The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the hunk CLI tool and a bundled Python helper script to manage review sessions.
Evidence: SKILL.md and references/live-review.md describe various hunk session commands. The scripts/apply_comments.py script uses subprocess.run() with a list of arguments to invoke the hunk binary. This implementation avoids shell injection vulnerabilities by passing arguments directly to the executable rather than through a shell interpreter.
[INDIRECT_PROMPT_INJECTION]: The skill processes review comment data from external JSON files, which represents a potential injection surface.
Ingestion points: The scripts/apply_comments.py script reads a user-supplied JSON file specified via the command line.
Boundary markers: The skill instructions emphasize using a validated JSON operation. The helper script acts as a gatekeeper for data entering the Hunk session.
Capability inventory: The skill is limited to reading local files for review purposes and writing comments back to the Hunk session via the CLI. It does not perform network operations or access sensitive system files.
Sanitization: The scripts/apply_comments.py script performs strict schema validation. It parses the input as JSON and verifies that only a whitelist of allowed keys (e.g., filePath, summary, newLine) is present and that values match expected types (strings and positive integers).

hunk-review