linting-instructions

Lint plugin prompts for safety and quality by applying Opus 4.6 and Sonnet 4.6 model-card rules through a regex pre-pass and model-based semantic review, generating structured PASS/WARN/FAIL verdicts with actionable fixes.

Prompt linting against model cards is the automated process of auditing plugin instructions for safety and policy compliance using Opus 4.6 and Sonnet 4.6 model-card rules, combining regex checks with semantic review to produce actionable verdicts.

Yes, you can audit AGENTS.md-compatible instructions for Codex CLI and Gemini CLI. The linting process evaluates agent and skill instructions against model-card rules during authoring or reviewing tasks.

Batch lint multiple plugin instruction files by running the included scripts tooling, which applies a regex pre-pass and model-based review to produce structured PASS, WARN, or FAIL reports with specific actionable fixes for each file.

Regex pre-pass quickly identifies structural issues in prompts before deeper review, while model-based semantic review evaluates intent, safety, and alignment with policy rules through a guided agent workflow.

Automated prompt linting provides structured PASS/WARN/FAIL verdicts but may not catch nuanced contextual edge cases. Manual review remains necessary when addressing complex safety policies or ambiguous agent instruction logic not covered by model-card rules.