The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill is designed to ingest and process academic papers, which are untrusted external data sources.
Ingestion points: The agent uses Read, Grep, and Glob tools to load the content of papers provided by the user into its context.
Boundary markers: The instructions lack explicit delimiters or system-level warnings to ignore natural language instructions that might be embedded within the papers being analyzed.
Capability inventory: The skill has access to high-impact tools including Bash, Write, Edit, WebFetch, and Task. If an attacker embeds instructions in a paper (e.g., in a LaTeX comment or hidden text), the agent might execute those commands via the Bash tool.
Sanitization: There is no evidence of sanitization, escaping, or validation of the paper's content before it is processed or used to generate subsequent tool arguments.
[Command Execution] (SAFE): The Bash tool is enabled, which is high-risk. However, in the context of a 'paper-workflow' orchestrator, this is likely intended for file management and running local audit scripts, matching the primary purpose of the skill.

paper-workflow