The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute several Python scripts (scripts/site_crawler.py, scripts/dfs_client.py, and scripts/report_publish.py) using subprocess calls. These commands include placeholders for user-supplied data such as URLs and report filenames.
Evidence in SKILL.md includes:
python scripts/site_crawler.py --url <url> --trace-redirects --pretty
python scripts/dfs_client.py backlinks --target https://example.com/old-page --pretty
python scripts/report_publish.py <report>.md
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (URLs, CSV files of redirects) and incorporates them into command-line arguments. This presents a potential attack surface for command injection or prompt manipulation if a user provides a malicious URL containing shell metacharacters.
Ingestion points: User-provided list of URLs and CSV files of old/new URL pairs described in the 'Inputs' section of SKILL.md.
Boundary markers: None identified. The instructions do not specify using delimiters or 'ignore embedded instructions' warnings when handling the external URLs.
Capability inventory: The skill uses subprocess execution via python for crawling, backlink analysis, and report generation.
Sanitization: The instructions do not explicitly mention sanitization, validation, or escaping of the user-provided URLs before they are passed to the script arguments.

redirect-analysis