The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to interact with the local repository and orchestration tools. It executes git diff to gather code changes and runs route.mjs from the t3-orchestrator skill to select reviewer models. These are standard operations for the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze potentially untrusted source code and diffs, which are then interpolated into prompts for sub-agent reviewers. This creates a surface where malicious instructions hidden in code comments or data could attempt to influence the agent's behavior.
Ingestion points: The skill identifies changes via git diff or by reading specific files and packages this content for reviewer sub-agents in Step 1.
Boundary markers: The references/reviewer-prompt.md template encapsulates the code content within a specific section, providing a structural boundary.
Capability inventory: Reviewer sub-agents are invoked as "read-only" and are explicitly instructed to run no commands, which limits the potential impact of any successful injection.
Sanitization: The skill relies on structured templates and a rigorous review rubric to maintain control over the sub-agents' focus, but does not implement explicit technical sanitization of the code content.

code-review-panel