---
name: parallel-agent-task-dispatcher
description: "Splits independent work into self-contained briefs for several agents running at once, then merges their results with checks. Use when a large job has pieces that could run at once and each needs a self-contained brief for a separate agent. Do not use for working a single task alone or running a plan step by step in one session. Related topics: AI, productivity."
---

# Parallel Agent Task Dispatcher

Splits independent work into self-contained briefs for several agents running at once, then merges their results with checks.

## When to use

Use this skill when a large job has pieces that could run at once and each needs a self-contained brief for a separate agent. Do not use it for working a single task alone or running a plan step by step in one session.

## Workflow

1. Ask the user for each input listed below in a single message, skipping any they have already given. Where options are listed, offer them, but accept other answers too.
2. Test every candidate piece for independence from the others and for shared file writes.
3. Merge or sequence any pieces that fail, stating the reason for each.
4. Write a standalone brief per piece with goal, allowed files, context, done definition, return format and stop rule.
5. Read each returned result critically for conflicts, duplicates and unmet done criteria.
6. Merge in a stated order, run the verification command and report each agent's contribution.

## Inputs

- **The job** (longer free text)
- **Candidate pieces** (longer free text)
- **Shared files or resources** (longer free text)
- **Maximum agents** (number from 2 to 8)
- **How results come back** (choose one: Written report, Branch per agent, Patch file)
- **Verification after merge** (free text, e.g. "Command that checks the combined result")

## Output

A dispatch plan listing each brief in full, followed after the run by a merge report with conflicts found, the order merged and the verification result.

## Prompt template

Fill the inputs into this template and follow it as the full instruction.

### Parallel Agent Task Dispatcher

Split this job into parallel work and write a brief for each agent.

- The job: {{The job}}
- Candidate pieces: {{Candidate pieces}}
- Shared files or resources: {{Shared files or resources}}
- Maximum agents: {{Maximum agents}}
- How results come back: {{How results come back}}
- Verification after merge: {{Verification after merge}}

First test each piece for independence: it needs no output from another piece, and no two pieces write the same file. Merge or sequence pieces that fail the test, and say why. Write each brief to stand alone, because the agent sees nothing else: the goal, the files it may touch and those it must not, the context it needs, the definition of done, the format of its return and a rule to stop and report instead of guessing. Do not share secrets in briefs. After the agents finish, read every result critically, check for conflicting edits or duplicated work, merge in a stated order, run the verification command and report what each agent contributed.

## Pitfalls

- Briefs that rely on context from the main conversation, which the receiving agent cannot see.
- Two agents editing the same file, which leaves a conflict nobody owns.
- Merging results unread because each agent said it succeeded.