The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of the opencli command and various passthrough commands to external tools like gh, docker, and vercel. It also allows for the registration and execution of local custom CLI tools.
[EXTERNAL_DOWNLOADS]: The skill promotes the installation of the @jackwener/opencli npm package and provides an internal command (opencli install) to download and install additional third-party CLI tools.
[DATA_EXFILTRATION]: The skill provides capabilities to read highly sensitive information from logged-in user accounts across nearly 80 platforms, including private messages, browsing history, and personal profiles. While no direct exfiltration is observed, the extensive data access poses a significant risk if the agent is compromised.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection due to the ingestion of content from numerous external websites and applications.
Ingestion points: Content is retrieved through commands like web read, youtube transcript, and various site-specific search and read commands.
Boundary markers: There are no specified delimiters or instructions to ignore malicious instructions embedded in the retrieved data.
Capability inventory: The agent can perform actions like posting to social media, sending messages, and executing system commands via passthrough CLIs.
Sanitization: The documentation does not mention any sanitization or validation of data retrieved from external sources.
[REMOTE_CODE_EXECUTION]: The 'Self-Repair' feature encourages the agent to modify the tool's TypeScript source code and retry execution, which is a form of self-modifying code behavior.

qiaomu-opencli-usage