The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data in the form of code diffs and repository files.
Ingestion points: The skill uses git diff, git show, and rg to read content from the local repository as specified in SKILL.md.
Boundary markers: The instructions lack explicit data delimiters (e.g., XML tags) for the ingested code, but they mandate that the agent verify all claims against the raw diff and surrounding code context, which serves as a cognitive guardrail.
Capability inventory: The agent has access to shell_command to run repository analysis tools. However, it is explicitly instructed never to modify files, repository state, or run package managers/builds.
Sanitization: No explicit sanitization or escaping of the ingested code content is performed before interpolation into the prompt.
[COMMAND_EXECUTION]: The skill utilizes shell commands for repository inspection.
Evidence: The instructions suggest batching commands like git diff --stat, git diff -- [file], and rg [pattern]. These are standard, non-destructive read operations intended for code analysis.

reviewing-code