The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's auditing and validation workflow involves processing untrusted data from the repository, which is a standard surface for indirect prompt injection.
Ingestion points: The detect-gaps.sh script scans the filesystem (Assets/Game/, design/gdd/), and validate-commit.sh reads git staged changes via git diff --cached during the commit process.
Boundary markers: Absent in the utility scripts; however, agent definitions (e.g., ai-programmer.md, unity-specialist.md) provide strict path-based isolation and role boundaries.
Capability inventory: Multiple agents are authorized to use the Bash tool for local operations. Utility hooks execute powershell.exe for UI notifications and python for JSON validation.
Sanitization: notify.sh implements single-quote escaping for values passed to the PowerShell command-line string to mitigate command injection risks.
[COMMAND_EXECUTION]: The skill uses local system commands to provide developer feedback and ensure data integrity.
Evidence: The notify.sh script executes powershell.exe to display Windows balloon tip notifications. The validate-assets.sh and validate-commit.sh hooks use python -m json.tool to validate the syntax of configuration files in Assets/Game/Data/.

game-studio-orchestration