v1.1.0 — Universal skill enhancement layer

First public release.

What it is: A universal grading layer that sits over any Claude Code skill producing non-measurable output (writing, plans, code, designs, structural docs). Reads the originating skill's SKILL.md on the fly, grades the output A-F against skill intent + user request, drives a revision pass.

Verified across 8 tests, 6 skills, 5 distinct output types:

• internal-comms (anthropics/skills) — strict-format text — lifted B → A
• write-like-ahmed — personal voice — qualitative lift, B held
• frontend-design (anthropics/skills) — HTML/CSS code — 4 cliché AI tells caught
• skill-creator (anthropics/skills) — SKILL.md — first-shot A
• writing-plans (obra/superpowers) — implementation plan — first-shot A
• Calibration control on already-A output — no false inflation
• Variance test (3 judge runs, identical input) — stable B/B/B
• Stress test (deliberately weak baseline) — correctly graded F, 9+ rule violations caught

Commands shipped:

• /second-pass:passing-grade <A-F>
• /second-pass:iteration-cap <n>
• /second-pass:confirm-revisions <true|false>
• /second-pass:strict — preset for high-stakes artifacts
• /second-pass:lenient — preset for quick drafts
• /second-pass:reset

Cost: ~2-3x baseline tokens in default mode, ~4x in strict. Linear with output length.

See README for installation and the full compatibility table.