awesome-openclaw-skills-zh
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The documentation instructs users to clone the skill library from GitHub (github.com/clawdbot-ai/awesome-openclaw-skills-zh.git) and install additional skills from the project's repository (clawdhub.com).
- [REMOTE_CODE_EXECUTION]: Provides instructions for installing remote skills via the
openclaw skill installcommand, which fetches and integrates external code into the agent's environment. - [COMMAND_EXECUTION]: Includes shell command examples for system configuration, skill installation, and environment variable management (e.g.,
export EMAIL_PASSWORD,source ~/.openclaw/.env). - [INDIRECT_PROMPT_INJECTION]: The skill documents tools that process untrusted external data such as emails and shared documents, creating a potential surface for indirect prompt injection.
- Ingestion points: Data entry occurs through methods like
AppleMailSkill.search_emails,GoogleWorkspaceSkill.gmail.get_messages, andGoogleWorkspaceSkill.drive.list_files. - Boundary markers: No explicit delimiters or instructions to ignore embedded prompts are visible in the provided code snippets.
- Capability inventory: The described skills possess capabilities for network communication and file system operations.
- Sanitization: The skill library mentions a specific 'email-prompt-injection-defense' task, indicating that defenses are intended to be managed by supplementary skills.
Audit Metadata