The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes two local Python scripts, scripts/review_check.py and scripts/selftest.py. These scripts are designed for offline validation of finding structures and self-testing the skill's logic. They use standard Python libraries (e.g., json, argparse, pathlib) and do not perform any network operations, access sensitive system files, or execute external commands.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze and review user-provided content (dramatic scripts, storyboard JSONL files, etc.). This inherently involves processing untrusted data. However, the skill instructions include strong guidelines to use specific identifiers (IDs, line numbers) and short citations rather than broad text interpolation. It explicitly warns against "ghostwriting" or copying full private inputs into reviews, which serves as a mitigation against data-based injection attacks.
[DATA_EXPOSURE]: The skill includes explicit instructions for a 'delivery_privacy' check (Category 10 in the workplan), specifically tasking the agent to look for and report leaks of private credentials, internal paths, or internal process text in the documents it reviews.

short-drama-review