The Agent Skills Directory

[COMMAND_EXECUTION]: The script render_relation_chart.py executes the system utility fc-list via subprocess.run to identify available Chinese fonts for rendering relationship charts. This call is hardcoded to a specific system binary and does not involve user-supplied arguments, serving the legitimate purpose of ensuring chart readability.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external text files containing novel content provided by the user (located in the 原文/ directory). While the skill's instructions explicitly direct the agent to remain objective and only use evidenced facts, the processing of untrusted external content represents a surface for potential indirect prompt injection.
Ingestion points: Novel content is read from local files specified by the user and stored in the 原文/ directory.
Boundary markers: The skill utilizes structural chapter markers and regular expressions to segment and index the content.
Capability inventory: The skill has the ability to write structured markdown and CSV files to the local filesystem and execute its bundled Python scripts.
Sanitization: Data is processed using character decoding and regex-based boundary detection to isolate chapter content.

story-long-analyze