@codemoot/core 0.2.6 → 0.2.9

package/dist/index.d.ts

@@ -1974,6 +1974,8 @@ interface ProgressCallbacks {
     onHeartbeat?: (elapsedSec: number) => void;
     /** Called on each stdout chunk for progress feedback. */
     onProgress?: (chunk: string) => void;
+    /** Called when the subprocess closes — flush any buffered state. */
+    onClose?: () => void;
 }
 interface CliCallOptions extends ProgressCallbacks {
     /** Total timeout in ms. Default: 600_000 (10 min). */
@@ -2179,6 +2181,7 @@ interface StepResult {
     durationMs: number;
     verdict?: 'approved' | 'needs_revision';
     feedback?: string;
+    score?: number;
 }
 declare class StepRunner {
     private registry;
@@ -2201,9 +2204,14 @@ interface LoopResult {
     iterations: number;
     approved: boolean;
     history: StepResult[];
+    finalScore?: number;
+}
+interface LoopOptions {
+    /** Score threshold for auto-approval on iteration >= 2. Only applies when set. */
+    autoApproveThreshold?: number;
 }
 declare class LoopController {
-    executeLoop(generateStep: ResolvedStep, reviewStep: ResolvedStep, inputs: Map<string, string>, task: string, maxIterations: number, runner: StepRunner, eventBus: EventBus): Promise<LoopResult>;
+    executeLoop(generateStep: ResolvedStep, reviewStep: ResolvedStep, inputs: Map<string, string>, task: string, maxIterations: number, runner: StepRunner, eventBus: EventBus, loopOptions?: LoopOptions): Promise<LoopResult>;
 }
 
 declare class CancellationToken {
@@ -2252,6 +2260,8 @@ interface PlanOptions {
     maxRounds?: number;
     stream?: boolean;
     workflowDir?: string;
+    /** Score threshold (1-10) for auto-approving plans on 2nd+ iteration. Default: 8. */
+    autoApproveThreshold?: number;
 }
 interface SessionResult {
     sessionId: string;
@@ -2669,6 +2679,6 @@ declare const DEFAULT_DLP_CONFIG: DlpConfig;
  */
 declare function sanitize(input: string, config?: Partial<DlpConfig>): DlpResult;
 
-declare const VERSION = "0.2.6";
+declare const VERSION = "0.2.8";
 
 export { type ArtifactRecord, ArtifactStore, type ArtifactType, type AssembledContext, type AttemptResult, BINARY_SNIFF_BYTES, type BridgeCapabilities, type BridgeOptions, type BridgeResumeOptions, type BudgetConfig, type BuildPhase, type BuildRun, BuildStore, type BuildSummary, type BuiltInRole, CLEANUP_TIMEOUT_SEC, CONTEXT_ACTIVE, CONTEXT_BUFFER, CONTEXT_RETRIEVED, CURRENT_VERSION, type CacheEntry, CacheStore, type CallModelOptions, CancellationError, CancellationToken, type ChatMessage, type CleanupConfidence, type CleanupFinding, type CleanupReport, type CleanupScope, type CleanupSource, CliAdapter, type CliAdapterConfig, type CliBridge, type CliCallOptions, type CliDetectionResult, ConfigError, type ContextBudget, ContextBuilder, type ContextBuilderOptions, type CostInput, type CostLogEntry, CostStore, type CostSummary, CostTracker, type CostUpdateEvent, DAYS_PER_YEAR, DEFAULT_CONFIG, DEFAULT_DLP_CONFIG, DEFAULT_MAX_TOKENS, DEFAULT_RULES, DEFAULT_TIMEOUT_SEC, DLP_MAX_CONTENT, DLP_MAX_PROCESSING_MS, DatabaseError, type DebateConfig, type DebateEngineState, type DebateInput, type DebateMessageRow, type DebateOptions, type DebatePattern, type DebateResponse, type DebateResult, DebateStore, type DebateTurnRow, type DebateTurnStatus, type DlpAuditEntry, type DlpConfig, type DlpMode, DlpReasonCode, type DlpRedaction, type DlpResult, type EngineEvent, type EnqueueOptions, ErrorCode, EventBus, type ExecutionMode, type FallbackConfig, HTTP_TOO_MANY_REQUESTS, type HandoffCommand, type HandoffEnvelopeOptions, type HostFindingInput, IMPLEMENTER_MAX_TOKENS, type JobLogRecord, type JobRecord, type JobStatus, JobStore, type JobType, type LogLevel, type Logger, type LoopConfig, LoopController, type LoopIterationEvent, type LoopResult, MCP_CONTENT_MAX_LENGTH, MCP_TASK_MAX_LENGTH, MCP_TIMEOUT_MAX, type MemoryCategory, type MemoryConfig, type MemoryInput, type MemoryRecord, MemoryStore, type MessageStatus, MessageStore, type MeteringSource, type ModelAdapter, type ModelCallResult, type ModelConfig, ModelError, type ModelPricing, type ModelProvider, ModelRegistry, Orchestrator, type OrchestratorOptions, type OutputConfig, type ParsedVerdict, type PlanInput, type PlanOptions, type PolicyContext, type PolicyDecision, type PolicyMode, type PolicyResult, type PolicyRule, type PresetName, type ProgressCallbacks, type ProjectConfig, type ProjectConfigInput, type PromptType, type PromptVariables, REVIEW_DIFF_MAX_CHARS, REVIEW_PRESETS, REVIEW_TEXT_MAX_CHARS, type ResolvedStep, type ResolvedWorkflow, type ResultStatus, type RetryConfig, type RetryOptions, type ReviewInput, type ReviewOptions, type ReviewPreset, type ReviewResult, type Role, type RoleConfig, RoleManager, type RunOptions, type Session, type SessionCompletedEvent, type SessionEvent, type SessionFailedEvent, SessionManager, type SessionOverflowStatus, type SessionResult, type SessionStartedEvent, type SessionStatus, SessionStore, type StepCompletedEvent, type StepDefinition, type StepFailedEvent, type StepResult, StepRunner, type StepStartedEvent, type StepType, TerminalReason, type TextDeltaEmitter, type TextDeltaEvent, type TextDoneEvent, type TokenUsage, type TranscriptEntry, type UnifiedSession, VERSION, type VerdictResult, type WorkflowDefinition, WorkflowEngine, WorkflowError, buildHandoffEnvelope, buildReconstructionPrompt, calculateCost, calculateDebateTokens, callModel, clearDetectionCache, computeThreeWayStats, computeTwoWayStats, costInputSchema, createIgnoreFilter, createLogger, debateInputSchema, debateOutputSchema, detectCli, estimateTokens, evaluatePolicy, generateId, generateSessionId, getModelPricing, getReviewPreset, getSchemaVersion, getTokenBudgetStatus, hashConfig, hashContent, hostFindingsSchema, isRateLimit, isRetryable, listPresetNames, listPresets, loadConfig, loadIgnorePatterns, loadPreset, memoryInputSchema, mergeThreeWay, mergeTwoWay, migrateConfig, openDatabase, parseDebateVerdict, parseVerdict, planInputSchema, preflightTokenCheck, projectConfigSchema, recalculateConfidenceStats, renderPrompt, reviewInputSchema, reviewOutputSchema, runAllScanners, runMigrations, sanitize, scanAntiPatterns, scanDeadCode, scanDuplicates, scanHardcoded, scanNearDuplicates, scanSecurity, scanUnusedDeps, scanUnusedExports, shouldIgnore, sleep, streamModel, validateConfig, withCanonicalRetry, withFallback, withRetry, writeConfig };

package/dist/index.js

@@ -2950,6 +2950,10 @@ var CliAdapter = class {
         if (settled) return;
         settled = true;
         cleanup();
+        try {
+          options?.onClose?.();
+        } catch {
+        }
         if (code !== 0) {
           reject(
             new ModelError(
@@ -3177,9 +3181,8 @@ var ModelRegistry = class _ModelRegistry {
   async healthCheckAll() {
     const results = /* @__PURE__ */ new Map();
     const detection = await detectCli("codex");
-    const healthy = detection.available && (detection.authOk ?? false);
     for (const alias of this.configs.keys()) {
-      results.set(alias, healthy);
+      results.set(alias, detection.available);
     }
     return results;
   }
@@ -3535,9 +3538,12 @@ ${vars.task}`
       "",
       "Provide specific, actionable feedback. Reference plan sections by name.",
       "",
-      "At the END of your review, you MUST include exactly one of these lines:",
+      "At the END of your review, you MUST include ALL of these lines:",
+      "SCORE: X/10",
       "VERDICT: APPROVED",
-      "VERDICT: NEEDS_REVISION"
+      "VERDICT: NEEDS_REVISION",
+      "",
+      "Where X is your numeric score (1-10) and VERDICT is exactly one of the two options."
     ].join("\n"),
     user: (vars) => `## Task:
 ${vars.task}
@@ -3605,9 +3611,9 @@ ${vars.plan ?? ""}`
       "",
       "Provide specific feedback with file names and references.",
       "",
-      "At the END of your review, you MUST include exactly one of:",
-      "VERDICT: APPROVED",
-      "VERDICT: NEEDS_REVISION"
+      "At the END of your review, you MUST include ALL of these lines:",
+      "SCORE: X/10",
+      "VERDICT: APPROVED or VERDICT: NEEDS_REVISION"
     ].join("\n"),
     user: (vars) => [
       `## Task:
@@ -3695,7 +3701,7 @@ var EventBus = class extends EventEmitter {
 
 // src/engine/loop-controller.ts
 var LoopController = class {
-  async executeLoop(generateStep, reviewStep, inputs, task, maxIterations, runner, eventBus) {
+  async executeLoop(generateStep, reviewStep, inputs, task, maxIterations, runner, eventBus, loopOptions) {
     const history = [];
     let currentInputs = new Map(inputs);
     let approved = false;
@@ -3706,7 +3712,12 @@ var LoopController = class {
       reviewInputs.set(generateStep.definition.id, generateResult.output);
       const reviewResult = await runner.execute(reviewStep, reviewInputs, task, iteration);
       history.push(reviewResult);
-      const verdict = reviewResult.verdict ?? "needs_revision";
+      let verdict = reviewResult.verdict ?? "needs_revision";
+      const score = reviewResult.score;
+      const threshold = loopOptions?.autoApproveThreshold;
+      if (verdict === "needs_revision" && threshold != null && score != null && score >= threshold && iteration >= 2) {
+        verdict = "approved";
+      }
       eventBus.emitEvent({
         type: "loop.iteration",
         stepId: reviewStep.definition.id,
@@ -3722,7 +3733,8 @@ var LoopController = class {
           finalOutput: generateResult.output,
           iterations: iteration,
           approved,
-          history
+          history,
+          finalScore: score
         };
       }
       if (iteration < maxIterations) {
@@ -4464,10 +4476,15 @@ var StepRunner = class {
     });
     let verdict;
     let feedback;
+    let score;
     if (stepType === "review") {
       const parsed = parseVerdict(result.text);
       verdict = parsed.verdict;
       feedback = parsed.feedback || void 0;
+      const scoreMatch = result.text.slice(-500).match(/SCORE:\s*(\d+)\/10/);
+      if (scoreMatch) {
+        score = Number.parseInt(scoreMatch[1], 10);
+      }
     }
     this.eventBus.emitEvent({
       type: "step.completed",
@@ -4483,7 +4500,8 @@ var StepRunner = class {
       usage: result.usage,
       durationMs,
       verdict,
-      feedback
+      feedback,
+      score
     };
   }
   resolvePromptType(stepType, stepId, iteration) {
@@ -4755,7 +4773,8 @@ var Orchestrator = class extends EventEmitter2 {
   async plan(task, options) {
     return this.executeSession(task, ["plan-review"], {
       maxIterations: options?.maxRounds,
-      stream: options?.stream
+      stream: options?.stream,
+      planAutoApproveThreshold: options?.autoApproveThreshold ?? 8
     });
   }
   /**
@@ -5073,7 +5092,8 @@ ${r.text}`)
           task,
           maxIterations,
           stepRunner,
-          this.eventBus
+          this.eventBus,
+          { autoApproveThreshold: options.planAutoApproveThreshold }
         );
         inputs.set("plan.output", planLoop.finalOutput);
         totalIterations += planLoop.iterations;
@@ -6080,7 +6100,7 @@ var hostFindingSchema = z3.object({
 var hostFindingsSchema = z3.array(hostFindingSchema);
 
 // src/index.ts
-var VERSION = "0.2.6";
+var VERSION = "0.2.8";
 export {
   ArtifactStore,
   BINARY_SNIFF_BYTES,