The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes untrusted data from external sources including pull request descriptions, linked issues, and comments which could contain instructions intended to influence the agent's review. \n
Ingestion points: Data enters the context via the pull request diff (Step 1), the pull request description, linked issues, and prior review comments (Step 3). \n
Boundary markers: The skill mitigates accidental obedience by instructing the agent to form an independent assessment (Step 2) before reading the author's narrative. \n
Capability inventory: The skill possesses the capability to read repository files and emit pull request comments (Step 7). \n
Sanitization: No specific sanitization or filtering of the external content is defined in the instructions. \n- Automated Communication Pathway: The skill provides a mechanism for other agentic tools to authorize the posting of pull request comments without a separate user confirmation. While this is a functional feature for automated workflows, it creates a pathway for comment emission based on instructions from other agents in the toolchain.

code-review