The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external, untrusted drama project files (scripts, storyboards, and prompts). This creates a vulnerability surface for indirect prompt injection where a malicious file could contain instructions to bypass review guidelines or manipulate the agent's output.
Ingestion points: Reads various project-specific files including 剧本.md, 分镜.md, 视频提示词.md, and asset definitions in references/.
Boundary markers: The skill employs an 'Evidence-Based Review Method' (REV-01 to REV-14) that requires the agent to quote evidence and use specific Markdown structures, which provides some delimiter-based separation but lacks formal sanitization.
Capability inventory: The agent can write review artifacts (Markdown and JSONL files) and is instructed to execute local validation scripts (scripts/review_check.py).
Sanitization: No explicit sanitization or escaping instructions are provided for handling content extracted from source files.
[COMMAND_EXECUTION]: The skill instructions include the execution of local Python scripts for validation, maintenance, and indexing tasks.
Evidence: SKILL.md contains instructions to run python3 scripts/selftest.py for installation and troubleshooting.
Evidence: references/rubric-source-analysis.md provides bash commands for verifying and checking the coverage of novel indexes using a local script (novel_index.py).

short-drama-review