The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill recommends installing 'google-generativeai' and 'gh'. These are trusted tools from established organizations (Google and GitHub).
[COMMAND_EXECUTION] (MEDIUM): The workflow relies on a 'gemini' executable that is not installed by the listed prerequisites. Using unverified binaries from unspecified sources is a security risk as they could perform arbitrary actions on the system.
[PROMPT_INJECTION] (LOW): (Category 8: Indirect Prompt Injection) The skill ingests untrusted data from GitHub issues via 'gh search issues' and interpolates it into prompts.
Ingestion points: Output from 'gh search issues' is passed directly into the LLM prompt.
Boundary markers: None identified; the content is directly concatenated into the prompt string.
Capability inventory: The skill provides text-based diagnostics and code fix recommendations but does not automate execution of those fixes.
Sanitization: None; external content is used without filtering.
Risk: A malicious GitHub issue could contain instructions designed to manipulate the agent's reasoning, although the impact is limited by the lack of autonomous execution capabilities in the provided scripts.

debug