The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes user-supplied drama scripts (剧本.md) which are then analyzed by local Python scripts. This creates a surface for indirect prompt injection where malicious instructions embedded in the drama script could potentially influence the agent's behavior or the execution of the indexing/estimation tools.
Ingestion points: Processes external script files (剧本.md) and user prompts containing script content as described in the workflow in SKILL.md.
Boundary markers: The skill defines several rule levels such as structural_invariant and reviewed_invariant to guide the agent and reviewer, which act as conceptual boundaries during processing.
Capability inventory: The skill utilizes local shell execution capabilities to run Python scripts (screenplay_index.py, duration_estimate.py, voice_sheet_check.py) for indexing, estimating duration, and verifying voice sheets.
Sanitization: There is no explicit evidence of sanitization for user-influenced variables, such as character names or file paths, when the agent constructs shell commands to execute the included scripts.
[COMMAND_EXECUTION]: The skill involves the execution of local Python scripts to perform metadata indexing and duration estimation. The instructions in SKILL.md direct the agent to generate and run shell commands (python3) to interact with these scripts. While the scripts perform benign tasks, the execution of commands using parameters derived from user input is a security-relevant behavior.

short-drama-write