The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script, scripts/loop.py, to manage task logic and state transitions. This script is a deterministic decision reducer that strictly validates JSON inputs against a defined schema and calculates integrity hashes (SHA-256) to ensure the consistency of the task ledger. It does not perform network operations or unauthorized file access.
[PROMPT_INJECTION]: The skill incorporates comprehensive defensive designs to mitigate injection risks. It establishes a 'Canonical Authority Order' that prioritizes platform and safety rules over user guidance and model inference. Furthermore, it implements a 'Safety and Authority' theorem for behavioral memory, requiring explicit user confirmation for any instructions retrieved from past sessions to prevent 'taint' or behavior overrides.
[DATA_EXFILTRATION]: No patterns associated with data exfiltration or unauthorized access to sensitive system files were found. The skill leverages host-provided durable storage (S2/S3 state classes) for persistence, ensuring that all data handling is confined to authorized environments.
[REMOTE_CODE_EXECUTION]: The skill does not perform remote code downloads or execution. It relies entirely on local scripts and established platform capabilities to perform its functions.
[CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or secrets were identified within the skill's instructions or scripts. The documentation explicitly advises redacting sensitive data and secrets from the task ledger.

nerd-loop