The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external code and review packets, which creates a potential surface for indirect injection.
Ingestion points: Repository files under review and response packets from sub-agent critics, as defined in references/review-topology.md.
Boundary markers: The workflow uses 'blind projections' and 'packet aliases' to hide real repository paths and context from sub-agents.
Capability inventory: The process involves reading file contents, aggregating peer evidence, and recording substantive findings in a central ledger.
Sanitization: Instructions explicitly mandate the removal of identity-bearing metadata, diff headers, and session lineage before data is shared with reviewers.
[SAFE]: The skill focuses on security best practices, including data isolation, audit trails, and strict constraints on sub-agent behavior to prevent unauthorized actions like code implementation or public posting.

subagent-review