task-delegate

作者 MOHJRNL已验证

Low-context multi-agent task delegation for coding agents. Delegate bounded implementation work to OpenCode, Codex, and Claude with compact briefs, safe execution, and structured results.

0
Stars
0
Forks
JavaScript
语言
2026/8/24
添加时间

⚠️ 第三方软件声明

本 Skill 为第三方开源软件,独立托管于 GitHub。SkillTip 仅为信息目录,不控制或维护底层仓库。所显示的安全检查为自动化且范围有限,安装前请自行审查源码。

阅读服务条款

安装

添加到你的 Claude Code skills 目录:

# Add to your Claude Code skills
git clone https://github.com/MOHJRNL/task-delegate

快速入门

使用 task-delegate 等 Skills 的指南。

安全报告

已验证

上次扫描:—

{
  "status": "PASSED",
  "issues": []
}

README.md

TaskDelegate

npm version npm downloads license Node.js GitHub release skills.sh

One portable interface for delegating bounded coding tasks to seven CLI agents—without handing them ownership of your repository.

What is TaskDelegate?

TaskDelegate is an open-source CLI and agent skill that delegates a narrowly scoped coding task to a supported local coding agent, verifies what changed, writes a normalized result, and leaves review and commit control with you.

It gives OpenCode, Codex, Claude Code, Kimi, z.ai, Grok, and Antigravity one consistent delegation workflow.

At a glance

  • One CLI and one portable agent skill
  • Seven live-verified delegation targets
  • Explicit target selection; no hidden auto-router
  • Plan-only and implementation modes
  • Clean-worktree protection
  • Bounded retries, timeouts, and output capture
  • Commit detection and Git diff inspection
  • Normalized task-delegate.result.v2 output
  • No automatic commit, push, publish, or acceptance

Supported targets:

  1. OpenCode
  2. Codex
  3. Claude Code
  4. Kimi
  5. z.ai through OpenCode
  6. Grok
  7. Antigravity

TaskDelegate v2.1.0 has been live-verified against all seven targets.

Why use TaskDelegate?

Using several coding agents usually means remembering different commands, permission modes, workspace behaviors, and output formats. TaskDelegate standardizes that operational layer while keeping the delegated agent bounded and reviewable.

How TaskDelegate works

TaskDelegate workflow: install, discover targets, delegate a bounded task, verify changes, and review the result

bounded task
  → discover seven supported targets
  → explicitly select one target
  → run it in the selected repository
  → inspect process, Git, and changed-file results
  → review the uncommitted diff

TaskDelegate is intentionally not an autonomous router. You select the target explicitly, and every implementation run remains subject to review.

Requirements

  • Node.js 18.17.0 or newer
  • Git for repository verification
  • At least one supported target CLI installed and authenticated

TaskDelegate installs its bundled skill, but it does not install or authenticate third-party coding CLIs.

Install

Run directly with npm

Run the current published release:

npx -y task-delegate@latest setup

Install the CLI globally

npm install -g task-delegate
task-delegate setup

Run from a local checkout

git clone https://github.com/MOHJRNL/task-delegate.git
cd task-delegate
npm link
task-delegate setup

Restart an already-open host CLI after setup so it can discover the installed skill.

Quick start

List targets:

task-delegate targets

Delegate a task:

task-delegate delegate \
  --to codex \
  --task "Fix the failing tests without changing public behavior" \
  --cd .

Run in plan-only mode:

task-delegate delegate \
  --to claude \
  --mode plan \
  --task "Propose the smallest safe fix for the authentication bug" \
  --cd .

Inspect the generated result under:

.task-delegate/runs/<run-id>/

Every run stores a compact brief, the delegated prompt, stdout, stderr, changed files, diff statistics, and a normalized result.json.

Supported coding agents and invocation

TaskDelegate can be invoked from these supported coding-agent hosts using each host's skill syntax:

HostInvocation
OpenCode/task-delegate
Claude Code/task-delegate
Codex$task-delegate
KimiUse task-delegate
GrokUse task-delegate
AntigravityUse task-delegate

The host should display the same seven targets in the canonical order and ask you to choose one. Auto-select is intentionally not exposed.

Main commands

task-delegate targets [--json]
task-delegate delegate [options]
task-delegate setup [--check|--dry-run]
task-delegate verify [--live] [--target <id>] [--jobs <n>] [--timeout-ms <n>] [--json]
task-delegate doctor
task-delegate hosts
task-delegate uninstall
task-delegate update

Delegate options

--to <opencode|codex|claude|kimi|zai|grok|agy>
--task <text>
--cd <path>
--mode <manual|plan>
--model <name>
--timeout-ms <number>
--retries <number>
--no-retry
--allow-dirty
--dry-run
--json

Defaults:

  • mode: manual
  • timeout: 180000 ms
  • retries: one retry for transient launch or timeout failures only
  • clean worktree required

Setup and diagnostics

Install the bundled skill into the supported host locations:

task-delegate setup

Verify binaries, skill paths, and installed versions:

task-delegate setup --check
task-delegate doctor
task-delegate hosts

Preview setup without writing:

task-delegate setup --dry-run

Remove installed skill copies:

task-delegate uninstall

Verification

Run deterministic checks:

npm run check
npm test
npm run pack:dry-run

Run real bounded smoke tests for installed and authenticated targets:

task-delegate verify \
  --live \
  --jobs 2 \
  --timeout-ms 180000

Verify one target:

task-delegate verify \
  --live \
  --target agy \
  --timeout-ms 180000

Security model

TaskDelegate reduces risk; it does not make third-party agents inherently safe.

Core controls:

  • explicit target selection
  • explicit canonical project directory
  • clean-worktree enforcement unless --allow-dirty is supplied
  • no automatic commit or push
  • detection and rejection of commits created by a backend
  • bounded timeout and process-tree termination
  • bounded stdout and stderr capture
  • transient-only bounded retry
  • symlink protection for run and installation paths
  • normalized result validation
  • changed-file and Git diff inspection
  • no shell pipeline in the live verifier
  • manual review required for every implementation run

Delegated prompts prohibit commits, pushes, history rewrites, secret access, writes outside the selected project, and unrequested destructive actions.

Antigravity

Antigravity print mode cannot prompt for write permission. TaskDelegate therefore uses workspace binding, sandbox mode, and non-interactive permission approval for reliable headless editing.

That permission bypass is broad within the Antigravity execution context. Run it only in repositories you trust and review every diff.

Grok

TaskDelegate runs Grok with the delegated process working directory set to the selected project. It does not currently pass a Grok-specific workspace or permission-approval flag.

Dirty worktrees

TaskDelegate rejects a dirty repository by default:

reason: dirty-worktree

You can override this deliberately:

task-delegate delegate \
  --to opencode \
  --task "Add the requested validation" \
  --cd . \
  --allow-dirty

When --allow-dirty is used, the result separates:

  • pre-existing changed files
  • newly changed files
  • files whose attribution may be ambiguous

Review with extra care.

Result contract

The normalized contract identifier is:

task-delegate.result.v2

A backend's own success message is not proof of completion. TaskDelegate also checks process outcome, timeout state, changed files, Git state, commit creation, and result validity.

Large logs remain separate from result.json.

Legacy compatibility

The older brief-based interface remains available for backward compatibility:

task-delegate run \
  --backend opencode \
  --mode plan \
  --brief examples/brief.sample.md \
  --cd .

New usage should prefer:

task-delegate delegate

See MIGRATION.md for upgrading from npm version 0.2.1.

Troubleshooting

A target is unavailable

Install and authenticate that target's CLI, then run:

task-delegate doctor
task-delegate setup --check

The installed skill is stale

Reinstall the bundled current version:

task-delegate setup
task-delegate setup --check

Restart the host CLI afterward.

A delegation times out

Increase the per-run timeout:

task-delegate delegate \
  --to codex \
  --task "Complete the bounded task" \
  --cd . \
  --timeout-ms 300000

TaskDelegate terminates the delegated process tree when the timeout is reached.

Antigravity reports success but creates nothing

Do not trust its response status alone. Check the normalized TaskDelegate result and actual changed files. TaskDelegate's live verifier validates the requested file content and Git state.

Frequently asked questions

Does TaskDelegate install the coding agents?

No. TaskDelegate installs its own bundled skill. OpenCode, Codex, Claude Code, Kimi, Grok, and Antigravity must be installed and authenticated separately. z.ai is used through the OpenCode adapter.

Does TaskDelegate choose an agent automatically?

No. Version 2.1.0 requires explicit target selection. Hidden or automatic routing is intentionally not exposed.

Can a delegated agent commit or push code?

TaskDelegate instructs targets not to commit or push, detects a changed Git HEAD, and fails the run if a backend creates a commit. You still need to inspect the resulting diff.

Where are delegation results stored?

Each run writes artifacts under .task-delegate/runs/<run-id>/, including result.json, logs, prompts, changed files, and diff statistics.

Is TaskDelegate an autonomous multi-agent framework?

No. It is a bounded delegation interface. It dispatches one scoped task to one selected target and returns control for review.

Development

npm run check
npm test
npm run delegate:dry-run
npm run pack:dry-run

The package has no runtime dependencies.

License

Apache License 2.0. See LICENSE and NOTICE.

常见问题

What is task-delegate?

task-delegate is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by MOHJRNL. Low-context multi-agent task delegation for coding agents. Delegate bounded implementation work to OpenCode, Codex, and Claude with compact briefs, safe execution, and structured results. It has 0 GitHub stars.

Is task-delegate safe to use?

Yes. task-delegate passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.

How do I install task-delegate?

Clone the repository with "git clone https://github.com/MOHJRNL/task-delegate" and add it to your Claude Code skills directory (see the Installation section above).

What programming language is task-delegate written in?

task-delegate is primarily written in JavaScript. It is open-source under MOHJRNL on GitHub, so you can review or fork the full source.

Are there alternatives to task-delegate?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh task-delegate against similar tools.

评论 (0)

暂无评论,成为第一个分享想法的人!

ECC

by affaan-m

10

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

242,21936,702JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情
15

An agentic skills framework & software development methodology that works.

234,96620,863Shell
AI 智能体ai-agentsbrainstorming
查看详情

hermes-agent

by NousResearch

10

The agent that grows with you

234,43747,175Python
AI 智能体ai-agentsagent-orchestration
查看详情

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

185,94028,768JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情

cc-switch

by farion1231

3

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

128,8688,826Rust
AI 智能体claude-codeai-tools
查看详情

claude-code

by anthropics

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

120,03119,897Shell
AI 智能体
查看详情

开发者还喜欢

基于喜欢此 Skill 的开发者投票和收藏

ECC

by affaan-m

10

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

242,21936,702JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情
15

An agentic skills framework & software development methodology that works.

234,96620,863Shell
AI 智能体ai-agentsbrainstorming
查看详情

hermes-agent

by NousResearch

10

The agent that grows with you

234,43747,175Python
AI 智能体ai-agentsagent-orchestration
查看详情

n8n

by n8n-io

12

Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.

201,88160,308TypeScript
MCP 服务器apisai-tools
查看详情

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

185,94028,768JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情

cc-switch

by farion1231

3

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

128,8688,826Rust
AI 智能体claude-codeai-tools
查看详情