The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to read sensitive local configuration files.
Evidence: The skill explicitly directs the agent to read api/src/environments/env.local.yml to determine the database type from the DATABASE_URL field.
Risk: Accessing environment files like .env or env.local.yml exposes sensitive credentials and connection strings to the model's context, increasing the risk of exposure.
[COMMAND_EXECUTION]: The skill encourages the execution of shell commands and database migrations.
Evidence: The documentation suggests running cd api && bun run db:up-dev to apply generated migration files.
Risk: If the migration files generated by the agent contain malicious SQL (e.g., due to indirect injection), these commands will execute that SQL against the developer's database.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of external project documentation.
Ingestion points: The skill parses PRD documents located at paths like docs/spec/{년}/{월}/pdj-*.md (SKILL.md).
Boundary markers: Absent. There are no instructions or delimiters provided to ensure the agent ignores instructions embedded within the PRD data.
Capability inventory: The skill has the capability to write files to the local filesystem (api/db/migrations/) and execute shell commands (bun run db:up-dev).
Sanitization: Absent. The skill maps input fields from the documentation directly into SQL templates without validation or escaping logic.

peach-gen-db