The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted input in the form of code diffs and skill packages, which presents an indirect prompt injection surface. However, the methodology includes robust defensive instructions requiring the agent to independently verify findings against original source materials and perform an adversarial refutation pass.
Ingestion points: Code diff text and local skill directory contents (SKILL.md, scripts, etc.).
Boundary markers: None explicitly defined for the input data.
Capability inventory: The skill utilizes file read operations and non-destructive command execution for syntax verification.
Sanitization: Instructions mandate direct source code verification and a dedicated step to attempt to disprove candidate findings before reporting them.
[COMMAND_EXECUTION]: The instructions direct the agent to verify command syntax within reviewed packages using help or dry-run flags. It explicitly forbids executing any command with side effects (creation, deletion, or modification) or those requiring credentials.
[EXTERNAL_DOWNLOADS]: The documentation includes an installation command using npx to fetch the skill from the author's official repository (soia-team/soia-open-dev-coding-skills). This is a standard vendor-provided resource.

soia-dev-review-panel