The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): Indirect Prompt Injection via untrusted data ingestion. ● Ingestion points: references/interfaces-integration.md describes a 'Standalone Quick Optimization' workflow that fetches job descriptions from user-provided URLs. ● Boundary markers: Absent. The instructions do not specify any delimiters or safety headers when processing the fetched content. ● Capability inventory: SKILL.md defines allowed-tools: Read,Write,Edit,WebFetch,WebSearch. ● Sanitization: Absent. The agent is instructed to directly extract keywords and skills from the untrusted content. ● Risk: An attacker could host a job description containing instructions to hijack the agent's logic and exfiltrate the user's career data using the WebFetch tool.
[EXTERNAL_DOWNLOADS] (MEDIUM): Reference to an untrusted external source. ● Evidence: SKILL.md and references/interfaces-integration.md both link to github.com/erichowens/cv-creator as a 'Production Implementation'. ● Risk: This repository is not on the trusted scope list. Users or agents following these instructions are exposed to unverified code and recommended execution of npm install.

cv-creator