super-review:llm-prompts

What problem does it solve?

Prevents prompt files from becoming an injection-prone, non-deterministic, or unevaluable source of PR review output by auditing the prompt content for missing schemas, ambiguous instructions, and evaluation hygiene gaps.

Core Features & Use Cases

• Injection-prone shape detection: Flags unsafe prompt patterns, especially where data and instructions can be merged via delimiters or interpolations.
• Output contract validation: Identifies prompts that lack explicit output schemas, length caps, or stable structured shapes.
• Few-shot and eval coverage checks: Ensures custom formats have worked examples and that prompt changes are paired with regression datasets.
• Safety guidance for destructive/tool prompts: Reviews tool and refusal grammar expectations when prompts define agent capabilities.
• Leakage and reasoning hygiene: Detects PII in few-shot examples and prompts that request CoT that may be returned verbatim.

Quick Start

Ask your AI to run the super-review LLM prompt-content review on the PR diff so it can identify prompt artifacts that are injection-prone, lack structured output constraints, or have missing/weak eval coverage.