The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages an AMINER_API_KEY to provide optional metadata enrichment services. It includes documentation and setup scripts to guide users on secure configuration. The instructions explicitly command the agent never to print the real token and to warn users against committing secrets to logs or version control.
[COMMAND_EXECUTION]: The skill invokes a local Python renderer located at scripts/render_html.py. This script is used to generate interactive HTML graphs and static SVG maps from analyzed citation data. Analysis of the script confirms it uses only the Python standard library and operates as a deterministic template engine for visualization.
[DATA_EXFILTRATION]: When the user opts into AMiner enrichment, the skill sends paper identifiers to the vendor's official APIs at aminer.cn. This network activity is documented as the primary function of the enrichment feature and targets the author's own infrastructure.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes untrusted content from academic papers.
Ingestion points: Data enters the context through the file parameter (PDF or text paths) or pasted text provided by the user, as specified in SKILL.md.
Boundary markers: The prompts in references/prompts.md use markdown-style tags (e.g., <citations>) to separate untrusted data, though they do not employ unique nonces or unpredictable delimiters.
Capability inventory: The skill has access to powerful tools including Read, Write, Edit, and Bash through its command definition in commands/paper-source-trace.md.
Sanitization: There is no evidence of automated sanitization or escaping of the input paper text before it is interpolated into LLM prompts.

paper-source-trace