The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the $ARGUMENTS variable (aliased to site_path) directly within shell command strings across multiple steps (e.g., mkdir -p {site_path}, uv run ... {site_path}/...). Since the input is not validated or escaped, a malicious user could provide a path string containing shell operators (like ;, |, or &) to execute arbitrary commands with the agent's privileges.\n- [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by downloading and analyzing untrusted HTML content from external websites using sub-agents.\n
Ingestion points: Raw and rendered HTML files sourced from external URLs and stored in the {site_path} directory.\n
Boundary markers: None. The sub-agent prompts (e.g., in Step 4) do not use delimiters or instructions to ignore embedded commands within the HTML files.\n
Capability inventory: The skill has access to shell execution (bash, uv run), file system operations (Read, Write), and agent orchestration (Agent).\n
Sanitization: No sanitization or filtering of the HTML content is performed before it is processed by the analysis agents.

scrape-spec