The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions advise the agent to tell the user to execute a remote script using curl -fsSL https://www.lazyweb.com/install.sh | bash. This is a high-risk pattern that executes unverified code with the user's shell privileges, bypassing standard package management and verification processes.
[CREDENTIALS_UNSAFE]: The skill and its included Python scripts (fetch-evidence.py, generate-prototypes.py) access sensitive files in the ~/.lazyweb/ directory, including lazyweb_mcp_token and openai_api_key. Furthermore, the skill instructs the agent to read the INTEGRITY file and pass its content as a tool argument, exposing local system identifiers and fingerprints to the AI model's context.
[COMMAND_EXECUTION]: The skill executes several local scripts and binaries. The generate-prototypes.py script uses subprocess.run to call the codex command-line tool. The SKILL.md also contains shell logic for browser setup, file system operations, and image uploads via curl.
[DATA_EXFILTRATION]: The skill's scripts use urllib.request to transmit data to lazyweb.com and api.openai.com. This data includes the locally retrieved tokens, system fingerprints, and search queries.
[PROMPT_INJECTION]: The skill processes untrusted data from external design database results and web search output (ingestion points). It lacks explicit boundary markers or delimiters when interpolating these results into report templates or internal prompts. The skill possesses extensive capabilities including shell execution, file writing, and network operations (capability inventory). There is no evidence of sanitization or filtering of the external content before it is processed (sanitization absent).

lazyweb-design-create