The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides documentation for Bun.$, a shell execution API, and includes logic for a self-update command using npx skills update.
[EXTERNAL_DOWNLOADS]: The update mechanism performs network requests to the vendor's GitHub repository (raw.githubusercontent.com/abpai/skills) to check for version updates.
[REMOTE_CODE_EXECUTION]: Includes the standard installation command for Bun, which uses a piped shell pattern (curl | bash) from the official bun.sh domain.
[PROMPT_INJECTION]: Contains meta-instructions directing the agent to perform "silent" version checks and handle update logic conditionally during the session.
[DATA_EXFILTRATION]: An indirect prompt injection surface exists due to the combination of data ingestion (via Bun.serve routes or WebSockets) and high-privilege capabilities such as shell execution and file system access.
Ingestion points: Data enters the agent's context through HTTP request parameters, JSON bodies, and WebSocket messages as defined in references/builtin-apis.md.
Boundary markers: No explicit boundary markers or "ignore instructions" warnings are included in the provided code snippets to prevent the agent from following instructions embedded in external data.
Capability inventory: The skill documents access to Bun.$ (shell execution), Bun.write (file system writes), sql (database queries), and Bun.cron (persistence via scheduled tasks).
Sanitization: Documentation notes that shell interpolations in Bun.$ are automatically escaped, which mitigates simple command injection but does not eliminate prompt injection risks.

bun-expert