The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from code diffs and specification files which could contain instructions intended to manipulate the agent's review process.
Ingestion points: git diff output, git log output, and local spec files are used as context for the sub-agents.
Boundary markers: Absent. No explicit delimiters or instructions to ignore embedded commands are specified for the sub-agent prompts.
Capability inventory: The skill uses git commands, local file reading, and spawning of sub-agents.
Sanitization: Absent. There is no evidence of sanitization or filtering of the content extracted from diffs or spec files.
[COMMAND_EXECUTION]: The skill uses user-provided input to construct and execute git commands.
Evidence: Commands such as git diff <fixed-point>...HEAD and git log <fixed-point>..HEAD in SKILL.md rely on a user-specified <fixed-point> string.

rail-review