The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill templates reference external GitHub Actions (trufflesecurity/trufflehog and reviewdog/action-detect-secrets). These organizations are not included in the 'Trusted GitHub Organizations' whitelist, making them unverifiable dependencies in this audit context.
COMMAND_EXECUTION (MEDIUM): The provided shell validation script uses indirect variable expansion (${!var}). While intended for validation logic, dynamic variable resolution is a form of dynamic code generation that can be exploitable if an attacker can influence the variable names being processed.
PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection through environment variables that it processes for validation.
Ingestion points: Environment variables are iterated and evaluated in the REQUIRED_VARS loop and the TypeScript validateEnv function.
Boundary markers: None are used to separate untrusted data from processing logic.
Capability inventory: The environment allows for network operations via curl and control flow manipulation via exit commands.
Sanitization: There is no evidence of sanitization or content validation beyond checking for variable existence.
DATA_EXFILTRATION (SAFE): The skill uses curl with secrets targeting a placeholder domain (api.example.com). It correctly implements ::add-mask:: to ensure sensitive values are redacted from CI/CD logs.

secrets-env-manager