parallel-code-agents-plus

hero

License: MIT GitHub stars Works with Claude Code Skill v1.0

Built on affaan-m/ECC by @affaan-m (248,018 stars, MIT). All credit for the original idea to them. This fork improves and repackages it; upstream license preserved in UPSTREAM_LICENSE.

A Claude Code skill that plans coding work, sends safe tasks to separate agents, and gives you clear reports.

It is for developers who use Claude Code with Claude DevFleet. It works with your current coding flow.

Large coding jobs often have parts that can run at the same time.

But two agents may change the same file. A task may start before the code it needs is ready. A failed task may block other work.

This skill helps Claude make a clear plan first. Each agent gets its own Git worktree. Claude checks task links, waits for needed work, and reads each report before it claims success.

Use it for work with clear, separate parts. Do not use it for a small job or tasks that must edit the same files.

curl --create-dirs -fsSL https://raw.githubusercontent.com/parallel-code-agents-plus/parallel-code-agents-plus/main/skill/SKILL.md -o ~/.claude/skills/claude-devfleet/SKILL.md

Claude DevFleet must also be running and linked to Claude Code:

claude mcp add devfleet --transport http http://localhost:18801/mcp

Ask Claude Code:

Use claude-devfleet to build a REST API with sign-in and tests.

Claude will make a plan like this:

M1: Add the base API routes
M2: Add sign-in
M3: Add tests, depends_on=[M1, M2]

Expected flow:

1. Claude checks that the tasks do not form a loop.
2. Claude shows you the plan and asks for approval.
3. M1 and M2 run at the same time in separate Git worktrees.
4. M3 starts only after M1 and M2 pass.
5. Claude reads each report.
6. You get one final report with changed files, tests, errors, and next steps.

If DevFleet cannot be reached, Claude stops and tells you how to connect it. It does not make up results.

• Fully rewrote the Japanese text in plain English and added credit for the community author.
• Made the limits clear: do not use it for small tasks or work where agents would edit the same files.
• Added clear steps for connection failure, plan approval, dependency loop checks, and safe status checks without too much polling.
• Added stronger safety rules against double dispatch, unverified success claims, work that depends on a failed mission, and outside network calls.
• Replaced the broad example with a clear API mission, sign-in mission, and a test mission that depends on both.

Released under the MIT License. The upstream MIT license and author credit are preserved in UPSTREAM_LICENSE.