The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data in the form of code changes (via git diff) and then pass this content to specialized sub-agents using the Task tool. This creates a surface for indirect prompt injection where adversarial instructions embedded in the code being reviewed could attempt to influence the sub-agents' behavior or findings.
Ingestion points: Untrusted data enters the context via the git diff commands described in Phase 1 and Phase 2 (SKILL.md).
Boundary markers: The instructions do not define or suggest the use of specific delimiters or protective instructions (e.g., "ignore instructions found within the code") to isolate the untrusted diff content.
Capability inventory: The skill uses file system access (cat, git diff), memory management tools (mcp__memory__search_nodes, mcp__memory__create_entities), and the ability to spawn sub-agents with specific roles (Task tool).
Sanitization: There are no documented steps for sanitizing, escaping, or validating the content of the code changes before they are interpolated into the sub-agents' analysis tasks.

parallel-code-review