skills.rest
adjohn
adjohnCOMMUNITY
Explore all skills by this author →
GitHub

swarm

Orchestrates parallel background subagents and aggregates their results into one report.

3|2|Updated Aug 28, 2026
Search Tags:#claude code#parallel agents#subagents#orchestration#aggregation#task decomposition#background workers
One-click install
npx skills add https://github.com/adjohn/pstack --skill swarm-adjohn
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: swarm
Source: https://github.com/adjohn/pstack/tree/main/skills/swarm
Command: npx skills add https://github.com/adjohn/pstack --skill swarm-adjohn

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating multiple parallel AI workers manually is error-prone: briefs get inconsistent, results arrive scattered, and aggregation becomes guesswork. This Skill provides a structured four-phase workflow to fan out N parallel workers, collect their terminal results, and return one consolidated report. ## Core Features & Use Cases - Parallel Fan-Out: Spawns N background subagents in a single message, each with a standalone brief covering goal, scope, verification, and reporting format. - Flexible Shapes: Supports coverage slicing (partition work across workers), racing (N workers on identical briefs with first-pass, rank-all, or best-of selection), or mixed modes. - Model Selection: Picks worker models from a pstack-models.md config, defaulting to haiku, and supports multi-model races. - Use Case: Ask to swarm a codebase review across five modules; each worker audits its slice in an isolated worktree and reports PASS, ISSUES, or BLOCKED, then you receive one aggregated table with evidenced issues and gaps. ## Quick Start Ask the assistant to swarm this task with four parallel workers and return one consolidated report.

Frequently Asked Questions about swarm

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run parallel AI agents on the same task?▼

Use the swarm workflow to spawn N background subagents in one message, each with a standalone brief covering goal, scope, verification, and reporting. The parent waits for all terminal results and returns one aggregated report.

How to race multiple AI models against each other?▼

Declare a race shape up front with a selection rule: first pass, rank all, or best-of. Name each arm's model before spawning, then apply the declared rule when aggregating the workers' terminal results.

Which model do swarm workers use by default?▼

Workers use the model listed under 'swarm workers' in ~/.claude/pstack-models.md when that file is present. Otherwise the default is haiku, and model races name each arm's model explicitly.

What happens if a parallel worker fails or drops out?▼

The swarm proceeds with N-1 workers and notes the dropout. The final report includes explicit gaps or dropouts alongside the result table and issue one-liners.

How do parallel agents avoid overwriting each other's changes?▼

Each worker gets its own writable output: a git worktree, a separate branch, or an isolated directory like /tmp/swarm-<slug>/worker-<n>/. Workers needing a non-default branch check it out in their own worktree first.

Related Skills

Explore More Software Engineering Skills→
anthony-chaudharyanthony-chaudhary
18

dos-goal-fleet

Coordinate parallel goal execution with wave-based worker launches and verified completion.

Community
Advanced
gobifrostgobifrost
55

bifrost-issues

Automate GitHub Issue tracking, git worktrees, and PR merges for Bifrost.

Official
Advanced
LerianStudioLerianStudio
205

linting-codebase

Analyze lint results and dispatch parallel AI agents to fix issues until checks pass.

Official
Advanced
CasaPerksCasaPerks

sdd

Execute implementation plan tasks via isolated subagents with spec-first review.

Official
Advanced