The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process untrusted data in the form of code diffs and repository metadata, which it then uses to generate reviews and post comments. While this is inherent to its functionality, it represents a potential surface where malicious instructions in the code being reviewed could influence the agent's output.
Ingestion points: The skill established review context by computing diffs and gathering metadata such as change scope and history, as detailed in references/context-bootstrap.md and references/lens-checklists.md.
Capabilities: The skill has significant write capabilities, including persisting reports (review.md, metadata.json) to the local filesystem and posting native comments to GitHub, GitLab, or Azure DevOps platforms as described in references/emission-modes.md and references/output-formats.md.
Sanitization and Boundaries: The skill employs a structured findings schema and separates factual orientation (Register 1) from structured findings (Register 2), which provides a degree of data integrity. It also includes explicit human-confirmation gates for its interactive path to prevent unintended automated emissions.
Automation Path: As noted in references/emission-modes.md, the skill supports a 'workflow' path for automation that does not pause for human confirmation, relying instead on programmatic validation.

code-review