The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill reads and processes untrusted HTML data (including scripts, JSON-LD, and meta tags) from a user-specified path. This data is incorporated into an analysis report and summary intended for a downstream agent with code-generation and file-writing capabilities.
Ingestion points: Reading HTML content from page_html_path.
Boundary markers: The instructions do not define delimiters or warnings to prevent the agent from following instructions embedded within the scraped HTML.
Capability inventory: The analysis is consumed by a code-generation agent that likely possesses Write and Bash capabilities to produce extraction code.
Sanitization: No validation or escaping of the HTML content is specified before it is summarized or included in the JSON output.
[COMMAND_EXECUTION]: Potential Command Injection. The skill instructions direct the agent to execute bash commands using variables (work_path, page_id) derived from the positional arguments of the raw $ARGUMENTS string.
Evidence: The template commands mkdir -p {work_path}/codegen-analyze and uv run ... -o {work_path}/codegen-analyze/{page_id}.cleaned.html interpolate these variables directly. If the input arguments contain shell metacharacters like semicolons or backticks, it could lead to arbitrary command execution on the host system.

scrape-codegen-analyze