The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of git diffs, which could contain malicious instructions embedded in source code or commit messages designed to influence the AI's behavior.
Ingestion points: Content is gathered via git diff and file reads in scripts/git-review.py.
Boundary markers: The output is structured with specific headers and file markers (e.g., === filename (status) ===) to distinguish diff content from metadata.
Capability inventory: The skill utilizes file system tools (Read, Write, Edit), git, and terminal interaction capabilities to process and act on these annotations.
Sanitization: While the script cleans technical metadata from the diff for readability, it does not sanitize the actual code content against adversarial instructions.
[COMMAND_EXECUTION]: The script executes various system commands to manage the interactive review process.
It invokes git for repository operations and interacts with terminal emulators (agterm, tmux, kitty, wezterm) to provide an overlay editor interface.
The script employs shlex.quote to escape command arguments and shutil.which to resolve binary paths, which helps prevent command injection via environment variables like $EDITOR.

git-review