The Agent Skills Directory

CREDENTIALS_UNSAFE (MEDIUM): SKILL.md provides an example diagnostic command env | grep IDENTITY to verify if secrets are correctly propagated through build layers. This command outputs the full environment variable string (e.g., IDENTITY=secret_value), which would cause sensitive credentials to be captured in diagnostic logs.
COMMAND_EXECUTION (LOW): The find-polluter.sh script automates the execution of test files using npm test based on a user-provided search pattern. This involves the execution of arbitrary code contained within test files on the local filesystem.
DATA_EXPOSURE (LOW): SKILL.md includes instructions to run macOS-specific commands (security list-keychains, security find-identity) that expose metadata about the system's keychain and cryptographic identities.
INDIRECT_PROMPT_INJECTION (LOW): The skill includes several 'Pressure Test' markdown files (test-pressure-1.md, etc.) designed as adversarial scenarios to test agent compliance. These represent a self-referential ingestion surface.
Ingestion points: Files test-pressure-1.md, test-pressure-2.md, and test-pressure-3.md.
Boundary markers: Absent; the files contain raw prompt-like scenarios without delimiters.
Capability inventory: Execution of shell commands via examples in SKILL.md and the find-polluter.sh script.
Sanitization: No sanitization or validation of the input scenarios is performed.

diagnostic-analysis