The Agent Skills Directory

[PROMPT_INJECTION]: The skill requires the agent to execute browser-act get-skills core and follow "operational directives" contained in the command's output. This creates a multi-step chain where external tool output dictates the agent's next instructions, establishing an indirect prompt injection surface.
Ingestion points: Output from the browser-act CLI tool in SKILL.md.
Boundary markers: None present; the skill explicitly forbids the agent from truncating output, which could otherwise act as a safety boundary.
Capability inventory: The agent has access to the Bash tool to execute browser interactions, navigation, and network capture.
Sanitization: No sanitization or validation of the directives received from the tool output is specified.
[PROMPT_INJECTION]: The instructions use authoritative directives such as "Do NOT skip this step" and "Do NOT truncate the output" to override the agent's default processing and context management procedures.
[EXTERNAL_DOWNLOADS]: The skill's metadata instructs the user to install the browser-act-cli package from PyPI using the uv tool.
[COMMAND_EXECUTION]: The skill functions by executing shell commands through a dedicated CLI tool to perform browser automation tasks.
[DATA_EXFILTRATION]: The skill documents the transmission of captcha challenge images to an external API (browseract.com) for verification assistance. Although limited to challenge images, this constitutes a network operation to a non-whitelisted domain.

browser-act