openclaw-kilo-agent

High-performance coding agent and browser automation orchestrator using the Kilo CLI. Use when you need to: (1) Offload heavy-duty coding tasks (refactoring,...

Security Scan

The skill's instructions match a Kilo-based automation agent, but the package metadata omits required binaries/installation and it encourages auto-approval and runtime npm downloads — mismatches that warrant caution.

Purpose & Capability

The SKILL.md clearly expects a locally installed Kilo CLI and the ability to launch MCP servers via npm/npx and Puppeteer; however the registry metadata declares no required binaries or installs. That omission is inconsistent: a Kilo orchestrator legitimately needs the Kilo binary (and often npm) available.

Instruction Scope

Instructions instruct the agent to run Kilo commands that control browsers (Puppeteer) and to use `--auto` to bypass interactive prompts. While browser automation is within the stated purpose, the guidance to always auto-approve operations and to invoke npm/npx MCP servers at runtime expands the agent's autonomy and can enable actions (navigation, form-filling, scraping) without explicit user confirmation.

Install Mechanism

No install spec is provided (lowest static installation risk). However the documentation requires runtime use of npm/npx to launch MCP servers (e.g., @modelcontextprotocol/server-puppeteer). That implies on-demand downloads and execution of third-party npm packages, which is higher-risk behavior not captured in the metadata.

Credentials

The skill declares no required environment variables or credentials, yet mentions integration with model providers, GitHub, and MCP servers which typically require tokens/keys. The absence of declared env requirements is inconsistent and may hide the need to supply sensitive credentials at runtime. Additionally, auto-approval increases the chance those credentials could be used without further prompts.

Persistence & Privilege

The skill is not forced-always and model invocation is allowed (normal). However the skill explicitly recommends `--auto` to bypass prompts, effectively increasing its ability to act autonomously. That autonomy combined with runtime npm downloads and possible credential use raises the blast radius if misused.

What to consider before installing

This skill looks like a thin instruction wrapper for the external Kilo CLI and Puppeteer MCP servers, but the registry metadata doesn't declare that dependency. Before installing or invoking: (1) Confirm you have a trusted, up-to-date Kilo CLI and understand its config at ~/.config/kilo/kilo.json; (2) be cautious about allowing the skill to run with `--auto` — that bypasses interactive approvals and can let automation act without asking you; (3) recognize that runtime use of `npx` will download and execute third-party npm packages (review the packages and their source); (4) do not provide sensitive tokens (GitHub, model provider keys, etc.) unless you trust the environment and have reviewed how they will be used; and (5) ask the skill author to update metadata to declare required binaries/env and to remove or justify the blanket recommendation to use auto-approval so you can make an informed risk decision.

This skill coordinates the Kilo CLI to perform complex execution tasks, from deep codebase refactoring to automated browser interactions.

Core Capabilities

• Heavy-Lifting Coding: Offload file-intensive tasks to the specialized Kilo agent.
• Browser Automation: Control web browsers via Puppeteer MCP integration.
• Session Continuity: Resume or fork existing Kilo sessions to maintain context.
• Autonomous Execution: Run Kilo with --auto for hands-free automation in OpenClaw.

Setting Up Kilo

If Kilo is not yet configured, follow the instructions in setup-guide.md.

Key Config Location: ~/.config/kilo/kilo.json

Workflow Pattern

For a detailed breakdown of the "Brain and Hands" execution strategy, see workflow.md.

1. Basic Task Execution

To run a one-shot task with Kilo:

kilo run --model <model> --auto "<prompt>"

2. Browser Automation Task

To run a browser-based task using the Puppeteer MCP:

kilo run --model <model> --auto "Use your puppeteer MCP tool to navigate to <URL>, <actions>, and report back <results>."

3. Session Management

To list, continue, or fork sessions:

# List all sessions
kilo session list

# Continue the last session
kilo run --continue --model <model> --auto "<next task>"

# Fork a specific session
kilo run --session <sessionID> --fork --model <model> --auto "<experimental task>"

4. Monitoring Progress

If a task takes a long time, use the process tool to poll the Kilo run.

Best Practices

• Specify the Model: Always use --model <provider/model> to ensure consistency (e.g., kilo/minimax/minimax-m2.5:free).
• Use Auto-Approval: Always include the --auto flag when running Kilo from OpenClaw to avoid hanging on permission prompts.
• Prompt Engineering: Be specific about the desired output. Kilo is powerful but benefits from clear, structured instructions.
• Check Stats: Use kilo stats to monitor token usage and costs.