The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes git diffs which are externally controlled data sources and could contain malicious instructions designed to influence the agent's review.
Ingestion points: Git repository history and working tree via --rev and --cwd arguments as described in SKILL.md.
Boundary markers: The tool employs byte-bounding ("clipped patch") and structured schemas ("capped by schema") to limit the data ingested into the model context.
Capability inventory: The skill is configured with the exec tool to run the diff processing programs.
Sanitization: Output is truncated to prevent context flooding, but the content itself is parsed from the repository without explicit sanitization mentioned.
[COMMAND_EXECUTION]: The skill uses the exec tool to invoke diff-slice and diff-review with arguments including directory paths and git revisions, which are executed in the local environment.

algal-diff