The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it instructs the agent to systematically process and evaluate untrusted external data (scientific manuscripts and grant proposals). Malicious instructions embedded within these documents could potentially influence the agent's assessment or override its instructions. * Ingestion points: The agent reads and analyzes manuscript sections as part of the core workflow defined in SKILL.md. * Boundary markers: There are no explicit instructions or delimiters defined to separate the manuscript content from the agent's instructions, nor are there warnings to the agent to ignore any instructions found within the manuscript. * Capability inventory: While the skill itself is purely instructional, the risk depends on the broader capabilities of the agent which could be abused if an injection is successful. * Sanitization: No techniques for sanitizing or filtering the content of the manuscripts are described in the skill files.
[NO_CODE]: The skill is composed entirely of informational markdown files (SKILL.md, references/common_issues.md, and references/reporting_standards.md). It contains no scripts, configuration files for package managers, or binary executables that would lead to code execution.

peer-review