@aion0/forge 0.5.26 → 0.5.28

package/.forge/worktrees/pipeline-4dd8dc2d/lib/pipeline.ts

@@ -0,0 +1,1565 @@
+/**
+ * Pipeline Engine — DAG-based workflow orchestration on top of the Task system.
+ *
+ * Workflow YAML → Pipeline instance → Nodes executed as Tasks
+ * Supports: dependencies, output passing, conditional routing, parallel execution, notifications.
+ */
+
+import { randomUUID } from 'node:crypto';
+import { existsSync, readdirSync, readFileSync, writeFileSync, mkdirSync } from 'node:fs';
+import { join } from 'node:path';
+import YAML from 'yaml';
+import { createTask, getTask, onTaskEvent, taskModelOverrides } from './task-manager';
+import { getProjectInfo } from './projects';
+import { loadSettings } from './settings';
+import { getAgent, listAgents } from './agents';
+import type { Task } from '@/src/types';
+import { getDataDir } from './dirs';
+
+const PIPELINES_DIR = join(getDataDir(), 'pipelines');
+const WORKFLOWS_DIR = join(getDataDir(), 'flows');
+
+// Track pipeline task IDs so terminal notifications can skip them (persists across hot-reloads)
+const pipelineTaskKey = Symbol.for('mw-pipeline-task-ids');
+const gPipeline = globalThis as any;
+if (!gPipeline[pipelineTaskKey]) gPipeline[pipelineTaskKey] = new Set<string>();
+export const pipelineTaskIds: Set<string> = gPipeline[pipelineTaskKey];
+
+// ─── Types ────────────────────────────────────────────────
+
+export interface WorkflowNode {
+  id: string;
+  project: string;
+  prompt: string;
+  mode?: 'claude' | 'shell' | 'plugin';  // default: 'claude', 'shell' runs command, 'plugin' runs plugin action
+  agent?: string;              // agent ID (default: from settings)
+  branch?: string;             // auto checkout this branch before running (supports templates)
+  worktree?: boolean;          // default: true. Set false to skip worktree isolation (run in project dir directly)
+  // Plugin mode fields
+  plugin?: string;             // plugin ID (e.g., 'jenkins', 'docker')
+  pluginAction?: string;       // action name (e.g., 'trigger', 'build'), defaults to plugin's defaultAction
+  pluginParams?: Record<string, any>;  // per-use parameters
+  pluginWait?: boolean;        // auto-run 'wait' action after main action
+  dependsOn: string[];
+  outputs: { name: string; extract: 'result' | 'git_diff' | 'stdout' | 'plugin' }[];
+  routes: { condition: string; next: string }[];
+  maxIterations: number;
+}
+
+// ─── Conversation Mode Types ──────────────────────────────
+
+export interface ConversationAgent {
+  id: string;           // logical ID within this conversation (e.g., 'architect', 'implementer')
+  agent: string;        // agent registry ID (e.g., 'claude', 'codex', 'aider')
+  role: string;         // system prompt / role description
+  project?: string;     // project context (optional, defaults to workflow input.project)
+}
+
+export interface ConversationMessage {
+  round: number;
+  agentId: string;      // logical ID from ConversationAgent
+  agentName: string;    // display name (resolved from registry)
+  content: string;
+  timestamp: string;
+  taskId?: string;      // backing task ID
+  status: 'pending' | 'running' | 'done' | 'failed';
+}
+
+export interface ConversationConfig {
+  agents: ConversationAgent[];
+  maxRounds: number;           // max back-and-forth rounds
+  stopCondition?: string;      // e.g., "all agents say DONE", "any agent says DONE"
+  initialPrompt: string;       // the seed prompt to kick off the conversation
+  contextStrategy?: 'full' | 'window' | 'summary';  // how to pass history, default: 'summary'
+  contextWindow?: number;      // for 'window'/'summary': how many recent messages to include in full (default: 4)
+  maxContentLength?: number;   // truncate each message to this length (default: 3000)
+}
+
+// ─── Workflow ─────────────────────────────────────────────
+
+export interface Workflow {
+  name: string;
+  type?: 'dag' | 'conversation';  // default: 'dag'
+  description?: string;
+  vars: Record<string, string>;
+  input: Record<string, string>;  // required input fields
+  nodes: Record<string, WorkflowNode>;
+  // Conversation mode fields (only when type === 'conversation')
+  conversation?: ConversationConfig;
+}
+
+export type PipelineNodeStatus = 'pending' | 'running' | 'done' | 'failed' | 'skipped';
+
+export interface PipelineNodeState {
+  status: PipelineNodeStatus;
+  taskId?: string;
+  outputs: Record<string, string>;
+  iterations: number;
+  startedAt?: string;
+  completedAt?: string;
+  error?: string;
+}
+
+export interface Pipeline {
+  id: string;
+  workflowName: string;
+  type?: 'dag' | 'conversation';  // default: 'dag'
+  status: 'running' | 'done' | 'failed' | 'cancelled';
+  input: Record<string, string>;
+  vars: Record<string, string>;
+  nodes: Record<string, PipelineNodeState>;
+  nodeOrder: string[];  // for UI display
+  createdAt: string;
+  completedAt?: string;
+  // Conversation mode state
+  conversation?: {
+    config: ConversationConfig;
+    messages: ConversationMessage[];
+    currentRound: number;
+    currentAgentIndex: number;  // index into config.agents
+  };
+}
+
+// ─── Workflow Loading ─────────────────────────────────────
+
+// ─── Built-in workflows ──────────────────────────────────
+
+export const BUILTIN_WORKFLOWS: Record<string, string> = {
+  'issue-fix-and-review': `
+name: issue-fix-and-review
+description: "Fetch GitHub issue → fix code → create PR → review PR → notify"
+input:
+  issue_id: "GitHub issue number"
+  project: "Project name"
+  base_branch: "Base branch (default: auto-detect)"
+  extra_context: "Additional instructions for the fix (optional)"
+nodes:
+  setup:
+    mode: shell
+    project: "{{input.project}}"
+    prompt: |
+      cd "$(git rev-parse --show-toplevel)" && \
+      if [ -n "$(git status --porcelain)" ]; then echo "ERROR: Working directory has uncommitted changes. Please commit or stash first." && exit 1; fi && \
+      ORIG_BRANCH=$(git branch --show-current || git rev-parse --short HEAD) && \
+      REPO=$(gh repo view --json nameWithOwner -q .nameWithOwner 2>/dev/null || git remote get-url origin | sed 's/.*github.com[:/]//;s/.git$//') && \
+      BASE="{{input.base_branch}}" && \
+      if [ -z "$BASE" ] || [ "$BASE" = "auto-detect" ]; then BASE=$(git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null | sed 's@^refs/remotes/origin/@@' || echo main); fi && \
+      git checkout "$BASE" 2>/dev/null || true && \
+      git pull origin "$BASE" 2>/dev/null || true && \
+      OLD_BRANCH=$(git branch --list "fix/{{input.issue_id}}-*" | head -1 | tr -d ' *') && \
+      if [ -n "$OLD_BRANCH" ]; then git branch -D "$OLD_BRANCH" 2>/dev/null || true; fi && \
+      echo "REPO=$REPO" && echo "BASE=$BASE" && echo "ORIG_BRANCH=$ORIG_BRANCH"
+    outputs:
+      - name: info
+        extract: stdout
+  fetch-issue:
+    mode: shell
+    project: "{{input.project}}"
+    depends_on: [setup]
+    prompt: |
+      ISSUE_ID="{{input.issue_id}}" && \
+      if [ -z "$ISSUE_ID" ]; then echo "__SKIP__ No issue_id provided" && exit 0; fi && \
+      SETUP_INFO=$'{{nodes.setup.outputs.info}}' && \
+      REPO=$(echo "$SETUP_INFO" | grep REPO= | cut -d= -f2) && \
+      gh issue view "$ISSUE_ID" --json title,body,labels,number -R "$REPO"
+    outputs:
+      - name: issue_json
+        extract: stdout
+  fix-code:
+    project: "{{input.project}}"
+    depends_on: [fetch-issue]
+    prompt: |
+      A GitHub issue needs to be fixed. Here is the issue data:
+
+      {{nodes.fetch-issue.outputs.issue_json}}
+
+      Steps:
+      1. Create a new branch from the current branch (which is already on the base). Name format: fix/{{input.issue_id}}-<short-description> (e.g. fix/3-add-validation, fix/15-null-pointer). Any old branch for this issue has been cleaned up.
+      2. Analyze the issue and fix the code.
+      3. Stage and commit with a message referencing #{{input.issue_id}}.
+
+      Base branch info: {{nodes.setup.outputs.info}}
+
+      Additional context from user: {{input.extra_context}}
+    outputs:
+      - name: summary
+        extract: result
+      - name: diff
+        extract: git_diff
+  push-and-pr:
+    mode: shell
+    project: "{{input.project}}"
+    depends_on: [fix-code]
+    prompt: |
+      SETUP_INFO=$'{{nodes.setup.outputs.info}}' && \
+      REPO=$(echo "$SETUP_INFO" | grep REPO= | cut -d= -f2) && \
+      BRANCH=$(git branch --show-current) && \
+      git push -u origin "$BRANCH" --force-with-lease 2>&1 && \
+      PR_URL=$(gh pr create --title "Fix #{{input.issue_id}}" \
+        --body "Auto-fix by Forge Pipeline for issue #{{input.issue_id}}." -R "$REPO" 2>/dev/null || \
+        gh pr view "$BRANCH" --json url -q .url -R "$REPO" 2>/dev/null) && \
+      echo "$PR_URL"
+    outputs:
+      - name: pr_url
+        extract: stdout
+  review:
+    project: "{{input.project}}"
+    depends_on: [push-and-pr]
+    prompt: |
+      Review the code changes for issue #{{input.issue_id}}.
+
+      Fix summary: {{nodes.fix-code.outputs.summary}}
+
+      Git diff:
+      {{nodes.fix-code.outputs.diff}}
+
+      Check for:
+      - Bugs and logic errors
+      - Security vulnerabilities
+      - Performance issues
+      - Whether the fix actually addresses the issue
+
+      Respond with:
+      1. APPROVED or CHANGES_REQUESTED
+      2. Specific issues found with file paths and line numbers
+    outputs:
+      - name: review_result
+        extract: result
+  cleanup:
+    mode: shell
+    project: "{{input.project}}"
+    depends_on: [review]
+    prompt: |
+      SETUP_INFO=$'{{nodes.setup.outputs.info}}' && \
+      ORIG=$(echo "$SETUP_INFO" | grep ORIG_BRANCH= | cut -d= -f2) && \
+      PR_URL=$'{{nodes.push-and-pr.outputs.pr_url}}' && \
+      if [ -n "$(git status --porcelain)" ]; then
+        echo "Issue #{{input.issue_id}} — PR: $PR_URL (staying on $(git branch --show-current))"
+      else
+        git checkout "$ORIG" 2>/dev/null || true
+        echo "Issue #{{input.issue_id}} — PR: $PR_URL (switched back to $ORIG)"
+      fi
+    outputs:
+      - name: result
+        extract: stdout
+`,
+  'multi-agent-collaboration': `
+name: multi-agent-collaboration
+type: conversation
+description: "Two agents collaborate: one designs, one implements"
+input:
+  project: "Project name"
+  task: "What to build or fix"
+agents:
+  - id: architect
+    agent: claude
+    role: "You are a software architect. Round 1: design the solution with clear steps. Later rounds: review the implementation and say DONE if satisfied."
+  - id: implementer
+    agent: claude
+    role: "You are a developer. Implement what the architect designs. After implementing, say DONE."
+max_rounds: 3
+stop_condition: "both agents say DONE"
+initial_prompt: "Task: {{input.task}}"
+`,
+  'review-mr': `
+name: review-mr
+description: "Review PR — AI code review with GitHub comment"
+input:
+  project: "Project name"
+  branch: "Branch name or PR number (empty = auto-detect latest open PR)"
+  base_branch: "Target branch (default: main)"
+vars:
+  default_base: main
+nodes:
+  resolve-pr:
+    mode: shell
+    project: "{{input.project}}"
+    worktree: false
+    prompt: |
+      INPUT_BRANCH="{{input.branch}}" && \\
+      BASE="{{input.base_branch}}" && \\
+      if [ -z "$BASE" ] || echo "$BASE" | grep -q '{{'; then BASE="main"; fi && \\
+      if [ -z "$INPUT_BRANCH" ] || echo "$INPUT_BRANCH" | grep -q '{{'; then \\
+        INPUT_BRANCH=$(gh pr list --state open --base "$BASE" --json number -q '.[0].number' 2>/dev/null); \\
+        if [ -z "$INPUT_BRANCH" ]; then echo "ERROR: No open PR found targeting $BASE" && exit 1; fi; \\
+      fi && \\
+      if echo "$INPUT_BRANCH" | grep -qE '^[0-9]+$'; then \\
+        PR_NUM="$INPUT_BRANCH"; \\
+      else \\
+        PR_NUM=$(gh pr list --state open --head "$INPUT_BRANCH" --json number -q '.[0].number' 2>/dev/null); \\
+        if [ -z "$PR_NUM" ]; then echo "ERROR: No open PR for branch $INPUT_BRANCH" && exit 1; fi; \\
+      fi && \\
+      echo "$PR_NUM"
+    outputs:
+      - name: pr_number
+        extract: stdout
+  fetch-diff:
+    mode: shell
+    project: "{{input.project}}"
+    worktree: false
+    depends_on: [resolve-pr]
+    prompt: "gh pr diff {{nodes.resolve-pr.outputs.pr_number}}"
+    outputs:
+      - name: diff
+        extract: stdout
+  fetch-files:
+    mode: shell
+    project: "{{input.project}}"
+    worktree: false
+    depends_on: [resolve-pr]
+    prompt: |
+      PR_NUM="{{nodes.resolve-pr.outputs.pr_number}}" && \\
+      echo "=== PR #$PR_NUM ===" && \\
+      gh pr view "$PR_NUM" --json title,author,additions,deletions,changedFiles,commits,body --jq '"Title: " + .title + "\\nAuthor: " + .author.login + "\\nFiles: " + (.changedFiles|tostring) + " changed, +" + (.additions|tostring) + "/-" + (.deletions|tostring) + "\\nCommits: " + (.commits|length|tostring) + "\\n\\n=== PR Description ===\\n" + (.body // "(no description)")' && \\
+      echo "" && \\
+      echo "=== Changed Files ===" && \\
+      gh pr diff "$PR_NUM" --name-only
+    outputs:
+      - name: stats
+        extract: stdout
+  review:
+    project: "{{input.project}}"
+    worktree: false
+    depends_on: [fetch-diff, fetch-files, resolve-pr]
+    prompt: |
+      You are a senior code reviewer. Perform a thorough code review of this PR.
+
+      ## PR Info & Description
+      {{nodes.fetch-files.outputs.stats}}
+
+      ## Diff
+      {{nodes.fetch-diff.outputs.diff}}
+
+      ## Review Requirements
+
+      **First**: Verify the PR description against actual changes:
+      - Is every claimed change actually implemented?
+      - Any claimed changes that are NOT in the diff?
+      - Any changes in the diff NOT mentioned in the description?
+
+      **Then**: Review code quality:
+      1. Bug risk — logic errors, edge cases, null references
+      2. Security — injection, hardcoded secrets, sensitive data exposure
+      3. Performance — obvious bottlenecks
+      4. Code quality — readability, naming, DRY
+
+      ## Output
+
+      Write the full review report to /tmp/forge-review-pr{{nodes.resolve-pr.outputs.pr_number}}.md in this format:
+
+      ## 🤖 Forge AI Code Review — PR #{{nodes.resolve-pr.outputs.pr_number}}
+
+      ### 📋 Summary
+      - Verdict: ✅ Approve / ⚠️ Request Changes / ❌ Reject
+      - One-line summary
+
+      ### ✅ PR Description Verification
+      List each change claimed in the PR description, mark ✓ implemented / ✗ not implemented / ⚠️ partial
+
+      ### 🔴 Blockers (must fix)
+      (write "None" if none)
+
+      ### 🟡 Suggestions
+
+      ### 🟢 Nice-to-have
+
+      ### 💡 Highlights
+
+      ---
+      _Generated by [Forge](https://github.com/aiwatching/forge) Pipeline_
+
+      **You MUST write the complete report to /tmp/forge-review-pr{{nodes.resolve-pr.outputs.pr_number}}.md. This is the most important step.**
+    outputs:
+      - name: report
+        extract: result
+  post-comment:
+    mode: shell
+    project: "{{input.project}}"
+    worktree: false
+    depends_on: [review, resolve-pr]
+    prompt: |
+      PR_NUM="{{nodes.resolve-pr.outputs.pr_number}}" && \\
+      REPORT="/tmp/forge-review-pr\${PR_NUM}.md" && \\
+      if [ ! -f "$REPORT" ]; then echo "ERROR: Review report not found at $REPORT" && exit 1; fi && \\
+      gh pr comment "$PR_NUM" --body-file "$REPORT" && \\
+      rm -f "$REPORT" && \\
+      echo "Comment posted to PR #$PR_NUM"
+    outputs:
+      - name: result
+        extract: stdout
+`,
+};
+
+export interface WorkflowWithMeta extends Workflow {
+  builtin?: boolean;
+}
+
+export function listWorkflows(): WorkflowWithMeta[] {
+  // User workflows
+  const userWorkflows: WorkflowWithMeta[] = [];
+  if (existsSync(WORKFLOWS_DIR)) {
+    for (const f of readdirSync(WORKFLOWS_DIR).filter(f => f.endsWith('.yaml') || f.endsWith('.yml'))) {
+      try {
+        userWorkflows.push({ ...parseWorkflow(readFileSync(join(WORKFLOWS_DIR, f), 'utf-8')), builtin: false });
+      } catch {}
+    }
+  }
+
+  // Built-in workflows (don't override user ones with same name)
+  const userNames = new Set(userWorkflows.map(w => w.name));
+  const builtins: WorkflowWithMeta[] = [];
+  for (const [, yaml] of Object.entries(BUILTIN_WORKFLOWS)) {
+    try {
+      const w = parseWorkflow(yaml);
+      if (!userNames.has(w.name)) {
+        builtins.push({ ...w, builtin: true });
+      }
+    } catch {}
+  }
+
+  return [...builtins, ...userWorkflows];
+}
+
+export function getWorkflow(name: string): WorkflowWithMeta | null {
+  return listWorkflows().find(w => w.name === name) || null;
+}
+
+function parseWorkflow(raw: string): Workflow {
+  const parsed = YAML.parse(raw);
+  const workflowType = parsed.type || 'dag';
+  const nodes: Record<string, WorkflowNode> = {};
+
+  for (const [id, def] of Object.entries(parsed.nodes || {})) {
+    const n = def as any;
+    nodes[id] = {
+      id,
+      project: n.project || '',
+      prompt: n.prompt || '',
+      mode: n.mode || (n.plugin ? 'plugin' : 'claude'),
+      agent: n.agent || undefined,
+      branch: n.branch || undefined,
+      worktree: n.worktree !== undefined ? n.worktree : undefined,
+      plugin: n.plugin || undefined,
+      pluginAction: n.plugin_action || n.pluginAction || undefined,
+      pluginParams: n.plugin_params || n.pluginParams || n.params || undefined,
+      pluginWait: n.plugin_wait || n.pluginWait || n.wait || false,
+      dependsOn: n.depends_on || n.dependsOn || [],
+      outputs: (n.outputs || []).map((o: any) => ({
+        name: o.name,
+        extract: o.extract || 'result',
+      })),
+      routes: (n.routes || []).map((r: any) => ({
+        condition: r.condition || 'default',
+        next: r.next,
+      })),
+      maxIterations: n.max_iterations || n.maxIterations || 3,
+    };
+  }
+
+  // Parse conversation config
+  let conversation: ConversationConfig | undefined;
+  if (workflowType === 'conversation' && parsed.agents) {
+    conversation = {
+      agents: (parsed.agents as any[]).map((a: any) => ({
+        id: a.id,
+        agent: a.agent || 'claude',
+        role: a.role || '',
+        project: a.project || undefined,
+      })),
+      maxRounds: parsed.max_rounds || parsed.maxRounds || 10,
+      stopCondition: parsed.stop_condition || parsed.stopCondition || undefined,
+      initialPrompt: parsed.initial_prompt || parsed.initialPrompt || '',
+      contextStrategy: parsed.context_strategy || parsed.contextStrategy || 'summary',
+      contextWindow: parsed.context_window || parsed.contextWindow || 4,
+      maxContentLength: parsed.max_content_length || parsed.maxContentLength || 3000,
+    };
+  }
+
+  return {
+    name: parsed.name || 'unnamed',
+    type: workflowType,
+    description: parsed.description,
+    vars: parsed.vars || {},
+    input: parsed.input || {},
+    nodes,
+    conversation,
+  };
+}
+
+// ─── Pipeline Persistence ─────────────────────────────────
+
+function ensureDir() {
+  if (!existsSync(PIPELINES_DIR)) mkdirSync(PIPELINES_DIR, { recursive: true });
+}
+
+function savePipeline(pipeline: Pipeline) {
+  ensureDir();
+  writeFileSync(join(PIPELINES_DIR, `${pipeline.id}.json`), JSON.stringify(pipeline, null, 2));
+}
+
+export function getPipeline(id: string): Pipeline | null {
+  try {
+    return JSON.parse(readFileSync(join(PIPELINES_DIR, `${id}.json`), 'utf-8'));
+  } catch {
+    return null;
+  }
+}
+
+export function deletePipeline(id: string): boolean {
+  const filePath = join(PIPELINES_DIR, `${id}.json`);
+  try {
+    if (existsSync(filePath)) {
+      const { unlinkSync } = require('node:fs');
+      unlinkSync(filePath);
+      return true;
+    }
+  } catch {}
+  return false;
+}
+
+export function listPipelines(): Pipeline[] {
+  ensureDir();
+  return readdirSync(PIPELINES_DIR)
+    .filter(f => f.endsWith('.json'))
+    .map(f => {
+      try {
+        return JSON.parse(readFileSync(join(PIPELINES_DIR, f), 'utf-8')) as Pipeline;
+      } catch {
+        return null;
+      }
+    })
+    .filter(Boolean) as Pipeline[];
+}
+
+// ─── Template Resolution ──────────────────────────────────
+
+/** Escape a string for safe embedding in single-quoted shell strings */
+function shellEscape(s: string): string {
+  // Replace single quotes with '\'' (end quote, escaped quote, start quote)
+  return s.replace(/'/g, "'\\''");
+}
+
+/** Escape a string for safe embedding in $'...' shell strings (ANSI-C quoting) */
+function shellEscapeAnsiC(s: string): string {
+  return s
+    .replace(/\\/g, '\\\\')
+    .replace(/'/g, "\\'")
+    .replace(/\n/g, '\\n')
+    .replace(/\r/g, '\\r')
+    .replace(/\t/g, '\\t');
+}
+
+function resolveTemplate(template: string, ctx: {
+  input: Record<string, string>;
+  vars: Record<string, string>;
+  nodes: Record<string, PipelineNodeState>;
+}, shellMode?: boolean): string {
+  return template.replace(/\{\{(.*?)\}\}/g, (_, expr) => {
+    const path = expr.trim();
+    let value = '';
+
+    // {{input.xxx}}
+    if (path.startsWith('input.')) value = ctx.input[path.slice(6)] || '';
+    // {{vars.xxx}}
+    else if (path.startsWith('vars.')) value = ctx.vars[path.slice(5)] || '';
+    // {{nodes.xxx.outputs.yyy}}
+    else {
+      const nodeMatch = path.match(/^nodes\.([\w-]+)\.outputs\.([\w-]+)$/);
+      if (nodeMatch) {
+        const [, nodeId, outputName] = nodeMatch;
+        value = ctx.nodes[nodeId]?.outputs[outputName] || '';
+      } else {
+        return `{{${path}}}`;
+      }
+    }
+
+    return shellMode ? shellEscapeAnsiC(value) : value;
+  });
+}
+
+// ─── Project-level pipeline lock ─────────────────────────
+const projectPipelineLocks = new Map<string, string>(); // projectPath → pipelineId
+
+function acquireProjectLock(projectPath: string, pipelineId: string): boolean {
+  const existing = projectPipelineLocks.get(projectPath);
+  if (existing && existing !== pipelineId) {
+    // Check if the existing pipeline is still running
+    const p = getPipeline(existing);
+    if (p && p.status === 'running') return false;
+    // Stale lock, clear it
+  }
+  projectPipelineLocks.set(projectPath, pipelineId);
+  return true;
+}
+
+function releaseProjectLock(projectPath: string, pipelineId: string) {
+  if (projectPipelineLocks.get(projectPath) === pipelineId) {
+    projectPipelineLocks.delete(projectPath);
+  }
+}
+
+// ─── Pipeline Execution ───────────────────────────────────
+
+export function startPipeline(workflowName: string, input: Record<string, string>): Pipeline {
+  const workflow = getWorkflow(workflowName);
+  if (!workflow) throw new Error(`Workflow not found: ${workflowName}`);
+
+  // Conversation mode — separate execution path
+  if (workflow.type === 'conversation' && workflow.conversation) {
+    return startConversationPipeline(workflow, input);
+  }
+
+  const id = randomUUID().slice(0, 8);
+  const nodes: Record<string, PipelineNodeState> = {};
+  const nodeOrder = topologicalSort(workflow.nodes);
+
+  for (const nodeId of nodeOrder) {
+    nodes[nodeId] = {
+      status: 'pending',
+      outputs: {},
+      iterations: 0,
+    };
+  }
+
+  const pipeline: Pipeline = {
+    id,
+    workflowName,
+    status: 'running',
+    input,
+    vars: { ...workflow.vars },
+    nodes,
+    nodeOrder,
+    createdAt: new Date().toISOString(),
+  };
+
+  savePipeline(pipeline);
+
+  // Start nodes that have no dependencies
+  scheduleReadyNodes(pipeline, workflow);
+
+  // Listen for task completions
+  setupTaskListener(pipeline.id);
+
+  return pipeline;
+}
+
+// ─── Conversation State Type (extracted to avoid Turbopack parse issues) ──
+type ConversationState = {
+  config: ConversationConfig;
+  messages: ConversationMessage[];
+  currentRound: number;
+  currentAgentIndex: number;
+};
+
+// ─── Conversation Mode Execution ──────────────────────────
+
+function startConversationPipeline(workflow: Workflow, input: Record<string, string>): Pipeline {
+  const conv = workflow.conversation!;
+  const id = randomUUID().slice(0, 8);
+
+  // Resolve agent display names
+  const agentNames: Record<string, string> = {};
+  const allAgents = listAgents();
+  for (const ca of conv.agents) {
+    const found = allAgents.find(a => a.id === ca.agent);
+    agentNames[ca.id] = found?.name || ca.agent;
+  }
+
+  const pipeline: Pipeline = {
+    id,
+    workflowName: workflow.name,
+    type: 'conversation',
+    status: 'running',
+    input,
+    vars: { ...workflow.vars },
+    nodes: {},
+    nodeOrder: [],
+    createdAt: new Date().toISOString(),
+    conversation: {
+      config: {
+        ...conv,
+        // Store resolved initial prompt so buildConversationContext uses it
+        initialPrompt: conv.initialPrompt.replace(/\{\{input\.(\w+)\}\}/g, (_, key) => input[key] || ''),
+      },
+      messages: [],
+      currentRound: 1,
+      currentAgentIndex: 0,
+    },
+  };
+
+  savePipeline(pipeline);
+
+  const resolvedPrompt = pipeline.conversation!.config.initialPrompt;
+
+  // Start the first round
+  scheduleNextConversationTurn(pipeline, resolvedPrompt, agentNames);
+
+  return pipeline;
+}
+
+function scheduleNextConversationTurn(pipeline: Pipeline, contextForAgent: string, agentNames?: Record<string, string>) {
+  const conv = pipeline.conversation!;
+  const config = conv.config;
+  const agentDef = config.agents[conv.currentAgentIndex];
+
+  if (!agentDef) {
+    pipeline.status = 'failed';
+    pipeline.completedAt = new Date().toISOString();
+    savePipeline(pipeline);
+    return;
+  }
+
+  // Resolve project
+  const projectName = agentDef.project || pipeline.input.project || '';
+  const projectInfo = getProjectInfo(projectName);
+  if (!projectInfo) {
+    pipeline.status = 'failed';
+    pipeline.completedAt = new Date().toISOString();
+    savePipeline(pipeline);
+    notifyPipelineComplete(pipeline);
+    return;
+  }
+
+  // Build the prompt: role context + conversation history + new message
+  const rolePrefix = agentDef.role ? `[Your role: ${agentDef.role}]\n\n` : '';
+  const fullPrompt = `${rolePrefix}${contextForAgent}`;
+
+  // Create a task for this agent's turn
+  const task = createTask({
+    projectName: projectInfo.name,
+    projectPath: projectInfo.path,
+    prompt: fullPrompt,
+    mode: 'prompt',
+    agent: agentDef.agent,
+    conversationId: '', // fresh session — no resume for conversation mode
+  });
+  pipelineTaskIds.add(task.id);
+
+  // Add pending message
+  const names = agentNames || resolveAgentNames(config.agents);
+  conv.messages.push({
+    round: conv.currentRound,
+    agentId: agentDef.id,
+    agentName: names[agentDef.id] || agentDef.agent,
+    content: '',
+    timestamp: new Date().toISOString(),
+    taskId: task.id,
+    status: 'running',
+  });
+
+  savePipeline(pipeline);
+
+  // Listen for this task to complete
+  setupConversationTaskListener(pipeline.id, task.id);
+}
+
+function resolveAgentNames(agents: ConversationAgent[]): Record<string, string> {
+  const allAgents = listAgents();
+  const names: Record<string, string> = {};
+  for (const ca of agents) {
+    const found = allAgents.find(a => a.id === ca.agent);
+    names[ca.id] = found?.name || ca.agent;
+  }
+  return names;
+}
+
+function setupConversationTaskListener(pipelineId: string, taskId: string) {
+  const cleanup = onTaskEvent((evtTaskId, event, data) => {
+    if (evtTaskId !== taskId) return;
+    if (event !== 'status') return;
+    if (data !== 'done' && data !== 'failed') return;
+
+    cleanup(); // one-shot listener
+
+    const pipeline = getPipeline(pipelineId);
+    if (!pipeline || pipeline.status !== 'running' || !pipeline.conversation) return;
+
+    const conv = pipeline.conversation;
+    const config = conv.config;
+    const msgIndex = conv.messages.findIndex(m => m.taskId === taskId);
+    if (msgIndex < 0) return;
+
+    const task = getTask(taskId);
+
+    if (data === 'failed' || !task) {
+      conv.messages[msgIndex].status = 'failed';
+      conv.messages[msgIndex].content = task?.error || 'Task failed';
+      pipeline.status = 'failed';
+      pipeline.completedAt = new Date().toISOString();
+      savePipeline(pipeline);
+      notifyPipelineComplete(pipeline);
+      return;
+    }
+
+    // Task completed — extract response
+    const response = task.resultSummary || '';
+    conv.messages[msgIndex].status = 'done';
+    conv.messages[msgIndex].content = response;
+
+    // Check stop condition
+    if (checkConversationStopCondition(conv, response)) {
+      finishConversation(pipeline, 'done');
+      return;
+    }
+
+    // Move to next agent in round, or next round
+    conv.currentAgentIndex++;
+    if (conv.currentAgentIndex >= config.agents.length) {
+      // Completed a full round
+      conv.currentAgentIndex = 0;
+      conv.currentRound++;
+
+      if (conv.currentRound > config.maxRounds) {
+        finishConversation(pipeline, 'done');
+        return;
+      }
+    }
+
+    savePipeline(pipeline);
+
+    // Build context for next agent: accumulate conversation history
+    const contextForNext = buildConversationContext(conv);
+    scheduleNextConversationTurn(pipeline, contextForNext);
+  });
+}
+
+/**
+ * Build context string for the next agent in conversation.
+ *
+ * Three strategies:
+ *   - 'full'    : pass ALL history as-is (token-heavy, good for short convos)
+ *   - 'window'  : pass only the last N messages in full, drop older ones
+ *   - 'summary' : pass older messages as one-line summaries + last N in full (default)
+ */
+function buildConversationContext(conv: ConversationState): string {
+  const config = conv.config;
+  const strategy = config.contextStrategy || 'summary';
+  const windowSize = config.contextWindow || 4;
+  const maxLen = config.maxContentLength || 3000;
+
+  const doneMessages = conv.messages.filter(m => m.status === 'done' && m.content);
+
+  let context = `[Conversation — Round ${conv.currentRound}]\n\n`;
+  context += `Task: ${config.initialPrompt}\n\n`;
+
+  if (doneMessages.length === 0) {
+    context += `--- Your Turn ---\nYou are the first to respond. Please address the task above. If you believe the task is complete, include "DONE" in your response.`;
+    return context;
+  }
+
+  context += `--- Conversation History ---\n\n`;
+
+  if (strategy === 'full') {
+    // Full: all messages, truncated per maxLen
+    for (const msg of doneMessages) {
+      context += formatMessage(msg, config, maxLen);
+    }
+  } else if (strategy === 'window') {
+    // Window: only last N messages
+    const recent = doneMessages.slice(-windowSize);
+    if (doneMessages.length > windowSize) {
+      context += `[... ${doneMessages.length - windowSize} earlier messages omitted ...]\n\n`;
+    }
+    for (const msg of recent) {
+      context += formatMessage(msg, config, maxLen);
+    }
+  } else {
+    // Summary (default): older messages as one-line summaries, recent in full
+    const cutoff = doneMessages.length - windowSize;
+    if (cutoff > 0) {
+      context += `[Previous rounds summary]\n`;
+      for (let i = 0; i < cutoff; i++) {
+        const msg = doneMessages[i];
+        const summary = extractSummaryLine(msg.content);
+        context += `  R${msg.round} ${msg.agentName}: ${summary}\n`;
+      }
+      context += `\n`;
+    }
+    // Recent messages in full
+    const recent = doneMessages.slice(Math.max(0, cutoff));
+    for (const msg of recent) {
+      context += formatMessage(msg, config, maxLen);
+    }
+  }
+
+  context += `--- Your Turn ---\nRespond based on the conversation above. If you believe the task is complete, include "DONE" in your response.`;
+  return context;
+}
+
+function formatMessage(msg: ConversationMessage, config: ConversationConfig, maxLen: number): string {
+  const agentDef = config.agents.find(a => a.id === msg.agentId);
+  const label = `${msg.agentName} (${agentDef?.id || '?'})`;
+  const content = msg.content.length > maxLen
+    ? msg.content.slice(0, maxLen) + '\n[... truncated]'
+    : msg.content;
+  return `[${label} — Round ${msg.round}]:\n${content}\n\n`;
+}
+
+/** Extract a one-line summary from agent output (first meaningful line or first 120 chars) */
+function extractSummaryLine(content: string): string {
+  const lines = content.split('\n').map(l => l.trim()).filter(l => l.length > 10);
+  const first = lines[0] || content.slice(0, 120);
+  return first.length > 120 ? first.slice(0, 117) + '...' : first;
+}
+
+function checkConversationStopCondition(conv: ConversationState, latestResponse: string): boolean {
+  const condition = conv.config.stopCondition;
+  if (!condition) return false;
+
+  const lower = condition.toLowerCase();
+
+  // "any agent says DONE"
+  if (lower.includes('any') && lower.includes('done')) {
+    return latestResponse.toUpperCase().includes('DONE');
+  }
+
+  // "all agents say DONE" / "both agents say DONE"
+  if ((lower.includes('all') || lower.includes('both')) && lower.includes('done')) {
+    // Only check messages from the CURRENT round — don't mix rounds
+    const currentRound = conv.currentRound;
+    const agentIds = conv.config.agents.map(a => a.id);
+    const roundMessages = new Map<string, string>();
+    for (const msg of conv.messages) {
+      if (msg.status === 'done' && msg.round === currentRound && msg.agentId !== 'user') {
+        roundMessages.set(msg.agentId, msg.content);
+      }
+    }
+    // All agents in this round must have responded AND said DONE
+    return agentIds.every(id => {
+      const content = roundMessages.get(id);
+      return content && content.toUpperCase().includes('DONE');
+    });
+  }
+
+  // Default: check if latest response contains DONE
+  return latestResponse.toUpperCase().includes('DONE');
+}
+
+/** Cleanly finish a conversation — cancel any still-running tasks, mark messages */
+function finishConversation(pipeline: Pipeline, status: 'done' | 'failed') {
+  const conv = pipeline.conversation!;
+  for (const msg of conv.messages) {
+    if (msg.status === 'running' && msg.taskId) {
+      // Cancel the running task
+      try { const { cancelTask } = require('./task-manager'); cancelTask(msg.taskId); } catch {}
+      msg.status = status === 'done' ? 'done' : 'failed';
+      if (!msg.content) msg.content = status === 'done' ? '(conversation ended)' : '(conversation failed)';
+    }
+    if (msg.status === 'pending') {
+      msg.status = 'failed';
+    }
+  }
+  pipeline.status = status;
+  pipeline.completedAt = new Date().toISOString();
+  savePipeline(pipeline);
+  notifyPipelineComplete(pipeline);
+}
+
+/** Cancel a conversation pipeline */
+export function cancelConversation(pipelineId: string): boolean {
+  const pipeline = getPipeline(pipelineId);
+  if (!pipeline || pipeline.status !== 'running' || !pipeline.conversation) return false;
+
+  // Cancel any running task
+  for (const msg of pipeline.conversation.messages) {
+    if (msg.status === 'running' && msg.taskId) {
+      const { cancelTask } = require('./task-manager');
+      cancelTask(msg.taskId);
+    }
+    if (msg.status === 'pending') msg.status = 'failed';
+  }
+
+  pipeline.status = 'cancelled';
+  pipeline.completedAt = new Date().toISOString();
+  savePipeline(pipeline);
+  return true;
+}
+
+/**
+ * Inject a user message into a running conversation.
+ * Waits for current agent to finish, then sends the injected message
+ * as additional context to the specified agent on the next turn.
+ */
+export function injectConversationMessage(pipelineId: string, targetAgentId: string, message: string): boolean {
+  const pipeline = getPipeline(pipelineId);
+  if (!pipeline || pipeline.status !== 'running' || !pipeline.conversation) {
+    throw new Error('Pipeline not running or not a conversation');
+  }
+
+  const conv = pipeline.conversation;
+  const agentDef = conv.config.agents.find(a => a.id === targetAgentId);
+  if (!agentDef) throw new Error(`Agent not found: ${targetAgentId}`);
+
+  // Add a "user" message to the conversation
+  conv.messages.push({
+    round: conv.currentRound,
+    agentId: 'user',
+    agentName: 'Operator',
+    content: `[@${targetAgentId}] ${message}`,
+    timestamp: new Date().toISOString(),
+    status: 'done',
+  });
+
+  savePipeline(pipeline);
+
+  // If no agent is currently running, immediately schedule the target agent
+  const hasRunning = conv.messages.some(m => m.status === 'running');
+  if (!hasRunning) {
+    // Point to the target agent for next turn
+    const targetIdx = conv.config.agents.findIndex(a => a.id === targetAgentId);
+    if (targetIdx >= 0) {
+      conv.currentAgentIndex = targetIdx;
+      savePipeline(pipeline);
+      const context = buildConversationContext(conv);
+      scheduleNextConversationTurn(pipeline, context);
+    }
+  }
+  // If an agent IS running, the injected message will be included in the next context build
+
+  return true;
+}
+
+// ─── Conversation Recovery ────────────────────────────────
+
+function recoverConversationPipeline(pipeline: Pipeline) {
+  const conv = pipeline.conversation!;
+  const runningMsg = conv.messages.find(m => m.status === 'running');
+  if (!runningMsg || !runningMsg.taskId) return;
+
+  const task = getTask(runningMsg.taskId);
+  if (!task) {
+    // Task gone — mark as done with empty content, try next turn
+    runningMsg.status = 'done';
+    runningMsg.content = '(no response — task was cleaned up)';
+    savePipeline(pipeline);
+    advanceConversation(pipeline);
+    return;
+  }
+  if (task.status === 'done') {
+    runningMsg.status = 'done';
+    runningMsg.content = task.resultSummary || '';
+    savePipeline(pipeline);
+    advanceConversation(pipeline);
+  } else if (task.status === 'failed' || task.status === 'cancelled') {
+    runningMsg.status = 'failed';
+    runningMsg.content = task.error || 'Task failed';
+    pipeline.status = 'failed';
+    pipeline.completedAt = new Date().toISOString();
+    savePipeline(pipeline);
+  } else {
+    // Still running — re-attach listener
+    setupConversationTaskListener(pipeline.id, runningMsg.taskId);
+  }
+}
+
+function advanceConversation(pipeline: Pipeline) {
+  const conv = pipeline.conversation!;
+  const config = conv.config;
+  const lastDoneMsg = [...conv.messages].reverse().find(m => m.status === 'done');
+
+  if (lastDoneMsg && checkConversationStopCondition(conv, lastDoneMsg.content)) {
+    finishConversation(pipeline, 'done');
+    return;
+  }
+
+  conv.currentAgentIndex++;
+  if (conv.currentAgentIndex >= config.agents.length) {
+    conv.currentAgentIndex = 0;
+    conv.currentRound++;
+    if (conv.currentRound > config.maxRounds) {
+      finishConversation(pipeline, 'done');
+      return;
+    }
+  }
+
+  savePipeline(pipeline);
+  const contextForNext = buildConversationContext(conv);
+  scheduleNextConversationTurn(pipeline, contextForNext);
+}
+
+// ─── Recovery: check for stuck pipelines ──────────────────
+
+function recoverStuckPipelines() {
+  const pipelines = listPipelines().filter(p => p.status === 'running');
+  for (const pipeline of pipelines) {
+    // Conversation mode recovery
+    if (pipeline.type === 'conversation' && pipeline.conversation) {
+      recoverConversationPipeline(pipeline);
+      continue;
+    }
+
+    const workflow = getWorkflow(pipeline.workflowName);
+    if (!workflow) continue;
+
+    let changed = false;
+    for (const [nodeId, node] of Object.entries(pipeline.nodes)) {
+      if (node.status === 'running' && node.taskId) {
+        const task = getTask(node.taskId);
+        if (!task) {
+          // Task gone — mark node as done (task completed and was cleaned up)
+          node.status = 'done';
+          node.completedAt = new Date().toISOString();
+          changed = true;
+        } else if (task.status === 'done') {
+          // Extract outputs
+          const nodeDef = workflow.nodes[nodeId];
+          if (nodeDef) {
+            for (const outputDef of nodeDef.outputs) {
+              if (outputDef.extract === 'result' || outputDef.extract === 'stdout') node.outputs[outputDef.name] = task.resultSummary || '';
+              else if (outputDef.extract === 'git_diff') node.outputs[outputDef.name] = task.gitDiff || '';
+            }
+          }
+          node.status = 'done';
+          node.completedAt = new Date().toISOString();
+          changed = true;
+        } else if (task.status === 'failed' || task.status === 'cancelled') {
+          node.status = 'failed';
+          node.error = task.error || 'Task failed';
+          node.completedAt = new Date().toISOString();
+          changed = true;
+        }
+      }
+    }
+
+    if (changed) {
+      savePipeline(pipeline);
+      // Re-setup listener and schedule next nodes
+      setupTaskListener(pipeline.id);
+      scheduleReadyNodes(pipeline, workflow);
+    }
+  }
+}
+
+// Run recovery every 30 seconds
+setInterval(recoverStuckPipelines, 30_000);
+// Also run once on load
+setTimeout(recoverStuckPipelines, 5000);
+
+export function cancelPipeline(id: string): boolean {
+  const pipeline = getPipeline(id);
+  if (!pipeline || pipeline.status !== 'running') return false;
+
+  // Conversation mode
+  if (pipeline.type === 'conversation') {
+    return cancelConversation(id);
+  }
+
+  pipeline.status = 'cancelled';
+  pipeline.completedAt = new Date().toISOString();
+
+  // Cancel all running tasks
+  for (const [, node] of Object.entries(pipeline.nodes)) {
+    if (node.status === 'running' && node.taskId) {
+      const { cancelTask } = require('./task-manager');
+      cancelTask(node.taskId);
+    }
+    if (node.status === 'pending') node.status = 'skipped';
+  }
+
+  savePipeline(pipeline);
+  return true;
+}
+
+// ─── Node Scheduling ──────────────────────────────────────
+
+async function scheduleReadyNodes(pipeline: Pipeline, workflow: Workflow) {
+  const ctx = { input: pipeline.input, vars: pipeline.vars, nodes: pipeline.nodes };
+
+  for (const nodeId of pipeline.nodeOrder) {
+    const nodeState = pipeline.nodes[nodeId];
+    if (nodeState.status !== 'pending') continue;
+
+    const nodeDef = workflow.nodes[nodeId];
+    if (!nodeDef) continue;
+
+    // Check all dependencies are done
+    const depsReady = nodeDef.dependsOn.every(dep => {
+      const depState = pipeline.nodes[dep];
+      return depState && depState.status === 'done';
+    });
+
+    // Check if any dependency failed (skip this node)
+    const depsFailed = nodeDef.dependsOn.some(dep => {
+      const depState = pipeline.nodes[dep];
+      return depState && (depState.status === 'failed' || depState.status === 'skipped');
+    });
+
+    if (depsFailed) {
+      nodeState.status = 'skipped';
+      savePipeline(pipeline);
+      continue;
+    }
+
+    if (!depsReady) continue;
+
+    // Resolve templates
+    const isShell = nodeDef.mode === 'shell';
+    const project = resolveTemplate(nodeDef.project, ctx);
+    const prompt = resolveTemplate(nodeDef.prompt, ctx, isShell);
+
+    const projectInfo = getProjectInfo(project);
+    if (!projectInfo) {
+      nodeState.status = 'failed';
+      nodeState.error = `Project not found: ${project}`;
+      savePipeline(pipeline);
+      notifyStep(pipeline, nodeId, 'failed', nodeState.error);
+      continue;
+    }
+
+    // All pipeline steps use worktree for isolated execution.
+    // Shell steps receive env vars: FORGE_WORKTREE, FORGE_WORKTREE_BRANCH, FORGE_PROJECT_ROOT.
+    // Set worktree: false on a node to skip (e.g. read-only gh commands that don't need isolation).
+    let effectivePath = projectInfo.path;
+    const useWorktree = nodeDef.worktree !== false;
+    const branchName = nodeDef.branch ? resolveTemplate(nodeDef.branch, ctx) : `pipeline/${pipeline.id.slice(0, 8)}`;
+    if (useWorktree) try {
+      const { execSync } = require('node:child_process');
+      const worktreePath = `${projectInfo.path}/.forge/worktrees/${branchName.replace(/\//g, '-')}`;
+      const { mkdirSync } = require('node:fs');
+      mkdirSync(`${projectInfo.path}/.forge/worktrees`, { recursive: true });
+
+      // Create branch if needed
+      try { execSync(`git branch ${branchName}`, { cwd: projectInfo.path, stdio: 'pipe' }); } catch {}
+
+      // Create or reuse worktree
+      try {
+        execSync(`git worktree add "${worktreePath}" ${branchName}`, { cwd: projectInfo.path, stdio: 'pipe' });
+        console.log(`[pipeline] Created worktree: ${worktreePath} (branch: ${branchName})`);
+      } catch {
+        const { existsSync } = require('node:fs');
+        if (existsSync(worktreePath)) {
+          console.log(`[pipeline] Reusing worktree: ${worktreePath}`);
+        } else {
+          try { execSync(`git worktree remove "${worktreePath}" --force`, { cwd: projectInfo.path, stdio: 'pipe' }); } catch {}
+          execSync(`git worktree add "${worktreePath}" ${branchName}`, { cwd: projectInfo.path, stdio: 'pipe' });
+          console.log(`[pipeline] Recreated worktree: ${worktreePath}`);
+        }
+      }
+
+      effectivePath = worktreePath;
+      (nodeState as any).worktreePath = worktreePath;
+      (nodeState as any).worktreeBranch = branchName;
+    } catch (e: any) {
+      console.warn(`[pipeline] Worktree creation failed, falling back to project dir: ${e.message}`);
+    }
+
+    // ── Plugin mode: execute plugin action directly ──
+    if (nodeDef.mode === 'plugin' && nodeDef.plugin) {
+      nodeState.status = 'running';
+      nodeState.startedAt = new Date().toISOString();
+      savePipeline(pipeline);
+      notifyStep(pipeline, nodeId, 'running');
+
+      try {
+        const { getInstalledPlugin } = await import('./plugins/registry');
+        const { executePluginWithWait } = await import('./plugins/executor');
+
+        const inst = getInstalledPlugin(nodeDef.plugin);
+        if (!inst) throw new Error(`Plugin "${nodeDef.plugin}" not installed`);
+        if (!inst.enabled) throw new Error(`Plugin "${nodeDef.plugin}" is disabled`);
+
+        // Resolve template params
+        const resolvedParams: Record<string, any> = {};
+        for (const [k, v] of Object.entries(nodeDef.pluginParams || {})) {
+          resolvedParams[k] = typeof v === 'string' ? resolveTemplate(v, ctx) : v;
+        }
+
+        const actionName = nodeDef.pluginAction || inst.definition.defaultAction || Object.keys(inst.definition.actions)[0];
+        const result = await executePluginWithWait(inst, actionName, resolvedParams, nodeDef.pluginWait);
+
+        if (result.ok) {
+          nodeState.status = 'done';
+          nodeState.completedAt = new Date().toISOString();
+          // Store plugin outputs
+          for (const [name, value] of Object.entries(result.output)) {
+            nodeState.outputs[name] = typeof value === 'string' ? value : JSON.stringify(value);
+          }
+          savePipeline(pipeline);
+          notifyStep(pipeline, nodeId, 'done');
+          console.log(`[pipeline] Plugin ${nodeDef.plugin}.${actionName}: done (${result.duration}ms)`);
+        } else {
+          throw new Error(result.error || 'Plugin action failed');
+        }
+      } catch (err: any) {
+        nodeState.status = 'failed';
+        nodeState.error = err.message;
+        nodeState.completedAt = new Date().toISOString();
+        savePipeline(pipeline);
+        notifyStep(pipeline, nodeId, 'failed', err.message);
+        console.error(`[pipeline] Plugin ${nodeDef.plugin}: failed — ${err.message}`);
+      }
+      continue;
+    }
+
+    // ── Shell/Agent mode: create task ──
+    // Inject worktree info into shell prompt so git commands work correctly
+    const taskMode = nodeDef.mode === 'shell' ? 'shell' : 'prompt';
+    let effectivePrompt = prompt;
+    if (taskMode === 'shell' && effectivePath !== projectInfo.path) {
+      // Prepend env vars so shell scripts know they're in a worktree
+      const envPrefix = `export FORGE_WORKTREE="${effectivePath}" FORGE_WORKTREE_BRANCH="${(nodeState as any).worktreeBranch || branchName}" FORGE_PROJECT_ROOT="${projectInfo.path}" && `;
+      effectivePrompt = envPrefix + prompt;
+    }
+    const task = createTask({
+      projectName: projectInfo.name,
+      projectPath: effectivePath,
+      prompt: effectivePrompt,
+      mode: taskMode as any,
+      agent: nodeDef.agent || undefined,
+    });
+    pipelineTaskIds.add(task.id);
+    // Pipeline tasks use the same model selection as normal tasks:
+    // agent model > global taskModel. No separate pipelineModel override.
+
+    nodeState.status = 'running';
+    nodeState.taskId = task.id;
+    nodeState.iterations++;
+    nodeState.startedAt = new Date().toISOString();
+    savePipeline(pipeline);
+
+    notifyStep(pipeline, nodeId, 'running');
+  }
+
+  // Check if pipeline is complete
+  checkPipelineCompletion(pipeline);
+}
+
+function checkPipelineCompletion(pipeline: Pipeline) {
+  const states = Object.values(pipeline.nodes);
+  const allDone = states.every(s => s.status === 'done' || s.status === 'skipped' || s.status === 'failed');
+
+  if (allDone && pipeline.status === 'running') {
+    const anyFailed = states.some(s => s.status === 'failed');
+    pipeline.status = anyFailed ? 'failed' : 'done';
+    pipeline.completedAt = new Date().toISOString();
+    savePipeline(pipeline);
+    notifyPipelineComplete(pipeline);
+
+    // Sync run status to project pipeline runs
+    try {
+      const { syncRunStatus } = require('./pipeline-scheduler');
+      syncRunStatus(pipeline.id);
+    } catch {}
+
+    // Log worktree info for user review
+    for (const [nodeId, state] of Object.entries(pipeline.nodes)) {
+      const wt = (state as any).worktreePath;
+      const branch = (state as any).worktreeBranch;
+      if (wt && branch) {
+        console.log(`[pipeline] Worktree preserved: ${wt} (branch: ${branch}) — review changes, then: git worktree remove "${wt}"`);
+      }
+    }
+
+    // Release project lock
+    const workflow = getWorkflow(pipeline.workflowName);
+    if (workflow) {
+      const projectNames = new Set(Object.values(workflow.nodes).map(n => n.project));
+      for (const pName of projectNames) {
+        const pInfo = getProjectInfo(resolveTemplate(pName, { input: pipeline.input, vars: pipeline.vars, nodes: pipeline.nodes }));
+        if (pInfo) releaseProjectLock(pInfo.path, pipeline.id);
+      }
+    }
+  }
+}
+
+// ─── Task Event Listener ──────────────────────────────────
+
+const activeListeners = new Set<string>();
+
+function setupTaskListener(pipelineId: string) {
+  if (activeListeners.has(pipelineId)) return;
+  activeListeners.add(pipelineId);
+
+  const cleanup = onTaskEvent((taskId, event, data) => {
+    if (event !== 'status') return;
+    if (data !== 'done' && data !== 'failed') return;
+
+    const pipeline = getPipeline(pipelineId);
+    if (!pipeline || pipeline.status !== 'running') {
+      cleanup();
+      activeListeners.delete(pipelineId);
+      return;
+    }
+
+    // Find the node for this task
+    const nodeEntry = Object.entries(pipeline.nodes).find(([, n]) => n.taskId === taskId);
+    if (!nodeEntry) return;
+
+    const [nodeId, nodeState] = nodeEntry;
+    const workflow = getWorkflow(pipeline.workflowName);
+    if (!workflow) return;
+
+    const nodeDef = workflow.nodes[nodeId];
+    const task = getTask(taskId);
+
+    if (data === 'done' && task) {
+      // Extract outputs
+      for (const outputDef of nodeDef.outputs) {
+        if (outputDef.extract === 'result') {
+          nodeState.outputs[outputDef.name] = task.resultSummary || '';
+        } else if (outputDef.extract === 'stdout') {
+          nodeState.outputs[outputDef.name] = task.resultSummary || '';
+        } else if (outputDef.extract === 'git_diff') {
+          nodeState.outputs[outputDef.name] = task.gitDiff || '';
+        }
+      }
+
+      // Convention: if stdout contains __SKIP__, mark node as skipped (downstream nodes will also skip)
+      const outputStr = task.resultSummary || '';
+      if (outputStr.includes('__SKIP__')) {
+        nodeState.status = 'skipped';
+        nodeState.completedAt = new Date().toISOString();
+        savePipeline(pipeline);
+        scheduleReadyNodes(pipeline, workflow);
+        checkPipelineCompletion(pipeline);
+        return;
+      }
+
+      // Check routes for conditional next step
+      if (nodeDef.routes.length > 0) {
+        const nextNode = evaluateRoutes(nodeDef.routes, nodeState.outputs, pipeline);
+        if (nextNode && nextNode !== nodeId) {
+          // Route to next node — mark this as done
+          nodeState.status = 'done';
+          nodeState.completedAt = new Date().toISOString();
+          // Reset next node to pending so it gets scheduled
+          if (pipeline.nodes[nextNode] && pipeline.nodes[nextNode].status !== 'done') {
+            pipeline.nodes[nextNode].status = 'pending';
+          }
+        } else if (nextNode === nodeId) {
+          // Loop back — check iteration limit
+          if (nodeState.iterations < nodeDef.maxIterations) {
+            nodeState.status = 'pending';
+            nodeState.taskId = undefined;
+          } else {
+            nodeState.status = 'done';
+            nodeState.completedAt = new Date().toISOString();
+          }
+        } else {
+          nodeState.status = 'done';
+          nodeState.completedAt = new Date().toISOString();
+        }
+      } else {
+        nodeState.status = 'done';
+        nodeState.completedAt = new Date().toISOString();
+      }
+
+      savePipeline(pipeline);
+      // No per-step done notification — only notify on start and failure
+    } else if (data === 'failed') {
+      nodeState.status = 'failed';
+      nodeState.error = task?.error || 'Task failed';
+      nodeState.completedAt = new Date().toISOString();
+      savePipeline(pipeline);
+      notifyStep(pipeline, nodeId, 'failed', nodeState.error);
+    }
+
+    // Schedule next ready nodes
+    scheduleReadyNodes(pipeline, workflow);
+  });
+}
+
+function evaluateRoutes(
+  routes: { condition: string; next: string }[],
+  outputs: Record<string, string>,
+  pipeline: Pipeline
+): string | null {
+  for (const route of routes) {
+    if (route.condition === 'default') return route.next;
+
+    // Simple "contains" check: {{outputs.xxx contains 'YYY'}}
+    const containsMatch = route.condition.match(/\{\{outputs\.(\w+)\s+contains\s+'([^']+)'\}\}/);
+    if (containsMatch) {
+      const [, outputName, keyword] = containsMatch;
+      if (outputs[outputName]?.includes(keyword)) return route.next;
+      continue;
+    }
+
+    // Default: treat as truthy check
+    return route.next;
+  }
+  return null;
+}
+
+// ─── Topological Sort ─────────────────────────────────────
+
+function topologicalSort(nodes: Record<string, WorkflowNode>): string[] {
+  const sorted: string[] = [];
+  const visited = new Set<string>();
+  const visiting = new Set<string>();
+
+  function visit(id: string) {
+    if (visited.has(id)) return;
+    if (visiting.has(id)) return; // cycle — skip
+    visiting.add(id);
+
+    const node = nodes[id];
+    if (node) {
+      for (const dep of node.dependsOn) {
+        visit(dep);
+      }
+      // Also add route targets
+      for (const route of node.routes) {
+        if (nodes[route.next] && !node.dependsOn.includes(route.next)) {
+          // Don't visit route targets in topo sort to avoid cycles
+        }
+      }
+    }
+
+    visiting.delete(id);
+    visited.add(id);
+    sorted.push(id);
+  }
+
+  for (const id of Object.keys(nodes)) {
+    visit(id);
+  }
+
+  return sorted;
+}
+
+// ─── Notifications ────────────────────────────────────────
+
+async function notifyStep(pipeline: Pipeline, nodeId: string, status: string, error?: string) {
+  const settings = loadSettings();
+  if (!settings.telegramBotToken || !settings.telegramChatId) return;
+
+  const icon = status === 'done' ? '✅' : status === 'failed' ? '❌' : status === 'running' ? '🔄' : '⏳';
+  const msg = `${icon} Pipeline ${pipeline.id}/${nodeId}: ${status}${error ? `\n${error}` : ''}`;
+
+  try {
+    await fetch(`https://api.telegram.org/bot${settings.telegramBotToken}/sendMessage`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        chat_id: settings.telegramChatId.split(',')[0].trim(),
+        text: msg,
+        disable_web_page_preview: true,
+      }),
+    });
+  } catch {}
+}
+
+async function notifyPipelineComplete(pipeline: Pipeline) {
+  const settings = loadSettings();
+  if (!settings.telegramBotToken || !settings.telegramChatId) return;
+
+  const icon = pipeline.status === 'done' ? '🎉' : '💥';
+  const nodes = Object.entries(pipeline.nodes)
+    .map(([id, n]) => `  ${n.status === 'done' ? '✅' : n.status === 'failed' ? '❌' : '⏭'} ${id}`)
+    .join('\n');
+
+  const msg = `${icon} Pipeline ${pipeline.id} (${pipeline.workflowName}) ${pipeline.status}\n\n${nodes}`;
+
+  try {
+    await fetch(`https://api.telegram.org/bot${settings.telegramBotToken}/sendMessage`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        chat_id: settings.telegramChatId.split(',')[0].trim(),
+        text: msg,
+        disable_web_page_preview: true,
+      }),
+    });
+  } catch {}
+}