The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download source code from an external GitHub repository.
Evidence: git clone https://github.com/TianyiDataScience/openclaw-control-center.git in SKILL.md.
[REMOTE_CODE_EXECUTION]: The installation process involves executing external code through Node.js package managers and build scripts.
Evidence: Sequential execution of npm install, npm run build, and npm run dev:ui after cloning the repository.
[COMMAND_EXECUTION]: Multiple shell commands are provided to set up and run the environment.
Evidence: Use of git clone, cd, npm install, cp .env.example .env, npm run build, npm test, and npm run dev:ui.
[DATA_EXFILTRATION]: The skill instructions involve reading and modifying sensitive local files and environment variables.
Evidence: Accessing ~/.openclaw/openclaw.json and managing .env files containing LOCAL_API_TOKEN and security flags.
[PROMPT_INJECTION]: The dashboard acts as a processing layer for untrusted agent data, creating a surface for indirect prompt injection.
Ingestion points: Agent activity, token usage, and collaboration messages are ingested via API endpoints like /api/hall/tasks/${task.id}/messages and /api/usage/metrics (referenced in SKILL.md).
Boundary markers: Absent; no explicit delimiters or instructions to ignore embedded content are mentioned.
Capability inventory: The installation scripts execute shell commands (npm install, npm run dev:ui) and the dashboard performs network operations via fetch (referenced in SKILL.md).
Sanitization: Absent; no validation or escaping of external agent content is described in the documentation.

openclaw-control-center