The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses npx xapi-to for all tool invocations and provides a sandbox run capability to execute arbitrary shell commands in a remote environment.
[EXTERNAL_DOWNLOADS]: Installation instructions rely on npx, which downloads and executes the xapi-to package from the public npm registry at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill provides extensive tools to ingest data from untrusted sources like Twitter, Reddit, and Weibo. This data is processed by the agent and could contain malicious instructions targeting the skill's powerful capabilities.
Ingestion points: Multiple social media API endpoints (e.g., twitter.tweet_detail in guides/twitter.md, reddit.api_v1_reddit_app_fetch__post__comments in guides/reddit.md).
Boundary markers: Instructions recommend using get for schema understanding but do not mandate specific delimiters or 'ignore instructions' wrappers for untrusted content.
Capability inventory: Includes shell execution (sandbox run), file writes, DNS modification, and code deployment (workers push).
Sanitization: No explicit sanitization or filtering of external content is described before interpolation into the agent's context.
[DYNAMIC_EXECUTION]: Managed sandboxes and hosted workers allow for the runtime execution and deployment of arbitrary code, which is a significant capability surface if the agent's logic is compromised by external data.
[CREDENTIALS_UNSAFE]: The skill manages a sensitive API key (XAPI_KEY), which is automatically saved to the local filesystem at ~/.xapi/config.json during the registration process.

xapi