The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates entirely through the superagent command-line utility, using it to manage repositories, worktrees, and terminals (e.g., superagent status, superagent worktree create, superagent terminal send).
[DATA_EXFILTRATION]: The skill exposes sensitive data access points through browser and terminal automation features.
Evidence:
superagent cookie get: Retrieves browser cookies.
superagent terminal read: Accesses terminal output history.
superagent snapshot and superagent screenshot: Captures page content and visual state.
superagent capture start: Intercepts console and network logs.
[REMOTE_CODE_EXECUTION]: The skill includes commands for dynamic code execution within the browser environment.
Evidence:
superagent eval --expression <js>: Executes arbitrary JavaScript in the page context.
superagent exec --command ...: Executes instructions via the automation engine's command surface.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted data from external sources.
Ingestion points: Terminal output (terminal read), accessibility trees (snapshot), and network/console logs (console, network).
Boundary markers: Absent. The instructions do not specify the use of delimiters or warnings to ignore instructions embedded in the ingested content.
Capability inventory: The agent can modify the filesystem (worktree create/rm), interact with terminals (terminal send), and perform browser actions (goto, click, eval, cookie delete).
Sanitization: Absent. There is no requirement to sanitize or validate content before processing.

superagent-cli