The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a suite of local Python scripts located in the scripts/ directory, such as audit_manuscript_state.py and audit_prose_patterns.py. These are used to audit manuscript files for terminology, prose patterns, and document structure. These scripts are executed by the agent as part of its normal operation and do not involve remote downloads or shell piping.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external manuscript data (DOCX, Markdown, etc.). While this represents an ingestion point for untrusted data, the instructions emphasize rigorous scientific integrity and treat manuscript content strictly as data for auditing purposes. This minimizes the risk of the agent interpreting embedded text as instructions.
Ingestion points: Local manuscript files and project state JSON files are read by the agent and processed by the provided scripts.
Boundary markers: The agent is instructed to use structured 'managed-project mode' and 'writing contracts', which provide a controlled environment for data analysis.
Capability inventory: The provided scripts can read local files and write a template state file. No network capabilities or arbitrary code execution paths were found.
Sanitization: Scripts utilize standard library modules like re for regex and xml.etree.ElementTree for DOCX parsing, ensuring safe data processing.
[DATA_EXPOSURE]: The skill accesses local project files (manuscripts and state files) to perform its functions. This access is limited to the local file system for the user's project, with no evidence of data exfiltration to external domains.

academic-writing-skills-zh-cn