soul-md-audit

What problem does it solve? A SOUL.md or system-prompt persona file can read well to a human yet fail in practice: hard rules stranded in the truncated middle, the same instruction restated five ways, or aspirational directives the model cannot enforce. This Skill audits persona files against how the Hermes runtime actually loads them and how models attend to long prompts, producing a scored report with concrete fixes. ## Core Features & Use Cases - 9-point binary checklist: Scores Hermes mechanics (scope, truncation safety, injection-scan cleanliness), voice quality, attention hygiene (front-loading, redundancy, enforceability), and readability. - Runtime-grounded mechanics checks: Verifies behavior against Hermes facts such as the 20,000-character head/tail truncation, HERMES_HOME-only loading, and the prompt-injection scanner that silently blocks tripped files. - Simplify pass: Rewrites bloated files through a constraint-preserving workflow that inventories every hard rule first and verifies none are dropped. - Use Case: A user shares a 25KB operator persona that "stopped working." The audit reveals the identity sits in the truncated middle and a phrase trips the injection scanner, then produces a prioritized fix list and a rewrite target from the template reference. ## Quick Start Audit my SOUL.md persona file and give me a scorecard with specific fixes.