@principles/pd-cli 1.80.0 → 1.81.0

package/src/commands/pain-retry.ts

@@ -0,0 +1,547 @@
+/**
+ * pd pain retry command — Retry a failed diagnosis by pain ID.
+ *
+ * Looks up the diagnostician task for a given painId, validates retry eligibility,
+ * then delegates to the existing diagnose run logic.
+ *
+ * Usage:
+ *   pd pain retry --pain-id <painId> --workspace <path> --runtime <kind> [runtime flags] [--json] [--force]
+ *
+ * IMPORTANT: --runtime is required (no default). test-double would generate fake
+ * candidates/ledger in a real workspace, so it must be explicitly requested.
+ */
+import {
+  RuntimeStateManager,
+  SqliteHistoryQuery,
+  SqliteContextAssembler,
+  SqliteDiagnosticianCommitter,
+  SqliteTrajectoryLocator,
+  SqliteSourceTraceLocator,
+  StoreEventEmitter,
+  DiagnosticianRunner,
+  DefaultDiagnosticianValidator,
+  TestDoubleRuntimeAdapter,
+  OpenClawCliRuntimeAdapter,
+  PiAiRuntimeAdapter,
+  PDRuntimeError,
+  resolveRuntimeConfig,
+  isRuntimeConfigError,
+  CandidateIntakeService,
+  run as diagnoseRun,
+} from '@principles/core/runtime-v2';
+import type { PDRuntimeAdapter, RuntimeConfig } from '@principles/core/runtime-v2';
+import type { PDTaskStatus } from '@principles/core/runtime-v2';
+import { PrincipleTreeLedgerAdapter } from '../principle-tree-ledger-adapter.js';
+import { resolveWorkspaceDir } from '../resolve-workspace.js';
+import * as path from 'path';
+
+// ── Types ──────────────────────────────────────────────────────────────────────
+
+interface PainRetryOptions {
+  painId: string;
+  workspace?: string;
+  json?: boolean;
+  force?: boolean;
+  runtime?: string;
+  openclawLocal?: boolean;
+  openclawGateway?: boolean;
+  agent?: string;
+  provider?: string;
+  model?: string;
+  apiKeyEnv?: string;
+  baseUrl?: string;
+  maxRetries?: number;
+  timeoutMs?: number;
+}
+
+/** Allowed task statuses for retry without --force. */
+const RETRYABLE_STATUSES: ReadonlySet<PDTaskStatus> = new Set([
+  'retry_wait',
+  'failed',
+  'needs_human_review',
+]);
+
+/**
+ * Resolve a painId to a diagnostician taskId.
+ *
+ * Convention: task_id = `diagnosis_<painId>` (see PainToPrincipleService).
+ * If the painId already has the `diagnosis_` prefix, reject to avoid double-prefixing.
+ */
+function resolveTaskIdFromPainId(painId: string): { taskId: string } | { reason: string; nextAction: string } {
+  if (painId.startsWith('diagnosis_')) {
+    return {
+      reason: `painId '${painId}' already has 'diagnosis_' prefix — this looks like a taskId, not a painId`,
+      nextAction: `Use the raw painId (without 'diagnosis_' prefix), or use 'pd diagnose run --task-id ${painId}' directly`,
+    };
+  }
+  return { taskId: `diagnosis_${painId}` };
+}
+
+/** Return value as a non-blank string if it is one, otherwise null. */
+function readNonBlankString(value: unknown): string | null {
+  return typeof value === 'string' && value.trim().length > 0 ? value : null;
+}
+
+/** Output a refused/not_found result, respecting --json mode. Exits with code 1. */
+function refuseExit(opts: PainRetryOptions, payload: { status?: string; painId: string; taskId?: string; reason: string; message?: string; nextAction: string }): never {
+  if (opts.json) {
+    console.log(JSON.stringify({
+      status: payload.status ?? 'refused',
+      painId: payload.painId,
+      taskId: payload.taskId ?? null,
+      reason: payload.reason,
+      ...(payload.message ? { message: payload.message } : {}),
+      nextAction: payload.nextAction,
+    }));
+  } else {
+    console.error(`error: ${payload.message ?? payload.reason}`);
+    console.error(`nextAction: ${payload.nextAction}`);
+  }
+  process.exit(1);
+}
+
+// ── Handler ────────────────────────────────────────────────────────────────────
+
+export async function handlePainRetry(opts: PainRetryOptions): Promise<void> {
+  const workspaceDir = resolveWorkspaceDir(opts.workspace);
+  const stateDir = `${workspaceDir}/.state`;
+
+  // Step 1: Resolve painId → taskId
+  const resolution = resolveTaskIdFromPainId(opts.painId);
+  if ('reason' in resolution) {
+    refuseExit(opts, { painId: opts.painId, reason: resolution.reason, nextAction: resolution.nextAction });
+  }
+
+  const { taskId } = resolution;
+
+  // Step 2: Look up task and validate
+  const stateManager = new RuntimeStateManager({ workspaceDir });
+
+  try {
+    await stateManager.initialize();
+
+    const task = await stateManager.getTask(taskId);
+    if (!task) {
+      refuseExit(opts, {
+        status: 'not_found',
+        painId: opts.painId,
+        taskId,
+        reason: 'task_not_found',
+        message: `No task found for painId '${opts.painId}' (looked for taskId '${taskId}')`,
+        nextAction: `Verify the painId is correct. Use 'pd task list --kind diagnostician' to see all diagnostician tasks.`,
+      });
+    }
+
+    if (task.taskKind !== 'diagnostician') {
+      refuseExit(opts, {
+        painId: opts.painId,
+        taskId,
+        reason: 'wrong_task_kind',
+        message: `Task '${taskId}' is not a diagnostician task (taskKind='${task.taskKind}')`,
+        nextAction: `pd pain retry only retries diagnostician tasks. Use 'pd diagnose run --task-id ${taskId}' for other task kinds.`,
+      });
+    }
+
+    const previousTaskStatus = task.status;
+    const previousLastError = task.lastError ?? null;
+
+    if (task.status === 'succeeded' && !opts.force) {
+      refuseExit(opts, {
+        painId: opts.painId,
+        taskId,
+        reason: 'already_succeeded',
+        message: `Task '${taskId}' already succeeded. Use --force to re-run a succeeded task.`,
+        nextAction: `Add --force to retry: pd pain retry --pain-id ${opts.painId} --force`,
+      });
+    }
+
+    if (!RETRYABLE_STATUSES.has(task.status) && task.status !== 'succeeded') {
+      refuseExit(opts, {
+        painId: opts.painId,
+        taskId,
+        reason: 'status_not_retryable',
+        message: `Task '${taskId}' has status '${task.status}' which is not retryable. Retryable statuses: ${[...RETRYABLE_STATUSES].join(', ')}`,
+        nextAction: `Wait for the task to reach a terminal state, or use 'pd diagnose run --task-id ${taskId}' directly.`,
+      });
+    }
+
+    // Step 3: Resolve runtime kind
+    // P1 fix: --openclaw-local and --openclaw-gateway are mutually exclusive.
+    // Must output JSON when --json is set (CLI operator gate).
+    if (opts.openclawLocal && opts.openclawGateway) {
+      refuseExit(opts, {
+        painId: opts.painId,
+        taskId,
+        reason: 'conflicting_flags',
+        message: '--openclaw-local and --openclaw-gateway are mutually exclusive',
+        nextAction: 'Provide exactly one of --openclaw-local or --openclaw-gateway, not both.',
+      });
+    }
+
+    // P1 fix: pd pain retry must NOT default to test-double.
+    // This command is for real workspace pain fixes — test-double would generate
+    // fake candidates/ledger in a real .pd/state.db. Require explicit --runtime
+    // or fall back to workflows.yaml config.
+    let runtimeKind = opts.runtime;
+    if (!runtimeKind) {
+      try {
+        const configResult = resolveRuntimeConfig(stateDir);
+        if (!isRuntimeConfigError(configResult) && configResult.runtimeKind) {
+          ({ runtimeKind } = configResult);
+        }
+      } catch {
+        // Config load failed — fall through to refusal
+      }
+    }
+
+    if (!runtimeKind) {
+      refuseExit(opts, {
+        painId: opts.painId,
+        taskId,
+        reason: 'missing_runtime',
+        message: 'No --runtime specified and no workflows.yaml config found. pd pain retry must not default to test-double to prevent fake data in real workspaces.',
+        nextAction: `Specify --runtime explicitly: pd pain retry --pain-id ${opts.painId} --runtime pi-ai --provider <provider> --model <model> --apiKeyEnv <ENV>`,
+      });
+    }
+
+    // Step 4: Build runtime adapter
+    let runtimeAdapter: PDRuntimeAdapter;
+
+    if (runtimeKind === 'openclaw-cli') {
+      const configResult = resolveRuntimeConfig(stateDir, { openclawLocal: opts.openclawLocal, openclawGateway: opts.openclawGateway, requestedRuntimeKind: 'openclaw-cli' });
+      if (isRuntimeConfigError(configResult)) {
+        refuseExit(opts, { painId: opts.painId, taskId, reason: configResult.reason, message: configResult.message, nextAction: configResult.nextAction });
+      }
+      const { openclawMode } = configResult;
+      if (!openclawMode) {
+        refuseExit(opts, {
+          painId: opts.painId,
+          taskId,
+          reason: 'missing_openclaw_mode',
+          message: 'runtimeKind is openclaw-cli but no mode resolved',
+          nextAction: 'Provide --openclaw-local or --openclaw-gateway',
+        });
+      }
+
+      runtimeAdapter = new OpenClawCliRuntimeAdapter({
+        runtimeMode: openclawMode,
+        workspaceDir,
+        agentId: opts.agent ?? 'main',
+      });
+    } else if (runtimeKind === 'test-double') {
+      runtimeAdapter = new TestDoubleRuntimeAdapter({
+        onPollRun: (_runId: string) => ({
+          runId: _runId,
+          status: 'succeeded',
+          startedAt: new Date().toISOString(),
+          endedAt: new Date().toISOString(),
+        }),
+        onFetchOutput: (_runId: string) => ({
+          runId: _runId,
+          payload: {
+            valid: true,
+            diagnosisId: `diag-retry-${Date.now()}`,
+            taskId,
+            summary: 'CLI retry test diagnosis — validate tool arguments before execution',
+            rootCause: 'Test root cause — missing argument validation',
+            violatedPrinciples: [],
+            evidence: [],
+            recommendations: [
+              { kind: 'principle', description: 'Always validate tool arguments before execution to prevent silent failures' },
+              { kind: 'rule', description: 'Use schema validation for external inputs' },
+            ],
+            confidence: 0.9,
+          },
+        }),
+      });
+    } else if (runtimeKind === 'pi-ai') {
+      let policyConfig: RuntimeConfig | null = null;
+      try {
+        const configResult = resolveRuntimeConfig(stateDir);
+        if (!isRuntimeConfigError(configResult)) {
+          policyConfig = configResult;
+        } else {
+          console.warn(`[pd pain retry] workflows.yaml policy load failed: ${configResult.message}. Using CLI flags if provided.`);
+        }
+      } catch (err: unknown) {
+        const detail = err instanceof Error ? err.message : String(err);
+        console.warn(`[pd pain retry] workflows.yaml policy load failed: ${detail}. Using CLI flags if provided.`);
+      }
+
+      const provider = opts.provider ?? policyConfig?.provider;
+      const model = opts.model ?? policyConfig?.model;
+      const apiKeyEnv = opts.apiKeyEnv ?? policyConfig?.apiKeyEnv;
+      const baseUrl = opts.baseUrl ?? policyConfig?.baseUrl;
+      const maxRetries = opts.maxRetries ?? policyConfig?.maxRetries;
+      const effectiveTimeoutMs = opts.timeoutMs ?? policyConfig?.timeoutMs;
+
+      // Validate required string fields: must be non-blank strings
+      const missing: string[] = [];
+      if (readNonBlankString(provider) === null) missing.push('provider');
+      if (readNonBlankString(model) === null) missing.push('model');
+      if (readNonBlankString(apiKeyEnv) === null) missing.push('apiKeyEnv');
+      if (missing.length > 0) {
+        refuseExit(opts, {
+          painId: opts.painId,
+          taskId,
+          reason: `missing_required_config: ${missing.join(', ')}`,
+          message: `Missing or blank required pi-ai config: ${missing.join(', ')}`,
+          nextAction: `Pass via --flag or add to workflows.yaml. Example: pd pain retry --pain-id ${opts.painId} --runtime pi-ai --provider openrouter --model anthropic/claude-sonnet-4 --apiKeyEnv OPENROUTER_API_KEY`,
+        });
+      }
+
+      // Validate numeric options: must be finite, integer, non-negative if provided
+      const invalidNumeric: string[] = [];
+      if (maxRetries !== undefined && maxRetries !== null && !(Number.isFinite(maxRetries) && Number.isInteger(maxRetries) && maxRetries >= 0)) {
+        invalidNumeric.push(`maxRetries (got: ${maxRetries})`);
+      }
+      if (effectiveTimeoutMs !== undefined && effectiveTimeoutMs !== null && !(Number.isFinite(effectiveTimeoutMs) && effectiveTimeoutMs > 0)) {
+        invalidNumeric.push(`timeoutMs (got: ${effectiveTimeoutMs})`);
+      }
+      if (invalidNumeric.length > 0) {
+        refuseExit(opts, {
+          painId: opts.painId,
+          taskId,
+          reason: `invalid_numeric_config: ${invalidNumeric.join(', ')}`,
+          message: `Invalid numeric pi-ai config: ${invalidNumeric.join(', ')}. maxRetries must be a non-negative integer; timeoutMs must be a positive number.`,
+          nextAction: 'Fix the numeric values and retry.',
+        });
+      }
+
+      // After validation, these are guaranteed non-blank strings.
+      const validProvider = readNonBlankString(provider);
+      const validModel = readNonBlankString(model);
+      const validApiKeyEnv = readNonBlankString(apiKeyEnv);
+      if (validProvider === null || validModel === null || validApiKeyEnv === null) {
+        refuseExit(opts, {
+          painId: opts.painId,
+          taskId,
+          reason: 'internal_validation_error',
+          message: 'Internal error: validated string fields became null after validation.',
+          nextAction: 'This should not happen. Please report this bug.',
+        });
+      }
+
+      if (!process.env[validApiKeyEnv]) {
+        refuseExit(opts, {
+          painId: opts.painId,
+          taskId,
+          reason: 'missing_api_key',
+          message: `Environment variable '${validApiKeyEnv}' is not set`,
+          nextAction: `Set the environment variable: export ${validApiKeyEnv}=<your-api-key>`,
+        });
+      }
+
+      runtimeAdapter = new PiAiRuntimeAdapter({
+        provider: validProvider,
+        model: validModel,
+        apiKeyEnv: validApiKeyEnv,
+        baseUrl,
+        maxRetries,
+        timeoutMs: effectiveTimeoutMs,
+        workspace: workspaceDir,
+      });
+    } else {
+      refuseExit(opts, {
+        painId: opts.painId,
+        taskId,
+        reason: `unknown_runtime: '${runtimeKind}'`,
+        message: `Unknown runtime kind '${runtimeKind}'`,
+        nextAction: 'Supported runtimes: openclaw-cli, test-double, pi-ai',
+      });
+    }
+
+    // Step 5: Build runner and execute (same as diagnose run)
+    const sqliteConn = stateManager.connection;
+    const { taskStore } = stateManager;
+    const { runStore } = stateManager;
+    const historyQuery = new SqliteHistoryQuery(sqliteConn);
+    const trajectoryLocator = new SqliteTrajectoryLocator(sqliteConn);
+    const sourceTraceLocator = new SqliteSourceTraceLocator(taskStore, trajectoryLocator);
+    const contextAssembler = new SqliteContextAssembler(taskStore, historyQuery, runStore, { sourceTraceLocator });
+
+    const eventEmitter = new StoreEventEmitter();
+    const committer = new SqliteDiagnosticianCommitter(sqliteConn);
+    const runner = new DiagnosticianRunner(
+      {
+        stateManager,
+        contextAssembler,
+        runtimeAdapter,
+        eventEmitter,
+        validator: new DefaultDiagnosticianValidator(),
+        committer,
+      },
+      {
+        owner: 'pd-cli-pain-retry',
+        runtimeKind,
+        pollIntervalMs: 100,
+        timeoutMs: 300_000,
+        agentId: opts.agent,
+      },
+    );
+
+    if (!opts.json) {
+      console.log(`\nRetrying diagnosis for pain: ${opts.painId}`);
+      console.log(`  Task ID:  ${taskId}`);
+      console.log(`  Previous: ${previousTaskStatus}${previousLastError ? ` (${previousLastError})` : ''}`);
+      console.log(`  Runtime:  ${runtimeKind}`);
+      console.log(`  Workspace: ${workspaceDir}\n`);
+    }
+
+    const result = await diagnoseRun({
+      taskId,
+      stateManager,
+      runner,
+    });
+
+    if (result.status !== 'succeeded') {
+      if (opts.json) {
+        console.log(JSON.stringify({
+          status: 'failed',
+          painId: opts.painId,
+          taskId,
+          runId: null,
+          runtimeKind,
+          previousTaskStatus,
+          previousLastError,
+          newTaskStatus: result.status,
+          errorCategory: result.errorCategory ?? null,
+          failureReason: result.failureReason ?? null,
+          nextAction: result.errorCategory === 'output_invalid'
+            ? 'The LLM output failed validation. Try a different model or provider.'
+            : 'Check the error category and retry with adjusted parameters.',
+        }, null, 2));
+      } else {
+        console.log(`\nRetry failed:`);
+        console.log(`  Status:         ${result.status}`);
+        console.log(`  Task ID:        ${result.taskId}`);
+        if (result.errorCategory) {
+          console.log(`  Error Category: ${result.errorCategory}`);
+        }
+        if (result.failureReason) {
+          console.log(`  Failure Reason: ${result.failureReason}`);
+        }
+        console.log('');
+      }
+      process.exit(1);
+      return;
+    }
+
+    // Step 6: Intake candidates
+    const candidates = await stateManager.getCandidatesByTaskId(taskId);
+    const intakeResults: { candidateId: string; ledgerEntryId?: string; status: string; error?: string; nextAction?: string }[] = [];
+    let intakeFailed = false;
+
+    const ledgerAdapter = new PrincipleTreeLedgerAdapter({ stateDir: path.join(workspaceDir, '.state') });
+    const intakeService = new CandidateIntakeService({ stateManager, ledgerAdapter });
+
+    for (const candidate of candidates) {
+      try {
+        const entry = await intakeService.intake(candidate.candidateId);
+        if (candidate.status !== 'consumed') {
+          await stateManager.updateCandidateStatus(candidate.candidateId, { status: 'consumed' });
+        }
+        intakeResults.push({
+          candidateId: candidate.candidateId,
+          ledgerEntryId: entry.id,
+          status: 'consumed',
+        });
+      } catch (intakeErr: unknown) {
+        intakeFailed = true;
+        const intakeErrorMessage = intakeErr instanceof Error ? intakeErr.message : String(intakeErr);
+        intakeResults.push({
+          candidateId: candidate.candidateId,
+          status: 'intake_failed',
+          error: intakeErrorMessage,
+          nextAction: `pd candidate intake --candidate-id ${candidate.candidateId} --workspace "${workspaceDir}"`,
+        });
+      }
+    }
+
+    const candidateIds = candidates.map((c) => c.candidateId);
+    const ledgerEntryIds = intakeResults
+      .filter((ir): ir is { candidateId: string; ledgerEntryId: string; status: string } => ir.status === 'consumed' && typeof ir.ledgerEntryId === 'string')
+      .map((ir) => ir.ledgerEntryId);
+
+    // Build nextAction: candidates generated but internalization not automatic
+    const internalizeNextAction = candidateIds.length > 0
+      ? `Candidates generated but internalization has NOT started automatically. To begin internalization, run:\n  ${candidateIds.map((id) => `pd candidate internalize --candidate-id ${id} --workspace "${workspaceDir}"`).join('\n  ')}`
+      : 'No candidates were generated from this diagnosis.';
+
+    if (opts.json) {
+      // Strict single JSON object output
+      console.log(JSON.stringify({
+        status: 'succeeded',
+        painId: opts.painId,
+        taskId,
+        runId: null,
+        runtimeKind,
+        previousTaskStatus,
+        previousLastError,
+        newTaskStatus: 'succeeded',
+        candidateIds,
+        ledgerEntryIds,
+        intake: {
+          candidates: intakeResults,
+        },
+        nextAction: internalizeNextAction,
+      }, null, 2));
+      if (intakeFailed) {
+        process.exit(1);
+      }
+      return;
+    }
+
+    // Text output
+    console.log(`\nRetry succeeded:`);
+    console.log(`  Pain ID:         ${opts.painId}`);
+    console.log(`  Task ID:         ${taskId}`);
+    console.log(`  Previous Status: ${previousTaskStatus}${previousLastError ? ` (${previousLastError})` : ''}`);
+    console.log(`  New Status:      succeeded`);
+    console.log(`  Candidates:      ${candidateIds.length}`);
+    if (result.contextHash) {
+      console.log(`  Context Hash:    ${result.contextHash.substring(0, 16)}...`);
+    }
+
+    if (intakeResults.length > 0) {
+      console.log(`\n  Candidate Intake:`);
+      for (const ir of intakeResults) {
+        if (ir.status === 'consumed') {
+          console.log(`    ${ir.candidateId}: consumed (ledger: ${ir.ledgerEntryId})`);
+        } else if (ir.status === 'intake_failed') {
+          console.log(`    ${ir.candidateId}: INTAKE FAILED — ${ir.error}`);
+          console.log(`      Next action: ${ir.nextAction}`);
+        }
+      }
+    }
+
+    console.log(`\n  Next Action:`);
+    console.log(`  ${internalizeNextAction}`);
+    console.log('');
+
+    if (intakeFailed) {
+      process.exit(1);
+    }
+  } catch (error: unknown) {
+    const message = error instanceof Error ? error.message : String(error);
+    let errorCategory = 'execution_failed';
+    if (error instanceof PDRuntimeError) {
+      errorCategory = error.category;
+    }
+    if (opts.json) {
+      console.log(JSON.stringify({
+        status: 'failed',
+        painId: opts.painId,
+        taskId: `diagnosis_${opts.painId}`,
+        errorCategory,
+        message,
+        nextAction: 'Check the error message and retry with adjusted parameters.',
+      }, null, 2));
+    } else {
+      console.error(`error: ${message} (${errorCategory})`);
+    }
+    process.exit(1);
+  } finally {
+    await stateManager.close();
+  }
+}

package/src/index.ts

@@ -8,6 +8,7 @@
 
 import { Command } from 'commander';
 import { handlePainRecord } from './commands/pain-record.js';
+import { handlePainRetry } from './commands/pain-retry.js';
 import { handleSamplesList } from './commands/samples-list.js';
 import { handleSamplesReview } from './commands/samples-review.js';
 import { handleEvolutionTasksList } from './commands/evolution-tasks-list.js';
@@ -76,6 +77,27 @@ painCmd
     await handlePainRecord(opts);
   });
 
+painCmd
+  .command('retry')
+  .description('Retry a failed diagnosis by pain ID')
+  .requiredOption('-p, --pain-id <painId>', 'Pain ID to retry diagnosis for')
+  .option('-w, --workspace <path>', 'Workspace directory')
+  .option('-r, --runtime <kind>', "Runtime kind: 'openclaw-cli', 'test-double', 'pi-ai'")
+  .option('--openclaw-local', 'Use local OpenClaw (mutually exclusive with --openclaw-gateway)')
+  .option('--openclaw-gateway', 'Use gateway OpenClaw (mutually exclusive with --openclaw-local)')
+  .option('-a, --agent <agentId>', 'Agent ID to invoke')
+  .option('--provider <name>', 'LLM provider (e.g., openrouter) — for pi-ai, falls back to policy')
+  .option('--model <id>', 'Model ID (e.g., anthropic/claude-sonnet-4) — for pi-ai, falls back to policy')
+  .option('--apiKeyEnv <name>', 'Env var name for API key — for pi-ai, falls back to policy')
+  .option('--baseUrl <url>', 'Custom base URL — for pi-ai, falls back to policy')
+  .option('--maxRetries <n>', 'Max retry attempts for LLM failures — for pi-ai, falls back to policy', parseInt)
+  .option('--timeoutMs <ms>', 'Timeout in milliseconds — for pi-ai, falls back to policy', parseInt)
+  .option('--force', 'Allow retry of already-succeeded tasks')
+  .option('--json', 'Output raw JSON')
+  .action(async (opts) => {
+    await handlePainRetry(opts);
+  });
+
 const samplesCmd = program
   .command('samples')
   .description('Correction sample management');