The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (validate_abstract.py and write_abstracts_md.py) to perform deterministic character counting and enforce formatting constraints on the generated abstracts. These scripts are bundled with the skill and do not interact with external networks or untrusted binaries.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes user-provided research descriptions. (1) Ingestion points: User-supplied research objects, problems, and content provided via input prompts or the info_form.md template. (2) Boundary markers: Output is delimited by specific marker tags (e.g., [ZH], [EN]) and Markdown headings as defined in config.yaml. (3) Capability inventory: Executes local Python scripts and writes to the local filesystem. (4) Sanitization: The write_abstracts_md.py script contains a _is_safe_filename function that prevents path traversal and restricts file writing to the current working directory, effectively mitigating common file-system-based injection risks.

nsfc-abstract