The Agent Skills Directory

[NO_CODE]: The skill is composed entirely of markdown documentation and templates. No executable scripts (Python, JavaScript, or Shell) are included, significantly reducing the attack surface.
[EXTERNAL_DOWNLOADS]: Several GitHub repositories are listed in the 'references/harness-source-notes.md' file. These are explicitly defined as research citations used to develop the skill's methodology and are not runtime dependencies or installation targets.
[DATA_EXFILTRATION]: The skill incorporates strong privacy guidelines, instructing the agent not to save sensitive information, publish transcripts, or mutate projects without explicit user authorization. No sensitive system paths or hardcoded credentials were detected.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external data (code and documentation) and user input, it includes structured safety measures such as a 'Question Contract' and a 'Closure Gate' to verify intent and maintain focus, effectively managing the data ingestion surface.
[DYNAMIC_CONTEXT_INJECTION]: The 'SKILL.md' file does not utilize dynamic command execution syntax (e.g., '!command'), ensuring that no shell commands are executed during skill loading.

deep-interview