The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to run the command curl -fsSL https://www.lazyweb.com/install.sh | bash. This represents a critical risk as it allows a remote server to execute arbitrary commands on the user's host without verification or oversight.
[COMMAND_EXECUTION]: The skill uses the Bash tool to run cat "$HOME/.lazyweb/VERSION". This demonstrates the ability to read from local hidden configuration directories, which can be a precursor to more sensitive file access or reconnaissance.
[DATA_EXFILTRATION]: The skill performs network requests to https://www.lazyweb.com/mcp and other sub-resources to send base64-encoded screenshots and product context. As this domain is not a whitelisted or established trusted technology service, it is flagged as an unverified data transfer path.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by ingesting external data (sign-up screenshots and product descriptions) which then influences the behavior of internal tools.
Ingestion points: image_b64 and context arguments in the signup_design_run tool call within SKILL.md.
Boundary markers: Absent; there are no specific instructions provided to the agent to ignore or delimit instructions found within the processed images or context.
Capability inventory: The skill is authorized to use Bash, Read, Write, Glob, and Grep tools.
Sanitization: Absent; the skill does not specify any validation or filtering mechanisms for the data retrieved from or sent to the external service.

lazyweb-optimize-sign-up