About parallel-agents
A skill for user-requested parallel agent work. It guides task decomposition, concurrent execution, and integration while limiting overlapping edits.
Key features
- 01Split work into independent lanes
- 02Run read-only investigations concurrently
- 03Set lane-specific models and effort
- 04Define allowed paths and output shape
- 05Integrate and verify changes
Use cases
- 01Research independent questions in parallel
- 02Divide repository exploration by directory or package
- 03Coordinate non-overlapping changes across agents
Install
Copy the command and run it in a terminal.
npx skills add https:/After installing, type/parallel-agentsin Claude Code to confirm it loaded.
Bundled filesagents/openai.yaml
Details
From the project READMEParallel Agents
User-invoked fan-out. Use when user asks for parallel or multi-agent work.
Fan out with the workflow tool. One run owns all lanes.
Set model and effort per lane with agent({ model, effort }).
Small fan-outs: Agent tool works too. Launch each lane, integrate in main thread.
Never one agent per one-line lookup.
Steps
-
Slice lanes. Read-only first: one agent per independent question, directory, package, axis. Launch concurrently.
-
Prompt tightly. Give each agent: lane, read/write mode, allowed paths, forbidden paths, expected output shape.
-
Write only when non-overlapping. Fixed-contract, non-overlapping file sets. Two lanes might touch same file: keep sequential.
-
Keep moving. While agents run: do local reading, planning, small safe edits. Do not poll.
-
Integrate. Read summaries. For writers: verify diffs. Resolve seams in main thread. Run narrowest useful check.
Rules
No duplicate lanes. No agents for one-line lookups. Never delegate ownership. Prefer more read agents over more write agents.