The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates a multi-step testing pipeline by executing various command-line tools, including terraform, tofu (OpenTofu), rg (ripgrep), rsync, perl, uv, and the GitHub CLI (gh).
[COMMAND_EXECUTION]: Several local scripts within the repository are invoked to perform specialized checks, such as scripts/validate_tailscale_large_scale_examples.py, scripts/validate_v3_final_polish_examples.py, smoke_v3_plan_matrix.py, and scripts/destroy.sh.
[DATA_EXFILTRATION]: The 'AI-Assisted Review' section includes commands that send the output of git diff to external AI platforms (OpenAI/Codex and Google Gemini). This is a functional feature for automated code review but involves transmitting local source code changes to third-party services.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by interpolating untrusted git diff data directly into prompts for external AI models.
Ingestion points: The output of $(git diff) is passed as a string argument to the codex and gemini commands in the review section.
Boundary markers: Absent; no delimiters or instructions are used to distinguish the code diff from the surrounding prompt instructions.
Capability inventory: The agent uses external AI models to analyze and provide feedback on the code.
Sanitization: The raw output of the git command is used without escaping or filtering.

test-changes