The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Python script (validate_prior_art.py) located in a sibling package's directory (code-prior-art-survey). Shell-level execution of scripts from external directories using variable paths creates a target for path manipulation or command injection.
[DYNAMIC_EXECUTION]: The execution path to the validation script is resolved dynamically at runtime (<producer-package>/scripts/). This dynamic loading from computed paths is a identified as a medium-severity security concern.
[INDIRECT_PROMPT_INJECTION]: The skill reviews artifacts (YAML/Markdown) that incorporate data from external repositories, such as descriptions, terms, and metadata. This ingested content is untrusted and could contain hidden instructions intended to influence the agent's VERDICT: approve|revise decision or other downstream behaviors. The skill mandates treating such content as data rather than instructions to mitigate this risk.
[EXTERNAL_DOWNLOADS]: The review workflow encourages "spot-checks" against external repository pages and potentially cloning repositories to verify information. This involves fetching and processing data from remote, untrusted sources during the skill's operation.

reviewing-code-prior-art-survey