SpecLock – Constraint enforcement for AI coding tools (Bolt.new, Claude
Enforces locks on AI behavior (not just memory) via semantic conflict detection; works with multiple platforms.
AI Constraint Engine — enforces CLAUDE.md, .cursorrules, AGENTS.md rules as laws. 51 MCP tools, 991 tests. Official MCP Registry. npx speclock protect
Semantic constraint enforcement for AI—Claude memory without guard rails. Nobody else does this.
Solo developers and teams using Claude Code, Cursor, Bolt.new, Lovable with concerns about AI guardrails and code integrity.
Claude's native memory · Mem0 · Cursor Memory Bank
The problem: Claude Code has auto-memory. Cursor has Memory Bank. But memory without enforcement is dangerous. Your AI remembers you use PostgreSQL — then switches to MongoDB because it "seemed better."
SpecLock adds active constraint enforcement: • You: "Don't ever touch the auth files" • AI: Locked • ...5 sessions later... • You: "Add social login" • AI: CONFLICT: This violates your lock. Should I proceed?
Key features: - Semantic conflict detection (synonym + negation, not just keywords) - Auto-guard related files (injects warnings into source code) - Works with Bolt.new, Lovable, Claude Code, Cursor - 19 MCP tools for memory, tracking, enforcement - Free & open source (MIT)
Demo: https://sgroy10.github.io/speclock/ GitHub: https://github.com/sgroy10/speclock npm: npm install speclock
Enforces locks on AI behavior (not just memory) via semantic conflict detection; works with multiple platforms.
Persistent agent memory for architectural rules via MCP—actually stops agents from redoing weeks of work.
Eight specialist agents catch what Claude Code misses, but it's prompts not actual code analysis.
Mathematically verified policies enforced outside the model—formal proof replaces prompt engineering.
Shadows Clojure's core operators so constraints look like normal code until you call satisfy.
Another Pomodoro timer, but this one forces you to stop working.