The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a cross-model judgment mechanism that sends the full content of the document under review to external well-known services (OpenAI, Anthropic, xAI, and Cursor) via their official command-line tools. This egress is a documented core feature but involves transmitting potentially sensitive user data to third parties.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents from GitHub issues and local filesystem paths. These inputs are used as context for sub-agents, creating a surface for indirect prompt injection where malicious instructions embedded in documents could influence the agent's behavior or automated fix suggestions.
[COMMAND_EXECUTION]: The skill utilizes several external command-line utilities, including the GitHub CLI (gh), Anthropic's claude, OpenAI's codex, xAI's grok, and the cursor-agent. These tools are invoked for document retrieval, update synchronization, and cross-model analysis.
[DYNAMIC_EXECUTION]: The peer-job-runner.py script manages background tasks using detached processes. It incorporates security safeguards such as verifying file ownership (os.fstat compared against os.geteuid) and using restrictive file permissions (0600) to mitigate risks like symlink attacks or unauthorized access in shared temporary directories.
[EXTERNAL_DOWNLOADS]: The skill interacts with various well-known services and their official package/CLI ecosystems. Specifically, it fetches data from GitHub and utilizes AI model endpoints from major providers. These interactions are conducted through established and trusted service interfaces.

doc-review