The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for untrusted data via scripts/fetch_page.py. SKILL.md provides boundary markers by explicitly instructing the agent to treat scraped content as untrusted and ignore embedded instructions. The skill's capabilities include network access and file system writes (via reporting scripts). Sanitization is implemented through instructions requiring the agent to label and verify evidence rather than through automated content filtering.
[COMMAND_EXECUTION]: Python scripts provided for brand scanning and citability scoring perform deterministic regex operations on text files. They do not invoke shell commands with untrusted input or use dangerous execution functions.
[DYNAMIC_EXECUTION]: A dynamic import of the sys module (__import__('sys')) is used in scripts/citability_scorer.py to read from standard input. This is a standard utility pattern and does not represent a malicious execution vector.

geo