The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions explicitly direct the agent to recommend a high-risk command to the user for installation or updates: curl -fsSL https://www.lazyweb.com/install.sh | sh. This pattern involves downloading a remote script and piping it directly into the system shell, which is a significant security risk as it executes unverified code.
[DATA_EXFILTRATION]: The skill is designed to build diagrams from 'real code' and 'real data payloads' found in the user's environment. This extracted information is then submitted to an external hosted service (lazyweb.com) via the lazyweb_propose_ui_changes tool, representing a transmission of potentially sensitive intellectual property or data to a third-party domain.
[PROMPT_INJECTION]: The skill processes data from an external source to determine its next actions. The agent polls lazyweb_get_proposal_decision to retrieve user verdicts from the external site. Because the agent is instructed to 'Apply' these changes using Write and Bash tools, a compromised or malicious response from the API could lead to indirect prompt injection, where the agent executes unintended file modifications or commands based on external input.
Ingestion points: The lazyweb_get_proposal_decision tool polls the external lazyweb.com API for decision data.
Boundary markers: No specific delimiters or safety instructions are provided to distinguish between the decision status and potentially malicious payloads in the API response.
Capability inventory: The skill has access to Write and Bash tools to modify the local environment.
Sanitization: The instructions lack requirements for sanitizing or validating the content of the 'accepted' suggestions before they are applied to the code.
[COMMAND_EXECUTION]: The skill uses powerful system tools (Bash, Write) to implement changes derived from remote API data, increasing the impact of any potential injection or exfiltration vulnerability.

lazyweb-propose-ui-changes