dispatching-parallel-agents · Parallel Task Dispatch
Facing multiple independent tasks (failures in different places, different subsystems), dispatch one focused agent per problem domain to work concurrently instead of investigating sequentially.
Core principle: one agent per independent domain; each gets focused, self-contained instructions with explicit constraints (don't touch other code) and expected output (what you found and fixed). Multiple dispatches in one round = parallel; one at a time = sequential. On return: check for conflicts, run the full test suite, spot-check.
Example invocation: "These three test files fail for different reasons — investigate each."
Full brief
Positioning
dispatching-parallel-agents is a parallel-investigation method: independent problem domains are dispatched to focused agents that work concurrently. It's about how to dispatch tasks — not a specific debugging technique.
Core capabilities
- Independent-domain identification: group failures by what's broken (different test files, subsystems, bugs); independence means no shared state and no sequential dependencies.
- Fit judgment: multiple different-root-cause failures, independently broken subsystems; not for related failures or exploratory debugging needing a full-system view.
- Focused prompt structure: focused (one domain) + self-contained (full context) + explicit output (root cause and fix summary) + constraints (don't change other code).
- Parallel dispatch: multiple dispatches in one round = parallel; one per round = sequential.
- Integration checks: read each summary, check for edits to the same code, run the full test suite, spot-check against systematic errors.
Workflow
- Identify independent problem domains (group the failures)
- Write a focused brief per domain (scope/goal/constraints/expected output)
- Dispatch all in the same round
- Read each returned summary
- Check conflicts, run the full suite, spot-check
Inputs & outputs
| Input | Notes |
|---|---|
| Multiple independent failures | Different-root-cause test failures, different-subsystem bugs |
| Per-domain context | Error messages, test names, relevant code locations |
Output: per-agent root-cause and fix summaries + an all-green full suite after integration.
Fit
- 3+ test files failing with different root causes
- Multiple subsystems broken independently
- Each problem understandable without the others' context
Not for
- Related failures (fixing one may fix others)
- Understanding that requires full system state
- Agents that would interfere (editing the same files, sharing resources)
Before you start
- No API key.
- Prompt quality decides success: too-broad ("fix all tests"), missing context, no constraints, and vague output are the four common mistakes.
dispatching-parallel-agents is part of the Aiglade Skill library. Invoke it from the Aiglade chat box in plain language.