← Back to Skills Marketplace
yonghaozhao722

Longrunning Agent

by Yonghao Zhao · GitHub ↗ · v1.1.1
cross-platform ⚠ suspicious
890
Downloads
0
Stars
4
Active Installs
5
Versions
Install in OpenClaw
/install longrunning-agent
Description
Enables AI agents to work on long-running projects across multiple sessions. Use when starting complex projects, resuming work on existing projects, managing...
README (SKILL.md)

OpenClaw Long-Running Agent Skill

This skill enables AI agents to work on long-running projects across multiple sessions.

Purpose

The longrunning-agent skill provides a structured workflow for:

  • Tracking progress across sessions
  • Managing task lists with priorities and dependencies
  • Making incremental, atomic progress on complex projects
  • Ensuring continuity when resuming work

Installation

  1. Copy this skill directory to your OpenClaw skills folder
  2. Ensure Claude Code CLI is installed and configured
  3. Create a project directory with the workflow files

Usage

Initialize a New Project

# Create project directory
mkdir my-project && cd my-project

# Initialize workflow files
claude -p "Initialize this project using the longrunning-agent workflow"

Workflow Files

The skill expects these files in the project directory:

  • CLAUDE.md - Project instructions and workflow guide
  • task.json - Task list with priorities and dependencies
  • progress.txt - Log of work completed
  • init.sh - Environment setup script (optional)

Task Format

{
  "tasks": [
    {
      "id": "task-1",
      "description": "Set up project structure",
      "priority": 1,
      "dependencies": [],
      "passes": false
    },
    {
      "id": "task-2",
      "description": "Implement core features",
      "priority": 2,
      "dependencies": ["task-1"],
      "passes": false
    }
  ]
}

Progress Format

[2024-01-15 10:30:00] Started session
[2024-01-15 10:35:00] Completed task: Set up project structure
[2024-01-15 10:40:00] Milestone: Core features implemented

Workflow Steps

  1. Read Progress - Check progress.txt for recent work
  2. Select Task - Find next passes: false task with met dependencies
  3. Initialize - Run init.sh if needed
  4. Implement - Work on one task incrementally
  5. Test - Run lint, build, and tests
  6. Document - Update progress.txt
  7. Mark Complete - Set passes: true in task.json
  8. Commit - Make atomic git commit

Best Practices

  • Work on ONE task per session
  • Make commits after each task completion
  • Keep progress.txt concise but informative
  • Use dependencies to manage task order
  • Test thoroughly before marking passes: true

Integration with Web UI

This skill integrates with the Agent Workflow Web App:

  • Tasks sync with the web database
  • Progress entries are captured
  • Session output is logged
  • Git commits are tracked

Templates

Templates for workflow files are in the templates/ directory:

  • CLAUDE.md.tpl - Project template
  • task.json.tpl - Task list template
Usage Guidance
This skill appears to implement a sensible long-running project workflow, but it has a few transparency issues you should check before installing: 1) Verify external dependencies: the SKILL.md asks you to have the Claude Code CLI configured and the manifest lists the GitHub CLI ('gh') — install and configure these yourself, and ensure you trust them. 2) Inspect any init.sh or template files before running them (they can execute arbitrary code). 3) Ask the author or inspect the referenced repository for details about the 'Agent Workflow Web App' integration — who hosts the web database, what endpoints are used, and what credentials (if any) are required. 4) Prefer skills that explicitly declare required binaries and environment variables; absence of those declarations here is the main reason for caution. If you can review the upstream repository or get confirmation about the web sync behavior and any required credentials, the risk assessment can be raised to higher confidence.
Capability Analysis
Type: OpenClaw Skill Name: longrunning-agent Version: 1.1.1 The skill instructs the AI agent to execute an optional `init.sh` script found in the project directory as part of its workflow (SKILL.md). This presents a significant Remote Code Execution (RCE) vulnerability, as the agent would execute arbitrary code from an untrusted or compromised project's `init.sh` without explicit sanitization or warning. Additionally, the `manifest.json` lists `gh` (GitHub CLI) as a required tool, granting the agent broad capabilities for interacting with GitHub repositories, which could be misused if the agent is later prompted maliciously. While these capabilities are risky, there is no direct evidence of intentional malicious behavior or self-exploitation within the provided skill files.
Capability Assessment
Purpose & Capability
The SKILL.md describes a project/workflow manager and the instructions (task.json, progress.txt, working on one task, git commits) are coherent with that purpose. However the manifest lists 'tools': ["gh"] and the README tells you to ensure the Claude Code CLI is installed — yet required binaries/env vars are empty. This mismatch between declared requirements and the instructions is unexpected.
Instruction Scope
Runtime instructions are focused on reading/writing local workflow files, optionally running an init.sh, running lint/build/tests, and making git commits — all reasonable for this purpose. The instructions also claim integration with an 'Agent Workflow Web App' (tasks sync with web database, session output logged) but provide no endpoints, credentials, or steps for that sync; that unspecified network behavior is worth flagging.
Install Mechanism
This is an instruction-only skill with no install script or downloaded code. The manifest's install destination is just a copy location. No archive downloads or third‑party packages are pulled by the skill itself.
Credentials
The skill does not request any environment variables or credentials, but the SKILL.md requires external tools (Claude Code CLI) and references syncing to a web database. If those integrations require API keys or tokens, they are not declared here — the omission reduces transparency and could lead to unexpected credential use by supporting tooling or templates.
Persistence & Privilege
The skill does not request always:true or other elevated persistence. It does not declare modifications to other skills or system-wide settings. Normal autonomous invocation is allowed (platform default).
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install longrunning-agent
  3. After installation, invoke the skill by name or use /longrunning-agent
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.1.1
Fix description: add explicit trigger phrases per Anthropic guidelines
v1.1.0
- Added YAML frontmatter to SKILL.md with name and description for improved metadata support. - No changes to core functionality or workflow; documentation was updated for better compatibility.
v1.0.2
Remove README.md, clean up SKILL.md
v1.0.1
Add authors/contributors section - credit Anthropic Claude
v1.0.0
Initial release: Track progress across sessions for long-running projects
Metadata
Slug longrunning-agent
Version 1.1.1
License
All-time Installs 4
Active Installs 4
Total Versions 5
Frequently Asked Questions

What is Longrunning Agent?

Enables AI agents to work on long-running projects across multiple sessions. Use when starting complex projects, resuming work on existing projects, managing... It is an AI Agent Skill for Claude Code / OpenClaw, with 890 downloads so far.

How do I install Longrunning Agent?

Run "/install longrunning-agent" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Longrunning Agent free?

Yes, Longrunning Agent is completely free (open-source). You can download, install and use it at no cost.

Which platforms does Longrunning Agent support?

Longrunning Agent is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Longrunning Agent?

It is built and maintained by Yonghao Zhao (@yonghaozhao722); the current version is v1.1.1.

💬 Comments