The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches interview question banks and roles from the vendor's GitHub repository (github.com/rohitg00/ai-engineering-from-scratch). These are used as content for the mock interviews and represent legitimate vendor-owned resources.
[INDIRECT_PROMPT_INJECTION]: The skill processes external content from local or remote repositories, which serves as a potential attack surface for indirect prompt injection findings.
Ingestion points: SKILL.md describes loading data from a local interviews/ directory or a public GitHub repository.
Boundary markers: Present; the AI is explicitly instructed to only display the "prompt" section of question files and wait for user input, which helps mitigate the risk of the model obeying hidden instructions in external data.
Capability inventory: The skill's capabilities are restricted to reading data and appending to a local state file (INTERVIEW-LEARNING.md). No arbitrary command execution, network exfiltration, or privileged file access capabilities were detected.
Sanitization: Not explicitly implemented; the skill relies on instructional constraints and structural parsing rules to manage external data safely.

interview-prep