The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a set of built-in shell scripts in the bin/ directory (e.g., gap-validate-finding, gap-orientation-preflight) to perform validation and codebase orientation. These scripts are properly scoped and utilize standard tools like git, grep, awk, and rg with argument validation to prevent shell injection.
[EXTERNAL_DOWNLOADS]: The gap-orientation-preflight script manages local checkouts by performing git clone and git fetch operations. These operations target configured repositories, typically hosted on GitHub, which is a well-known and trusted service.
[PROMPT_INJECTION]: The skill explicitly defines an 'Untrusted content boundary' in SKILL.md. It instructs the agent to ignore any directives found in analyzed materials (such as 'ignore previous instructions') and report them as suspected injections, effectively neutralizing indirect prompt injection vectors.
[DATA_EXFILTRATION]: Network activity is restricted to authenticated repository access and metadata retrieval via a tracker interface. The skill instructions specifically forbid the exfiltration of sensitive data or access to unauthorized credential stores.

om-gap-analysis