The Agent Skills Directory

[SAFE]: The skill implements a legitimate orchestration framework for code analysis and review. Its use of tools such as Bash, Read, and the Agent tool is appropriate for the task of analyzing code diffs and generating reports.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data in the form of code diffs and external requirement documents. It incorporates several design patterns to mitigate associated risks:
Ingestion points: Ingests code diffs via BASE_SHA..HEAD_SHA, project documentation through PRD_PATH, and user intent from INTENT_SNAPSHOT.
Boundary markers: Enforces a structured JSON output format for findings and uses specific agent identities to isolate review perspectives.
Capability inventory: The skill utilizes Bash and Grep for validation, and the Agent tool to delegate analysis to specialized sub-agents.
Sanitization: Implements a "Hallucination check" that uses Bash commands (e.g., [ -f "$file" ]) to verify that reviewer findings reference actual files and symbols present in the repository. It also uses an "Actionability filter" to reject findings that lack verifiable code references.

ql-deep-review