries) {
             const tid = String(fe.task_id ?? "");
             const parts = tid.split(":r");
             if (parts.length >= 2) {
                 const roundStr = parts[1].split(":")[0];
                 const fbRound = parseInt(roundStr, 10);
-                const fbScore = fe.quality_score ?? 0;
-                if (!Number.isNaN(fbRound) && fbScore > 0) {
-                    fbQuality.set(fbRound, Math.max(fbQuality.get(fbRound) ?? 0, fbScore));
+                if (!Number.isNaN(fbRound) && fe.success !== undefined) {
+                    fbSuccess.set(fbRound, fe.success);
                 }
             }
         }
         for (const entry of results) {
             const lineage = (entry.loop_lineage ?? entry.lineage ?? {});
             const rnd = lineage.round;
-            if (rnd && fbQuality.has(rnd)) {
-                lineage.quality_score = fbQuality.get(rnd);
-                entry.quality_score = fbQuality.get(rnd);
+            if (rnd && fbSuccess.has(rnd)) {
+                lineage.success = fbSuccess.get(rnd);
+                entry.success = fbSuccess.get(rnd);
             }
         }
         // Fallback: scan Markdown files if JSON vault had no matches
@@ -456,10 +469,6 @@ export class LoopForgeEngine {
                 entry.constraint_violations =
                     lineage.constraint_violations ?? [];
             }
-            // v1.7: Ensure quality_score is always populated for failure lineage weighting
-            if (entry.quality_score === undefined || entry.quality_score === null) {
-                entry.quality_score = lineage.quality_score ?? 0;
-            }
         }
         // Enrich: attach full prompt text from Markdown for L0 cache reuse
         for (const entry of finalResults) {
@@ -499,11 +508,6 @@ export class LoopForgeEngine {
         const success = fb.success;
         const violations = fb.constraint_violations ?? [];
         const fixes = fb.manual_fixes_needed ?? "";
-        const quality = scoreQuality({
-            success,
-            constraint_violations: violations,
-            manual_fixes_needed: fixes,
-        });
         // Loop-aware task_id for feedback→lineage backfill
         const loopId = request.loop_id;
         const loopRound = request.round;
@@ -517,7 +521,7 @@ export class LoopForgeEngine {
         const signal = {
             task_id: taskId,
             task_type: request.task.slice(0, 80),
-            quality_score: quality,
+            success,
             skill_used: request.skill_name ?? "",
             violations,
             manual_fixes: fixes,
@@ -525,13 +529,13 @@ export class LoopForgeEngine {
             round: loopRound,
         };
         this.persistFeedbackToVault(signal);
-        // Flush immediately so next compile cycle sees quality scores
+        // Flush immediately so next compile cycle sees success flags
         this.flushFeedbackBuffer();
         // Update state
         this.state.call_count++;
-        this.state.quality_trend.push(quality);
-        if (this.state.quality_trend.length > 20) {
-            this.state.quality_trend = this.state.quality_trend.slice(-20);
+        this.state.success_trend.push(success);
+        if (this.state.success_trend.length > 20) {
+            this.state.success_trend = this.state.success_trend.slice(-20);
         }
         // Circuit breaker
         if (this.shouldBreak()) {
@@ -542,7 +546,7 @@ export class LoopForgeEngine {
         }
         logEvent("round_complete", {
             technique: "feedback",
-            quality,
+            success,
             loopId: loopId ?? "unknown",
             round: loopRound ?? this.state.call_count,
         });
@@ -550,10 +554,10 @@ export class LoopForgeEngine {
             status: AgentStatus.OK,
             response: {
                 status: AgentStatus.OK,
-                prompt: `## Feedback Recorded\n\nQuality Score: ${quality}/5\nSignals: 1`,
+                prompt: `## Feedback Recorded\n\nSuccess: ${success}\nSignals: 1`,
                 analysis: makeAnalysis({
                     technique: "feedback",
-                    rationale: `quality=${quality}`,
+                    rationale: `success=${success}`,
                     independence: "n/a",
                     cognitive_load: "low",
                 }),
@@ -569,7 +573,7 @@ export class LoopForgeEngine {
      *  P0–P2: Also persists discovered_constraints, objective_refinement,
      *  and emerged_subtasks for the compiler to consume next round.
      *  Call this BEFORE invokeLoopCompile for the next round so that
-     *  hydrateLoopContext picks up the latest quality scores. */
+     *  hydrateLoopContext picks up the latest success flags. */
     autoFeedback(selfEval, loopId, round, task) {
         this.ensureInit({ task, mode: Mode.FEEDBACK, vault_config: makeVaultConfig(), feedback: null, skill_name: null, task_id: null });
         const fb = makeExecutionFeedback({
@@ -578,11 +582,6 @@ export class LoopForgeEngine {
             constraint_violations: selfEval.constraint_violations,
             manual_fixes_needed: "",
         });
-        const quality = scoreQuality({
-            success: fb.success,
-            constraint_violations: fb.constraint_violations,
-            manual_fixes_needed: fb.manual_fixes_needed,
-        });
         const taskId = `loop:${loopId}:r${round}:feedback`;
         // Multi-agent: Merge sub-agent discovered constraints into the main constraint flow
         const subDiscovered = (selfEval.worker_results ?? [])
@@ -594,7 +593,7 @@ export class LoopForgeEngine {
         const signal = {
             task_id: taskId,
             task_type: task.slice(0, 80),
-            quality_score: quality,
+            success: fb.success,
             skill_used: "",
             violations: selfEval.constraint_violations,
             manual_fixes: "",
@@ -630,9 +629,9 @@ export class LoopForgeEngine {
         }
         // Update state
         this.state.call_count++;
-        this.state.quality_trend.push(quality);
-        if (this.state.quality_trend.length > 20) {
-            this.state.quality_trend = this.state.quality_trend.slice(-20);
+        this.state.success_trend.push(fb.success);
+        if (this.state.success_trend.length > 20) {
+            this.state.success_trend = this.state.success_trend.slice(-20);
         }
         // Circuit breaker
         if (this.shouldBreak()) {
@@ -644,10 +643,10 @@ export class LoopForgeEngine {
         logEvent("round_complete", {
             loopId,
             round,
-            quality,
+            success: fb.success,
             technique: this.state?.last_technique ?? "feedback",
         });
-        return quality;
+        return fb.success;
     }
     // ═══════════════════════════════════════════════════════════════════════
     // Loop Compile (public, primary)
@@ -670,53 +669,23 @@ export class LoopForgeEngine {
             force_level: extras.force_level ?? "auto",
             health_check_interval: extras.health_check_interval ?? 1,
             external_context: extras.external_context ?? "",
+            max_rounds: extras.max_rounds ?? undefined,
         });
         // Convert last_round_result if present
         const lastRR = extras.last_round_result;
         if (lastRR) {
             if (typeof lastRR === "object" && !Array.isArray(lastRR)) {
                 const rr = lastRR;
-                // Parse P4 execution evidence
-                let executionEvidence;
-                const rawEvidence = rr.execution_evidence;
-                if (rawEvidence && typeof rawEvidence === "object") {
-                    const testResults = rawEvidence.test_results;
-                    executionEvidence = makeExecutionEvidence({
-                        files_changed: Array.isArray(rawEvidence.files_changed)
-                            ? rawEvidence.files_changed.filter((v) => typeof v === "string")
-                            : [],
-                        test_results: testResults && typeof testResults.passed === "number"
-                            ? {
-                                passed: testResults.passed,
-                                failed: testResults.failed ?? 0,
-                                skipped: testResults.skipped ?? 0,
-                            }
-                            : null,
-                        success_criteria_met: Array.isArray(rawEvidence.success_criteria_met)
-                            ? rawEvidence.success_criteria_met.filter((v) => typeof v === "string")
-                            : [],
-                        success_criteria_remaining: Array.isArray(rawEvidence.success_criteria_remaining)
-                            ? rawEvidence.success_criteria_remaining.filter((v) => typeof v === "string")
-                            : [],
-                        progress_estimate: typeof rawEvidence.progress_estimate === "number"
-                            ? Math.max(0, Math.min(1, rawEvidence.progress_estimate))
-                            : 0.0,
-                    });
-                }
-                // Parse P5 revised_success_criteria
-                const revisedCriteria = Array.isArray(rr.revised_success_criteria)
-                    ? rr.revised_success_criteria
-                        .filter((v) => typeof v === "object" && v !== null &&
-                        typeof v.old === "string" && typeof v.new === "string")
-                        .map((v) => ({ old: v.old, new: v.new }))
-                    : [];
+                // Parse P4 execution evidence (shared helper)
+                const executionEvidence = parseExecutionEvidence(rr.execution_evidence);
+                // Parse P5 revised_success_criteria (shared parser)
+                const revisedCriteria = parseCriterionRevisions(rr.revised_success_criteria);
                 lcr.last_round_result = makeLoopRoundResult({
                     round: rr.round ?? 0,
                     success: rr.success ?? false,
                     output_summary: rr.output_summary ?? "",
                     constraint_violations: rr.constraint_violations ?? [],
                     manual_fixes_needed: rr.manual_fixes_needed ?? "",
-                    quality_score: rr.quality_score ?? 0,
                     // P0–P2: Cognitive evolution fields
                     discovered_constraints: Array.isArray(rr.discovered_constraints)
                         ? rr.discovered_constraints.filter((v) => typeof v === "string")
@@ -737,21 +706,11 @@ export class LoopForgeEngine {
                     wrong_assumptions: Array.isArray(rr.wrong_assumptions)
                         ? rr.wrong_assumptions.filter((v) => typeof v === "string")
                         : [],
-                    // Multi-agent: Worker delegation results
-                    worker_results: Array.isArray(rr.worker_results)
-                        ? rr.worker_results
-                            .filter((v) => typeof v?.agentId === "string" && typeof v?.subTask === "string")
-                            .map((v) => ({
-                            agentId: v.agentId,
-                            subAgentType: typeof v.subAgentType === "string" ? v.subAgentType : "general-purpose",
-                            subTask: v.subTask,
-                            resultSummary: typeof v.resultSummary === "string" ? v.resultSummary : "",
-                            success: typeof v.success === "boolean" ? v.success : false,
-                            discoveredConstraints: Array.isArray(v.discoveredConstraints)
-                                ? v.discoveredConstraints
-                                : [],
-                        }))
-                        : [],
+                    // v1.10: Checkpoint boundary
+                    compression_checkpoint: typeof rr.compression_checkpoint === "boolean" ? rr.compression_checkpoint : false,
+                    checkpoint_label: typeof rr.checkpoint_label === "string" ? rr.checkpoint_label : "",
+                    // Multi-agent: Worker delegation results (shared parser)
+                    worker_results: parseWorkerResults(rr.worker_results),
                 });
             }
         }
@@ -788,15 +747,10 @@ export class LoopForgeEngine {
                 },
             };
         }
-        // Build prompt text from response
-        const promptLines = [
-            `## LoopForge Loop Compile — Round ${response.round}`,
-            `**Recompile Level**: ${response.recompile_level.toUpperCase()}`,
-            `**Loop ID**: ${response.loop_id}`,
-            `**Goal ID**: ${response.goal_id}`,
-            "",
-            response.prompt,
-        ];
+        // Build prompt text from response.
+        // Compile functions (buildL2Prompt / specialist compilers via buildHeader)
+        // already produce their own round-level headers with technique + identity.
+        const promptLines = [response.prompt];
         if (response.warnings.length) {
             promptLines.push("");
             promptLines.push("### Warnings");
@@ -816,8 +770,8 @@ export class LoopForgeEngine {
             const hasError = verificationFlags.some((f) => f.severity === "error");
             if (hasError) {
                 promptLines.push("");
-                promptLines.push("**Gate Verdict: CONTRADICTED** — this round's quality score has been excluded " +
-                    "from the quality trend. Address each 🚫 flag explicitly in your next response.");
+                promptLines.push("**Gate Verdict: CONTRADICTED** — this round's success flag has been excluded " +
+                    "from the success trend. Address each 🚫 flag explicitly in your next response.");
             }
         }
         if (response.loop_health) {
@@ -865,6 +819,7 @@ export class LoopForgeEngine {
                     reference_file: response.reference_file,
                 }),
                 error: null,
+                state_file_content: response.state_file_content,
             },
         };
     }
@@ -876,17 +831,18 @@ export class LoopForgeEngine {
             return false;
         const policy = getPolicy();
         const maxCB = policy.engine.max_circuit_breaker;
-        if (this.state.quality_trend.length < maxCB)
+        if (this.state.success_trend.length < maxCB)
             return false;
-        const recent = this.state.quality_trend.slice(-maxCB);
-        const isFlat = recent.every((val, i) => i === 0 || recent[i - 1] >= val);
-        if (isFlat) {
+        const recent = this.state.success_trend.slice(-maxCB);
+        // Trip only when all recent rounds are failures (no false-positives on all-success)
+        const allFailed = recent.every((v) => v === false);
+        if (allFailed) {
             logEvent("circuit_breaker", {
                 trend: recent,
-                totalRounds: this.state.quality_trend.length,
+                totalRounds: this.state.success_trend.length,
             });
         }
-        return isFlat;
+        return allFailed;
     }
 }
 // ═══════════════════════════════════════════════════════════════════════════