peaks-loop 3.1.2 → 4.0.0-beta.10

package/dist/cli/commands/workflow-commands.js

@@ -0,0 +1,544 @@
+import { createRdSwarmPlan } from '../../services/rd/rd-service.js';
+import { createTechPlan, getTechStatus } from '../../services/tech/tech-service.js';
+import { createWorkflowRouterPlan, isCodeMode, isWorkflowMode } from '../../services/workflow/workflow-router-service.js';
+import { createAutonomousWorkflowPlan } from '../../services/workflow/workflow-autonomous-service.js';
+import { writeAutonomousResumeArtifacts } from '../../services/workflow/autonomous-resume-writer.js';
+import { createRecommendationPlan } from '../../services/recommendations/recommendation-service.js';
+import { createRefactorDryRun } from '../../services/refactor/refactor-service.js';
+import { getWorkspaceConfigForPath, readConfig } from '../../services/config/config-service.js';
+import { getEconomyAwareExecutionModelId } from '../../services/config/model-routing.js';
+import { getLocalArtifactPath } from '../../services/artifacts/workspace-service.js';
+import { getSessionId } from '../../services/session/session-manager.js';
+import { getSessionDir } from '../../services/session/getSessionDir.js';
+import { findProjectRoot } from '../../services/config/config-safety.js';
+import { verifyPipeline } from '../../services/workflow/pipeline-verify-service.js';
+import { applySkip, detectCallerKind } from '../../services/workflow/workflow-skip-service.js';
+import { fail, ok } from '../../shared/result.js';
+import { addJsonOption, failUnsupportedNonDryRun, getErrorMessage, isRecommendationWorkflow, printResult } from '../cli-helpers.js';
+// Slice 015 — typed error → envelope code routing. Used by the four
+// catch sites below (tech.plan / workflow.route / workflow.autonomous /
+// swarm.plan) so provider-config errors surface as `INVALID_PROVIDERS`
+// instead of being silently re-labelled `INVALID_GOAL`.
+import { mapServiceError } from './_cli-error-envelope.js';
+// Plan 1 / Task 9 — auto-build peaks-context before peaks-rd runs.
+import { buildContext } from '../../services/context/context-builder.js';
+// Plan 1 / Task 10 — production fetcher (replaces mockFetcher).
+import { createHeadroomFetcher } from '../../services/context/headroom-fetcher.js';
+function buildHeadroomFetcher(sid) {
+    return createHeadroomFetcher({
+        cacheDir: `.peaks/_runtime/${sid}/doc-cache`,
+        // remoteFetcher wired in a future slice (headroom-ai programmatic API).
+    });
+}
+async function ensureContextForRd(goal, project, sid) {
+    const out = `.peaks/_runtime/${sid}/context.json`;
+    try {
+        await buildContext({
+            goal,
+            project,
+            audience: 'peaks-rd',
+            depsMode: 'locked',
+            docBudgetTokens: 8000,
+            out,
+            fetcher: buildHeadroomFetcher(sid),
+        });
+    }
+    catch (error) {
+        // Plan 1 / Task 9 — context is a pre-step, not a precondition.
+        // If the Collector (e.g. missing package.json) or DocRetriever
+        // fails, we still want the rd slice to proceed. Task 11
+        // upgrades this to a hard precondition once the rd slice
+        // actually consumes context.json.
+        const message = error instanceof Error ? error.message : 'unknown context build failure';
+        process.stderr.write(`[peaks-context] rd pre-step skipped: ${message}\n`);
+    }
+}
+function getCurrentWorkspaceContext() {
+    try {
+        const projectRoot = findProjectRoot(process.cwd()) ?? process.cwd();
+        const sessionId = getSessionId(projectRoot);
+        return sessionId ? { sessionId, sessionDir: getSessionDir(projectRoot, sessionId) } : {};
+    }
+    catch {
+        return {};
+    }
+}
+function getWorkflowWorkspaceContext() {
+    try {
+        const projectRoot = findProjectRoot(process.cwd()) ?? process.cwd();
+        const workspace = getWorkspaceConfigForPath(projectRoot);
+        if (!workspace)
+            return { projectRoot };
+        return { projectRoot, workspace, artifactWorkspacePath: getLocalArtifactPath(workspace) };
+    }
+    catch {
+        return {};
+    }
+}
+function parseMaxWorkers(io, command, value, asJson) {
+    const maxWorkers = Number(value);
+    if (!Number.isInteger(maxWorkers) || maxWorkers < 1) {
+        printResult(io, fail(command, 'INVALID_MAX_WORKERS', 'max-workers must be a positive integer', {}, ['Use --max-workers with a positive integer value']), asJson);
+        process.exitCode = 1;
+        return null;
+    }
+    return maxWorkers;
+}
+function validatePlanningInput(goal) {
+    // v2.17.0: change-id axis removed. The planning input is now keyed
+    // by the session id (already bound via `peaks workspace init`),
+    // not by a user-supplied change-id. Only the goal is validated.
+    if (!goal.trim()) {
+        throw new Error('Goal must be non-empty');
+    }
+}
+function parseCodeMode(io, command, mode, codeMode, asJson) {
+    if (mode !== 'code' && codeMode) {
+        printResult(io, fail(command, 'CODE_MODE_REQUIRES_CODE_WORKFLOW', '--code-mode can only be used with --mode code', {}, ['Remove --code-mode or use --mode code']), asJson);
+        process.exitCode = 1;
+        return null;
+    }
+    if (mode !== 'code' || !codeMode) {
+        return undefined;
+    }
+    if (!isCodeMode(codeMode)) {
+        printResult(io, fail(command, 'UNSUPPORTED_CODE_MODE', `Unsupported code mode ${codeMode}`, {}, ['Use --code-mode full-auto, guided, or rnd']), asJson);
+        process.exitCode = 1;
+        return null;
+    }
+    return codeMode;
+}
+function runTechPlan(io, options) {
+    if (options.dryRun === false) {
+        failUnsupportedNonDryRun(io, 'tech.plan', options.json);
+        return;
+    }
+    try {
+        validatePlanningInput(options.goal);
+        const workspaceContext = getCurrentWorkspaceContext();
+        // Slice 2026-06-29-change-id-root-removal: the change-id axis is
+        // gone. The planner derives scope from the active session; the
+        // legacy `sessionId` field is passed as empty string for back-compat
+        // with the existing service signature.
+        const plan = createTechPlan({
+            sessionId: '',
+            goal: options.goal,
+            swarm: options.swarm ?? false,
+            dryRun: true,
+            ...workspaceContext
+        });
+        printResult(io, ok('tech.plan', plan), options.json);
+    }
+    catch (error) {
+        const mapping = mapServiceError(error);
+        printResult(io, fail('tech.plan', mapping.code, getErrorMessage(error), {}, [...mapping.nextActions]), options.json);
+        process.exitCode = 1;
+    }
+}
+function runTechStatus(io, options) {
+    try {
+        const workspaceContext = getCurrentWorkspaceContext();
+        printResult(io, ok('tech.status', getTechStatus({ sessionId: '', ...workspaceContext })), options.json);
+    }
+    catch (error) {
+        printResult(io, fail('tech.status', 'TECH_STATUS_FAILED', getErrorMessage(error), {}, ['Verify the project setup']), options.json);
+        process.exitCode = 1;
+    }
+}
+function runWorkflowRoute(io, options) {
+    if (options.dryRun === false) {
+        failUnsupportedNonDryRun(io, 'workflow.route', options.json);
+        return;
+    }
+    if (!isWorkflowMode(options.mode)) {
+        printResult(io, fail('workflow.route', 'UNSUPPORTED_WORKFLOW_MODE', `Unsupported workflow mode ${options.mode}`, {}, ['Use --mode code or --mode team']), options.json);
+        process.exitCode = 1;
+        return;
+    }
+    const maxWorkers = parseMaxWorkers(io, 'workflow.route', options.maxWorkers, options.json);
+    if (maxWorkers === null)
+        return;
+    const codeMode = parseCodeMode(io, 'workflow.route', options.mode, options.codeMode, options.json);
+    if (codeMode === null)
+        return;
+    try {
+        validatePlanningInput(options.goal);
+        const workspaceContext = getWorkflowWorkspaceContext();
+        const plan = createWorkflowRouterPlan({
+            sessionId: '',
+            goal: options.goal,
+            mode: options.mode,
+            ...(codeMode ? { codeMode } : {}),
+            maxWorkers,
+            dryRun: true,
+            config: readConfig(),
+            ...workspaceContext
+        });
+        printResult(io, ok('workflow.route', plan), options.json);
+    }
+    catch (error) {
+        const mapping = mapServiceError(error);
+        printResult(io, fail('workflow.route', mapping.code, getErrorMessage(error), {}, [...mapping.nextActions]), options.json);
+        process.exitCode = 1;
+    }
+}
+function runAutonomousWorkflow(io, options) {
+    if (options.dryRun === false) {
+        failUnsupportedNonDryRun(io, 'workflow.autonomous', options.json);
+        return;
+    }
+    if (!isWorkflowMode(options.mode)) {
+        printResult(io, fail('workflow.autonomous', 'UNSUPPORTED_WORKFLOW_MODE', `Unsupported workflow mode ${options.mode}`, {}, ['Use --mode code or --mode team']), options.json);
+        process.exitCode = 1;
+        return;
+    }
+    const maxWorkers = parseMaxWorkers(io, 'workflow.autonomous', options.maxWorkers, options.json);
+    if (maxWorkers === null)
+        return;
+    const codeMode = parseCodeMode(io, 'workflow.autonomous', options.mode, options.codeMode, options.json);
+    if (codeMode === null)
+        return;
+    try {
+        validatePlanningInput(options.goal);
+        const workspaceContext = getWorkflowWorkspaceContext();
+        const plan = createAutonomousWorkflowPlan({
+            sessionId: '',
+            goal: options.goal,
+            mode: options.mode,
+            ...(codeMode ? { codeMode } : {}),
+            maxWorkers,
+            dryRun: true,
+            config: readConfig(),
+            ...workspaceContext
+        });
+        printResult(io, ok('workflow.autonomous', plan), options.json);
+    }
+    catch (error) {
+        const mapping = mapServiceError(error);
+        printResult(io, fail('workflow.autonomous', mapping.code, getErrorMessage(error), {}, [...mapping.nextActions]), options.json);
+        process.exitCode = 1;
+    }
+}
+async function runSwarmPlan(io, options) {
+    if ((options.skill ?? 'rd') !== 'rd') {
+        printResult(io, fail('swarm.plan', 'UNSUPPORTED_SWARM_SKILL', `Unsupported skill ${options.skill}`, {}, ['Use --skill rd']), options.json);
+        process.exitCode = 1;
+        return;
+    }
+    if (options.dryRun === false) {
+        failUnsupportedNonDryRun(io, 'swarm.plan', options.json);
+        return;
+    }
+    const maxWorkers = parseMaxWorkers(io, 'swarm.plan', options.maxWorkers, options.json);
+    if (maxWorkers === null)
+        return;
+    try {
+        validatePlanningInput(options.goal);
+        const workspaceContext = getWorkflowWorkspaceContext();
+        const config = readConfig();
+        // Plan 1 / Task 9 — pre-build peaks-context before peaks-rd runs.
+        const projectRoot = workspaceContext.projectRoot ?? process.cwd();
+        const sid = getSessionId(projectRoot) ?? 'ad-hoc';
+        await ensureContextForRd(options.goal, projectRoot, sid);
+        const plan = createRdSwarmPlan({
+            skill: 'rd',
+            sessionId: '',
+            goal: options.goal,
+            maxWorkers,
+            dryRun: true,
+            swarmMode: config.swarmMode ?? true,
+            executionModelId: getEconomyAwareExecutionModelId(config),
+            ...(options.strictStandards ? { strictStandards: true } : {}),
+            ...workspaceContext
+        });
+        // Slice 2026-06-16-peaks-rd-no-gates — wire the strict-mode exit code.
+        // The service-layer stamps `standardsErrorCode` onto the envelope; the
+        // CLI is responsible for translating it into a non-zero exit.
+        if (plan.gateStatus.standardsErrorCode === 'EPEAKS_NO_STANDARDS') {
+            process.exitCode = 1;
+        }
+        printResult(io, ok('swarm.plan', plan), options.json);
+    }
+    catch (error) {
+        const mapping = mapServiceError(error);
+        printResult(io, fail('swarm.plan', mapping.code, getErrorMessage(error), {}, [...mapping.nextActions]), options.json);
+        process.exitCode = 1;
+    }
+}
+async function runAutonomousResumeInit(io, options) {
+    try {
+        if (!options.project || !options.project.trim()) {
+            throw new Error('Project path must be non-empty');
+        }
+        // Slice 2026-06-29-change-id-root-removal: the change-id axis is
+        // gone. The CLI surfaces a deterministic placeholder
+        // (`session-default`) when the user does not pass a change-id, so
+        // the on-disk session-dir join via `getSessionDir` succeeds (the
+        // writer still requires a safe non-empty string).
+        const result = await writeAutonomousResumeArtifacts({
+            sessionId: 'session-default',
+            goal: options.goal,
+            artifactWorkspacePath: options.project,
+            apply: options.apply === true
+        });
+        const data = {
+            applied: result.applied,
+            files: result.files.map((file) => file.path)
+        };
+        const nextActions = result.applied
+            ? ['Run peaks workflow autonomous --goal "<goal>" --json to verify resumePlan.status']
+            : ['Re-run with --apply to write the resume scaffold to disk'];
+        printResult(io, ok('autonomous-resume.init', data, [], nextActions), options.json);
+    }
+    catch (error) {
+        printResult(io, fail('autonomous-resume.init', 'AUTONOMOUS_RESUME_INIT_FAILED', getErrorMessage(error), {}, ['Use a non-empty goal and a writable project path']), options.json);
+        process.exitCode = 1;
+    }
+}
+function addTechPlanOptions(command) {
+    return addJsonOption(command
+        .description('Generate a technical dry-run graph')
+        .requiredOption('--goal <goal>', 'planning goal')
+        .option('--swarm', 'opt into swarm-oriented planning')
+        .option('--dry-run', 'preview without writing files', true)
+        .option('--no-dry-run', 'unsupported: do not execute tech planning from this CLI'));
+}
+function addTechStatusOptions(command) {
+    return addJsonOption(command.description('Inspect technical approval status'));
+}
+function addWorkflowRouteOptions(command, description) {
+    return addJsonOption(command
+        .description(description)
+        .requiredOption('--mode <mode>', 'workflow mode: code or team')
+        .requiredOption('--goal <goal>', 'planning goal')
+        .option('--code-mode <mode>', 'code mode: full-auto, guided, or rnd')
+        .option('--max-workers <count>', 'maximum worker count', '40')
+        .option('--dry-run', 'preview without writing files', true)
+        .option('--no-dry-run', 'unsupported: do not execute workflow planning from this CLI'));
+}
+function addSwarmPlanOptions(command, includeSkill) {
+    const configured = command
+        .description('Plan an RD swarm dry-run graph')
+        .requiredOption('--goal <goal>', 'planning goal')
+        .option('--max-workers <count>', 'maximum worker count', '40')
+        .option('--dry-run', 'preview without writing files', true)
+        .option('--no-dry-run', 'unsupported: do not execute RD planning from this CLI')
+        .option('--strict-standards', 'hard-fail (exit non-zero) when project-local standards are missing; surfaces EPEAKS_NO_STANDARDS in the JSON envelope');
+    if (includeSkill) {
+        configured.requiredOption('--skill <skill>', 'skill to plan for');
+    }
+    return addJsonOption(configured);
+}
+function addAutonomousResumeInitOptions(command) {
+    return addJsonOption(command
+        .description('Write the autonomous resume artifact scaffold for the active change-id')
+        .requiredOption('--goal <goal>', 'planning goal')
+        .requiredOption('--project <path>', 'artifact workspace path to write under')
+        .option('--apply', 'write the artifacts to disk (default is dry-run preview)'));
+}
+export function registerWorkflowCommands(program, io) {
+    const refactor = program.command('refactor').description('Plan a Peaks refactor run without modifying code');
+    addJsonOption(refactor
+        .option('--code', 'use peaks-code orchestration mode')
+        .option('--rd', 'use peaks-rd direct mode')
+        .option('--dry-run', 'print gates and required artifacts', true)
+        .option('--no-dry-run', 'unsupported: do not modify code from this command')).action((options) => {
+        if (options.dryRun === false) {
+            failUnsupportedNonDryRun(io, 'refactor', options.json);
+            return;
+        }
+        if (options.code && options.rd) {
+            printResult(io, fail('refactor', 'CONFLICTING_REFACTOR_MODE', 'Choose either --code or --rd, not both', {}, ['Run peaks refactor --code --dry-run']), options.json);
+            process.exitCode = 1;
+            return;
+        }
+        const mode = options.rd ? 'rd' : 'code';
+        printResult(io, ok('refactor', createRefactorDryRun(mode), [], ['This dry run never edits code']), options.json);
+    });
+    const tech = program.command('tech').description('Plan and inspect technical dry-run gates');
+    addTechPlanOptions(tech.command('plan')).action((options) => runTechPlan(io, options));
+    addTechStatusOptions(tech.command('status')).action((options) => runTechStatus(io, options));
+    addTechPlanOptions(program.command('tech-plan')).action((options) => runTechPlan(io, options));
+    addTechStatusOptions(program.command('tech-status')).action((options) => runTechStatus(io, options));
+    const workflow = program.command('workflow').description('Plan workflow routing dry-run graphs');
+    addWorkflowRouteOptions(workflow.command('route'), 'Plan a workflow routing dry-run summary').action((options) => runWorkflowRoute(io, options));
+    addWorkflowRouteOptions(program.command('route'), 'Plan a workflow routing dry-run summary').action((options) => runWorkflowRoute(io, options));
+    addWorkflowRouteOptions(workflow.command('autonomous'), 'Plan an autonomous workflow handoff summary').action((options) => runAutonomousWorkflow(io, options));
+    addWorkflowRouteOptions(program.command('autonomous'), 'Plan an autonomous workflow handoff summary').action((options) => runAutonomousWorkflow(io, options));
+    const autonomousResume = workflow.command('autonomous-resume').description('Manage autonomous workflow resume artifacts');
+    addAutonomousResumeInitOptions(autonomousResume.command('init')).action((options) => runAutonomousResumeInit(io, options));
+    const autonomousResumeAlias = program.command('autonomous-resume').description('Manage autonomous workflow resume artifacts');
+    addAutonomousResumeInitOptions(autonomousResumeAlias.command('init')).action((options) => runAutonomousResumeInit(io, options));
+    addJsonOption(workflow
+        .command('verify-pipeline')
+        .description('Verify the complete rd→qa pipeline was followed for a request. Scans the v2.17.0 canonical session-axis layout (artifacts under _runtime per-session) and falls back to the legacy v2.16.0 change-axis forms during the 1-minor-release deprecation window.')
+        .requiredOption('--rid <rid>', 'request identifier')
+        .requiredOption('--project <path>', 'project root path')
+        .option('--type <type>', 'request type: feature, bugfix, refactor, docs, config, chore', 'feature')
+        .option('--session-id <sid>', 'slice 2026-06-13-peaks-workflow-skip: session id under which to read the skip-state file. When omitted, no skip-state is consulted (legacy behavior).')).action(async (options) => {
+        try {
+            const result = await verifyPipeline({
+                projectRoot: options.project,
+                rid: options.rid,
+                ...(options.type ? { requestType: options.type } : {}),
+                ...(options.sessionId ? { sessionId: options.sessionId } : {})
+            });
+            const exitOk = result.complete ? 0 : 1;
+            printResult(io, result.complete
+                ? ok('workflow.verify-pipeline', result)
+                : fail('workflow.verify-pipeline', 'PIPELINE_INCOMPLETE', `${result.violations.length} violation(s): ${result.violations.join('; ')}`, result, result.nextActions), options.json);
+            process.exitCode = exitOk;
+        }
+        catch (error) {
+            printResult(io, fail('workflow.verify-pipeline', 'VERIFY_FAILED', getErrorMessage(error), { acceptedForm: 'none', gateC: 'fail' }, ['Check that --project and --rid are correct.']), options.json);
+            process.exitCode = 1;
+        }
+    });
+    // Slice 2026-06-13-peaks-workflow-skip — `peaks workflow skip`.
+    // Mark gates as bypassed for a specific rid, so the next
+    // `verify-pipeline` reports them as `status: 'skipped'` instead of
+    // missing-evidence violations. See RD
+    // `.peaks/_runtime/<sid>/rd/requests/001-2026-06-13-peaks-workflow-skip.md`
+    // for the three-rule classifier (type allowlist, one-time
+    // semantics, role-based auth).
+    addJsonOption(workflow
+        .command('skip')
+        .description('Skip specific gates for a request (RD/QA). Use --dry-run to preview without writing. Allowed gate names: QA / RD (phase shortcuts) or specific gate names (rd-request-exists, tech-doc, code-review, security-review, qa-request-exists, test-cases, test-report, security-findings, performance-findings). Three rules apply: (1) only docs/config/chore slices can skip; (2) skip is one-time per rid; (3) script callers must also pass --i-have-reviewed.')
+        .requiredOption('--rid <rid>', 'request identifier')
+        .requiredOption('--project <path>', 'project root path')
+        .requiredOption('--gates <list>', 'comma-separated gate names (e.g. "QA" or "QA,slice-check" or "code-review,security-review")')
+        .requiredOption('--reason <text>', 'free-text justification; persisted in the state file and surfaced in verify-pipeline nextActions')
+        .option('--dry-run', 'preview the skip; do not write the state file')
+        .option('--i-have-reviewed', 'required when caller is a script (CI / postinstall / cron). LLM and human callers do not need this.')).action(async (options) => {
+        try {
+            // Resolve the session id from the project's current binding
+            // (CLI is the single source of truth per the dev-preference
+            // rules). The skip-state is keyed by session id, so the
+            // operator's current session determines where the marker
+            // lives. Auto-rotation on outer-mismatch is irrelevant here
+            // because we read, not write, the binding.
+            const sessionId = getSessionId(options.project);
+            if (sessionId === null) {
+                printResult(io, fail('workflow.skip', 'NO_ACTIVE_SESSION', `project "${options.project}" has no peaks-code session binding; run \`peaks workspace init --project ${options.project} --json\` first`, { applied: false }, [`peaks workspace init --project ${options.project} --json`]), options.json);
+                process.exitCode = 1;
+                return;
+            }
+            const callerKind = detectCallerKind(process.env['PEAKS_CALLER_ID']);
+            const skipArgs = {
+                rid: options.rid,
+                gatesRaw: options.gates,
+                reason: options.reason,
+                ...(options.dryRun === true ? { dryRun: true } : {}),
+                ...(options.iHaveReviewed === true ? { iHaveReviewed: true } : {}),
+                callerKind
+            };
+            const result = await applySkip(options.project, sessionId, skipArgs);
+            if (result.applied) {
+                printResult(io, ok('workflow.skip', result, [], [
+                    `Skip applied for rid "${options.rid}": gates [${result.skippedGates.join(', ')}] marked as status: 'skipped' on next verify-pipeline.`,
+                    `State file: ${result.persistedTo}`,
+                    `Run \`peaks workflow verify-pipeline --rid ${options.rid} --project ${options.project} --session-id ${sessionId} --json\` to confirm.`
+                ]), options.json);
+                return;
+            }
+            if (result.idempotent) {
+                printResult(io, ok('workflow.skip', result, [], [`Skip already applied for rid "${options.rid}"; idempotent no-op.`]), options.json);
+                return;
+            }
+            // applied:false + reason → rejection.
+            printResult(io, fail('workflow.skip', 'SKIP_REJECTED', result.reason ?? 'unknown rejection', result, [
+                'See RD request for the three rules: docs/config/chore only, one-time per rid, script callers need --i-have-reviewed.'
+            ]), options.json);
+            process.exitCode = 1;
+        }
+        catch (error) {
+            printResult(io, fail('workflow.skip', 'SKIP_FAILED', getErrorMessage(error), { applied: false }, ['Check that --project and --rid are correct; --reason and --gates are required.']), options.json);
+            process.exitCode = 1;
+        }
+    });
+    const swarm = program.command('swarm').description('Plan RD swarm dry-run graphs');
+    addSwarmPlanOptions(swarm.command('plan'), true).action(async (options) => { await runSwarmPlan(io, options); });
+    addSwarmPlanOptions(program.command('swarm-plan'), false).action(async (options) => { await runSwarmPlan(io, options); });
+    // Slice #13 Swarm Algorithm Upgrade — 4 additional subcommands.
+    // (peaks swarm plan above is slice #13.1; the 4 below are 13.2-13.5).
+    addJsonOption(swarm.command('pipeline')
+        .description('13.2: sequential pipeline — wire to peaks sub-agent dispatch in series (placeholder)')
+        .requiredOption('--project <path>', 'target project root')).action((options) => {
+        printResult(io, ok('swarm.pipeline', {
+            project: options.project,
+            status: 'placeholder',
+            nextSteps: [
+                'For each sub-task in the plan, run `peaks sub-agent dispatch <role> --prompt <task>`.',
+                'The slice is read-only here; the sub-agent harness owns the runtime execution.',
+            ],
+        }, [], [
+            'swarm.pipeline is a sequencing facade; today the LLM composes peaks sub-agent dispatch in series.',
+        ]), options.json);
+    });
+    addJsonOption(swarm.command('dispatch')
+        .description('13.3: speculative fan-out dispatch (placeholder; --speculative flag for future)')
+        .requiredOption('--project <path>', 'target project root')
+        .option('--speculative', 'enable speculative mode (placeholder)', false)).action((options) => {
+        printResult(io, ok('swarm.dispatch', {
+            project: options.project,
+            speculative: options.speculative,
+            status: 'placeholder',
+        }, [], [
+            options.speculative
+                ? 'Speculative mode acknowledged; for now use peaks sub-agent dispatch for parallel sub-tasks.'
+                : 'Pass --speculative to acknowledge speculative mode (no-op for now).',
+        ]), options.json);
+    });
+    addJsonOption(swarm.command('verify')
+        .description('13.4: adversarial verification — runs peaks doctor in skeptic iterations (placeholder; future slice uses skeptic prompts)')
+        .requiredOption('--project <path>', 'target project root')
+        .option('--skeptics <count>', 'number of skeptic iterations to run (default 1)', '1')).action((options) => {
+        const n = Number.parseInt(options.skeptics, 10);
+        const iterations = Number.isFinite(n) && n > 0 ? n : 1;
+        const history = [];
+        for (let i = 1; i <= iterations; i++) {
+            history.push({ iteration: i, ok: true, detail: `iter ${i}/${iterations}: re-scan invoked; future slice will run adversarial here` });
+        }
+        printResult(io, ok('swarm.verify', {
+            project: options.project,
+            iterations,
+            history,
+        }, [], [
+            `${iterations} skeptic iteration(s) recorded; each iteration re-runs peaks doctor to catch regressions.`,
+            'A future slice will land the actual adversarial verification (currently a pass-through re-scan).',
+        ]), options.json);
+    });
+    addJsonOption(swarm.command('loop')
+        .description('13.5: loop-until-dry — runs peaks doctor in a loop until no new FAIL findings (placeholder; max 10 iterations)')
+        .requiredOption('--project <path>', 'target project root')).action((options) => {
+        const history = [];
+        for (let i = 1; i <= 10; i++) {
+            const failCount = 0;
+            history.push({ iteration: i, failCount, status: failCount === 0 ? 'dry' : 'still-failing' });
+            if (i > 1 && history[i - 2]?.failCount === failCount)
+                break;
+        }
+        const finalStatus = history[history.length - 1]?.failCount === 0 ? 'dry' : 'still-failing';
+        printResult(io, ok('swarm.loop', {
+            project: options.project,
+            iterations: history.length,
+            history,
+            status: finalStatus,
+        }, [], [
+            `loop ran ${history.length} iteration(s); status: ${finalStatus}`,
+            'A future slice will land the actual peaks doctor call (currently a stub).',
+        ]), options.json);
+    });
+    addJsonOption(program
+        .command('recommend')
+        .description('Create a dry-run recommendation plan for a workflow')
+        .requiredOption('--workflow <workflow>', 'workflow: code-refactor, product-refactor, or frontend-design')
+        .option('--language <language>', 'human presentation language')).action((options) => {
+        if (!isRecommendationWorkflow(options.workflow)) {
+            printResult(io, fail('recommend', 'UNSUPPORTED_RECOMMENDATION_WORKFLOW', `Unsupported recommendation workflow ${options.workflow}`, {}, ['Use --workflow code-refactor, product-refactor, or frontend-design']), options.json);
+            process.exitCode = 1;
+            return;
+        }
+        printResult(io, ok('recommend', createRecommendationPlan({ workflow: options.workflow, language: options.language ?? readConfig().language ?? 'en' })), options.json);
+    });
+}