The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides several TypeScript (Bun) and Python scripts (zod-audit.ts, ai_stack_scan.py, zod-run.ts) intended to be executed locally by the agent to perform code audits, migration checks, and schema validation. These scripts are strictly for local repository analysis.
[DYNAMIC_EXECUTION]: The script scripts/zod-run.ts utilizes dynamic import() to load and execute Zod schemas from local modules provided as arguments. This is a functional requirement for exercising schema validation/encoding logic on user-provided code. The documentation explicitly warns users to avoid modules with heavy side effects when using this tool.
[DATA_EXPOSURE_AND_EXFILTRATION]: The scripts/ai_stack_scan.py utility is designed to scan for potentially exposed credentials such as Supabase service-role keys and JWTs. However, it contains robust sanitization logic that redacts these sensitive strings using a [redacted] placeholder before presenting the evidence in its output, ensuring sensitive data is not leaked into the AI context.
[INDIRECT_PROMPT_INJECTION]: As an audit tool, the skill processes untrusted local source files. While this creates a surface for indirect prompt injection (malicious instructions in code comments), the scripts are configured to redact secrets, and the tool is designed for developer-assisted auditing rather than autonomous execution on unknown data.

zod-v4