The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted data from the local repository (such as git diff output and document contents) and transmit it to a sub-agent for analysis. This creates a vulnerability where malicious instructions embedded in code comments or document text could influence the behavior of the sub-agent.
Ingestion points: SKILL.md describes identifying diff scopes and file paths to be read and passed to the sub-agent.
Boundary markers: While the skill structures requests using Objective and Scope sections, it does not implement strict sanitization for the content of the diffs themselves.
Capability inventory: The skill has the ability to read files and communicate with external model providers (Claude, OpenAI, etc.).
Sanitization: No explicit sanitization or filtering of the content being reviewed is described in the provided scripts.
[COMMAND_EXECUTION]: The skill relies on executing local bash scripts (run.sh, prepare.sh, dispatch.sh, etc.) and standard version control tools (git diff, git log, git show). These commands are used to gather the context necessary for the review delegation task and are constrained to the skill's subdirectory and standard git operations.

delegate-review