The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from the local environment into AI prompts, creating a surface for indirect prompt injection.
Ingestion points: Reads local source files (e.g., src/file.ts) and log files (e.g., logs/app.log) using cat within shell command substitutions to build prompts for the gemini CLI tool.
Boundary markers: Ingested file content is interpolated directly into the prompt string without using delimiters or instructions to the AI to ignore instructions embedded within the data.
Capability inventory: The workflow uses tools with significant capabilities, including codebase searching (rg), version control (git), GitHub interactions (gh), and AI analysis (gemini).
Sanitization: No sanitization, escaping, or filtering is applied to the content of the files before they are sent to the AI.
[COMMAND_EXECUTION]: The skill relies on a wide array of shell commands to perform its functions. This includes searching codebases with rg, interacting with GitHub via gh, and managing complex multi-terminal diagnostics using background processes (&), output redirection to /tmp/, and the wait command.
[EXTERNAL_DOWNLOADS]: The documentation instructs the user to install the google-generativeai Python package. This is a legitimate package from a well-known service provider used for the skill's primary AI reasoning capabilities.

debug