The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/scrape_thn.py fetches data from external sources including https://feeds.feedburner.com/TheHackersNews and https://thehackernews.com. This involves ingesting content from third-party websites into the agent's environment.
[COMMAND_EXECUTION]: The script scripts/update_blocklist.py automatically modifies execution/security_scan.py and SECURITY_GUARDRAILS.md. It generates Python code snippets using data extracted from external articles and injects them directly into these files. The generate_scan_entry function in scripts/update_blocklist.py does not sanitize the name or pattern fields, which could allow a malicious external source to inject arbitrary Python code into the security_scan.py script.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). It ingests untrusted data from external websites (ingestion point: scripts/scrape_thn.py), lacks boundary markers to isolate this data from the agent's execution logic, and possesses significant capabilities including the ability to write to and modify system security scripts (capability: file-write in scripts/update_blocklist.py). Sanitization of the ingested data is minimal and insufficient to prevent manipulation of the resulting code.

supply-chain-monitor