Parallel Tasks Skill#

Execute multiple tasks in parallel with enterprise-grade reliability: timeout protection, error isolation, and real-time progress feedback.

When to Use#

Use this skill when:

• User says "run these in parallel" or "do these simultaneously"
• Multiple independent tasks need to be executed at once
• User wants faster results by running tasks concurrently
• Tasks are slow and user wants to avoid waiting sequentially

Core Concept#

Serial vs Parallel:

text

SERIAL (slow):
Task 1 → Task 2 → Task 3  (5min + 5min + 5min = 15min)

PARALLEL (fast):
Task 1 ─┬─> (5min total, not 15min)
Task 2 ─┼─>
Task 3 ─┘

Usage#

Basic Parallel Execution#

text

/parallel
- Task 1: Search for docs
- Task 2: Search for code
- Task 3: Search for examples

Named Tasks with Custom Timeout#

text

/parallel timeout=300
- [search-docs] Search for relevant documentation
- [search-code] Find similar implementations
- [analyze] Analyze the results

CLI Usage (scripts/executor.ts)#

bash

# Simple usage
node scripts/executor.ts "Research AI trends" "Research market analysis"

# Named tasks with custom timeout
node scripts/executor.ts --timeout 600 \
  --task "[research] Research AI trends" \
  --task "[implement] Build the feature"

# Read from file (one task per line)
node scripts/executor.ts --tasks-file my-tasks.txt --max-concurrent 3

# Named task formats (all equivalent):
# - [name] description
# - - description (auto-named as task-1, task-2, ...)
# - 1. description

Implementation#

Core Execution Pattern#

The executor uses a semaphore pattern with configurable concurrency:

typescript

// 1. Parse tasks from input
const tasks = parseTaskInput(input)

// 2. Execute tasks with concurrency control
const results: TaskResult[] = []
const executing: Promise<void>[] = []

for (const task of tasks) {
  // Wait if at max concurrency
  if (executing.length >= maxConcurrent) {
    await Promise.race(executing)
  }

  const promise = runTask(task).then(result => {
    results.push(result)
    // Remove from executing list
    const idx = executing.indexOf(promise)
    if (idx > -1) executing.splice(idx, 1)
  })

  executing.push(promise)
}

await Promise.all(executing)

Task Execution via hermes cli#

typescript

async function executeTaskViaSpawn(
  task: Task,
  timeoutSeconds: number
): Promise<TaskResult> {
  const taskId = `parallel-${Date.now()}-${randomId()}`

  return new Promise((resolve) => {
    const proc = spawn('hermes', [
      'cli', '--',
      'sessions_spawn',
      '--task', `"${task.description}"`,
      '--label', `"${task.name}"`,
      '--timeout', String(timeoutSeconds),
      '--session-id', taskId
    ], { stdio: ['ignore', 'pipe', 'pipe'] })

    // Timeout handling
    const timeoutId = setTimeout(() => {
      proc.kill('SIGTERM')
      resolve({
        name: task.name,
        status: 'timeout',
        duration: Date.now() - startTime,
        error: `Exceeded ${timeoutSeconds}s timeout`
      })
    }, timeoutSeconds * 1000)

    proc.on('close', (code, signal) => {
      clearTimeout(timeoutId)
      if (signal === 'SIGTERM') {
        resolve({ name: task.name, status: 'timeout', ... })
      } else if (code === 0) {
        resolve({ name: task.name, status: 'fulfilled', ... })
      } else {
        resolve({ name: task.name, status: 'rejected', ... })
      }
    })
  })
}

Timeout Protection#

OptionDefaultDescriptiontimeout300Default timeout per task (seconds)Per-task timeout-Override global timeout for specific tasks

Behavior: Task auto-terminates after timeout, other tasks continue.

Error Isolation#

Each task runs in complete isolation:

ProblemSerialParallel (This Skill)One task failsAll others stopOnly failed task affectedOne task hangsBlocks entire flowOthers continue normallyOne task times outMay cascadeContained, others finish

Concurrency Control#

OptionDefaultDescriptionmaxConcurrent5Maximum tasks running simultaneously

Pattern: Semaphore-style - starts N tasks, when one completes, starts next.

Progress Feedback#

Real-time terminal output with colored status:

text

🚀 Starting 3 tasks in parallel (max 5 concurrent)...

🔄 [task-1] Starting (timeout: 300s)...
🔄 [task-2] Starting (timeout: 300s)...
🔄 [task-3] Starting (timeout: 300s)...
✅ [1/3] [task-1] Complete (23.5s)
✅ [2/3] [task-2] Complete (45.2s)
⏱️  [3/3] [task-3] Timeout after 300s

Task Input Formats#

1. Named Tasks (Recommended)#

text

[research] Research AI trends and write report
[implement] Build the feature
[test] Write comprehensive tests

2. Bullet List#

text

- Search for API documentation
- Find relevant code examples
- Check for existing implementations

3. Numbered List#

text

1. Research authentication patterns
2. Design database schema
3. Implement API endpoints

4. Plain Text (auto-named)#

text

Research AI trends
Build the feature
Write tests

→ Auto-named: task-1, task-2, task-3

Output Format#

Success Case#

text

✅ Parallel Execution Complete
   3 tasks: 2 succeeded, 1 failed (45.2s total)

┌─────────────────────┬────────────┬────────────┐
│ Task                │ Status     │ Duration   │
├─────────────────────┼────────────┼────────────┤
│ research            │ ✅ fulfilled│ 23.5s      │
│ implement           │ ✅ fulfilled│ 45.2s      │
│ test                │ ⏱️ timeout  │ 300.0s    │
└─────────────────────┴────────────┴────────────┘

❌ Failed Tasks:
   • test: Exceeded 300s timeout

Exit Codes#

CodeMeaning0All tasks succeeded1Some tasks failed or timed out

Options#

CLI Options#

OptionDefaultDescription--timeout, -to300Timeout per task (seconds)--max-concurrent, -m5Max concurrent tasks--stop-on-errorfalseStop all if one fails--no-progressfalseSuppress progress output--tasks-file, -f-Read tasks from file--parse-Parse stdin to JSON

Per-Task Options (in task description)#

text

[name] description (timeout=600)

Error Handling#

Error Types#

StatusCauseBehaviorfulfilledTask succeededReturns result valuetimeoutExceeded timeoutTask terminated, others continuerejectedProcess errorError captured, others continuecancelledUser cancelledAll running tasks terminateno_replyNo outputReported as warning

Best Practices#

1. Independent tasks first: Tasks should not depend on each other
2. Set reasonable timeouts: Don't set 5min if task should take 30s
3. Use named tasks: Easier to debug when something fails
4. Keep tasks focused: One clear goal per task
5. Mind concurrency: Don't set maxConcurrent higher than system can handle

Examples#

Example 1: Research Multiple Topics#

bash

node scripts/executor.ts \
  "Research Claude Code best practices" \
  "Find OpenClaw skill examples" \
  "Search for agent design patterns"

Example 2: Named Tasks from File#

bash

# tasks.txt:
# [research] Research AI trends
# [implement] Build the feature
# [test] Write tests

node scripts/executor.ts --tasks-file tasks.txt --timeout 600

Example 3: Parallel Implementation#

bash

node scripts/executor.ts --timeout 600 \
  --task "[backend] Implement user authentication API" \
  --task "[frontend] Build login form component" \
  --task "[database] Create users table migration"

Example 4: Web Scraping#

bash

node scripts/executor.ts \
  --task "[store1] Fetch product data from store1.com" \
  --task "[store2] Fetch product data from store2.com" \
  --task "[store3] Fetch product data from store3.com"

Anti-Patterns#

❌ Don't use for dependent tasks:

bash

# WRONG - second task depends on first!
node scripts/executor.ts \
  "Create user account" \
  "Send welcome email"

Use sequential execution instead.

❌ Don't use for very fast tasks:

bash

# WRONG - spawning overhead not worth it
node scripts/executor.ts "Read file A" "Read file B" "Read file C"

The overhead of spawning parallel sessions isn't worth it for sub-second tasks.

Related Skills#

• subagents - Background agent spawning
• batch-operations - Bulk file operations
• workflow-orchestrator - Complex multi-step workflows