healer

Healer v9.0

Universal Autonomous Codebase Health & Development Engine with Design Intelligence

44 commands | 27 flow presets | 36+ recipes | Karpathy enforcement | Postfix ? discoverability | Self-validating command catalog | Shared enforcement protocol | Integrated UI/UX design databases | Any project, any language, any platform

---

What is Healer?

Healer is a comprehensive command suite for Claude Code that turns your AI assistant into a research-augmented development lifecycle engine. Every command searches for best practices, patterns, and proven solutions before acting.

What's New in v9.0: Karpathy Principles Integration

v9 integrates Andrej Karpathy's coding principles directly into Healer's enforcement layer — upgrading them from passive CLAUDE.md suggestions to actively enforced HARD-GATEs:

• /healer:karpathy — focused code review against 4 Karpathy principles with dual per-principle + per-file report format
• Simplicity Protocol (P2) — HARD-GATE preventing over-engineering, speculative features, and single-use abstractions. Self-scoped: only activates when commands write source code.
• Surgical Changes Protocol (P3) — HARD-GATE enforcing minimal diffs with traceback test: "Every changed line traces to the user's request." Self-scoped.
• P1 enhancement — Research Protocol now requires surfacing tradeoffs and pushing back when simpler approaches exist
• P4 coverage — already enforced by Healer's Verification and Fix Verification Protocols since v4
• 5 anti-rationalization entries + 3 red-flag stop conditions targeting over-engineering patterns
• Flow presets: karpathy-review, karpathy-fix
• See docs/designs/2026-04-15-karpathy-integration.md for design rationale and CHANGELOG.md for full details.

What's New in v7.1: Postfix ? Help + Self-Validating Catalog

• ? postfix discoverability — type /healer:<command> ? (or --help) on ANY healer command for an instant six-section drill-down (purpose, what it does, expected input, concrete example with trace, input purpose, after/related). Works on flow presets too: /healer:flow feature ?.
• Sub-50ms help latency — pre-built data/help-index.json replaces the old slow path that scanned all 41 command files (~14k lines) per invocation.
• Self-validating command catalog — data/commands.yaml and data/flows.yaml (JSON-Schema-validated) are the single source of truth. A PostToolUse hook auto-rebuilds the index on every catalog/command edit and detects drift loudly.
• One-source documentation — README, HTML user guide, and help system all derive from the same YAML catalog (auto-generated marker sections, in progress).
• See docs/designs/2026-04-11-postfix-help-and-catalog.md for design rationale and docs/plans/2026-04-11-postfix-help-and-catalog.md for the implementation plan.

What's New in v6: Design Intelligence Integration

v6 integrates UI-UX-Pro-Max (MIT License) design intelligence into Healer, adding:

• 6 new commands: brand, logo, cip, banner, icon, slides
• 5 enhanced commands: design, design-system, design-review, implement, audit now query local design databases before web research
• Design data layer: 161 color palettes, 57 font pairings, 99 UX guidelines, 50+ UI styles, 14 stack-specific guideline files
• BM25 search engine: Python-based search across all design databases
• Auto-sync: SessionStart hook keeps design data fresh from upstream
• New flow presets: identity, visual, brand-to-prod + 4 new recipes

Design intelligence is the DATA LAYER. Healer's enforcement protocol is the PROCESS LAYER. Together they produce research-backed design that avoids AI slop.

v4 Foundation: Enforcement Protocol

v4 introduces a shared enforcement layer (_enforcement.md) that makes all commands actually effective:

• HARD-GATEs — absolute blockers that prevent skipping research, verification, or testing
• Explicit tool calls — every "search online" instruction now specifies exact tools (WebSearch, WebFetch, Context7 MCP)
• Evidence-before-claims — every "tests pass" claim requires actual command output, not assumptions
• Fix-verify cycles — apply one fix, run tests, verify, then move on (no batching)
• Anti-rationalization tables — blocks common excuses like "I already know how to fix this"
• Red-flag stop conditions — automatic halt when approaches aren't working (3+ failed fixes = stop and reassess)

This enforcement layer is what separates Healer v4 from tools that just describe best practices — Healer v4 enforces them.