MonoClaw

Delegation Patterns

Delegation lets Mona spawn child agents for parallel work. These patterns show when and how to use it effectively.

Pattern 1: Parallel research

Research multiple topics simultaneously:

I need to compare three database options for our project:
1. PostgreSQL — check features, licensing, hosted options
2. MongoDB — check features, licensing, hosted options
3. SQLite — check features, licensing, hosted options

Delegate each research task to a separate agent, then synthesize the results into a comparison table.

Pattern 2: Multi-file code review

Review multiple files in parallel:

Review these files for security issues:
- src/auth.ts
- src/middleware.ts
- src/routes/users.ts
- src/routes/payments.ts

Delegate one agent per file, then summarize all findings.

Pattern 3: Batch data processing

Process multiple data files:

I have 10 CSV files in data/2024/. For each file:
1. Validate the schema
2. Check for anomalies
3. Generate a summary

Delegate one agent per file, then combine the summaries into a report.

Pattern 4: A/B analysis

Compare two approaches:

I'm considering two architectures for our API:
Approach A: REST with OpenAPI
Approach B: GraphQL with Apollo

Delegate one agent to analyze Approach A, another to analyze Approach B. Compare pros, cons, and trade-offs. Recommend one with justification.

Pattern 5: Multi-step pipeline

Chain dependent tasks:

I need to:
1. Scrape https://example.com/products
2. Extract product names and prices
3. Compare with our pricing
4. Generate a competitive analysis

Step 1 can be delegated. Steps 2-4 depend on its output, so run sequentially after.

Pattern 6: Error recovery

Retry failed tasks with different parameters:

Try to build the project. If it fails:
1. Delegate an agent to diagnose the error
2. Delegate another agent to find a fix
3. Apply the fix and retry

When NOT to delegate

Don't delegate when:

  • The task is simple and quick (< 30 seconds)
  • The subtasks are highly dependent on each other
  • You're hitting API rate limits
  • Context window is already constrained

Configuration for delegation

# ~/.monoclaw/config.yaml
delegation:
  max_agents: 5
  default_timeout: 300
  share_memory: true
  retry_failed: true

Monitoring delegated tasks

monoclaw delegation status

Output:

Active delegations: 3
- Agent #1: Researching PostgreSQL (45s elapsed)
- Agent #2: Researching MongoDB (42s elapsed)
- Agent #3: Researching SQLite (40s elapsed)

Best practices

  • Keep tasks independent — Agents work best in parallel
  • Set timeouts — Prevent runaway agents
  • Limit concurrency — 3-5 agents is usually optimal
  • Synthesize results — Don't just concatenate outputs; combine them meaningfully
  • Handle failures — One failed agent shouldn't crash the whole task