The Agent Skills Directory

[SAFE]: The skill implements developmental review of scholarly works using local Python scripts for rubric validation and scoring.
[DATA_EXFILTRATION]: The code contains no network libraries (such as requests or urllib) and implements strict validation that rejects URLs in local reference fields to prevent accidental data transmission.
[CREDENTIALS_UNSAFE]: No hardcoded credentials or environment variable access found. The skill proactively rejects input JSON containing sensitive keys like email, ssn, password, or phone via a predefined PRIVATE_FIELD_KEYS blacklist in scripts/_common.py.
[DYNAMIC_EXECUTION]: The scripts use fixed schemas and standard library parsers. There is no usage of dangerous functions such as eval(), exec(), or pickle.load().
[COMMAND_EXECUTION]: Although the skill manifest requests the Bash tool, it is only used to provide documented examples for running the local Python scripts. The scripts themselves do not invoke subprocesses or shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (JSON/CSV). It mitigates this attack surface with comprehensive input sanitization, including depth limits, node count limits, and strict schema enforcement. The skill lacks the network or system-level capabilities required for an injection to cause material harm.

scholar-evaluation