The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a local Python script (数据追踪/拉取番茄数据.py) to fetch analytics data from supported platforms. This script is intended to be a local component of the skill, and its use is necessary for retrieving the data required for analysis.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes user-supplied chapter text, which creates a theoretical surface for indirect prompt injection if the prose contains malicious instructions.
Ingestion points: Project files containing novel prose (chapter text) are read and processed by the story-data-text-improvement-planner agent.
Boundary markers: The skill uses a rigid state-machine architecture where text reading only occurs after a separate agent has verified the analytics data; all agent handoffs are structured via strictly enforced JSON schemas.
Capability inventory: Agent roles assigned to analyze prose are restricted to read-only file system tools (Read, Glob, Grep) and are explicitly barred from executing shell commands or initiating network operations.
Sanitization: Although the prose itself is not filtered, the agent is instructed to treat the text as passive diagnostic data and is constrained to outputting structured improvement proposals, which significantly mitigates the risk of the model executing embedded instructions.

story-data-analyze