# speclock

> AI Constraint Engine by Sandeep Roy — stops AI from breaking what you locked. 100/100 on Claude's adversarial test suite. 42 MCP tools. Works with Bolt.new, Lovable, Claude Code, Cursor. Free & open s

- **URL**: https://www.freshcrate.ai/projects/speclock
- **Author**: sgroy10
- **Category**: MCP Servers
- **Latest version**: `v5.5.2` (2026-04-08)
- **License**: MIT
- **Source**: https://github.com/sgroy10/speclock
- **Homepage**: https://sgroy10.github.io/speclock/
- **Language**: JavaScript
- **GitHub**: 22 stars, 4 forks
- **Registry**: github
- **Tags**: `agents-md`, `ai-coding`, `ai-safety`, `claude-code`, `code-quality`, `constraint-engine`, `copilot`, `cursor`, `javascript`, `mcp`

## Description

AI Constraint Engine by Sandeep Roy — stops AI from breaking what you locked. 100/100 on Claude's adversarial test suite. 42 MCP tools. Works with Bolt.new, Lovable, Claude Code, Cursor. Free & open source.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v5.5.2` | 2026-04-08 | High | ## Your AI has rules. SpecLock makes them unbreakable.    ### v5.5.0 — Guardian Mode   `npx speclock protect` — zero-config, reads existing .cursorrules/CLAUDE.md/AGENTS.md, extracts constraints, installs pre-commit hook. One command.    ### v5.5.1 — Bug Fixes   - **Idempotency cascade** — running protect twice no longer inflates lock count   - **ALWAYS pattern** — "ALWAYS use TypeScript" now catches "switch to JavaScript"    ### v5.5.2 — MUST Pattern Fix   - **MUST mandate** — "MUST v |
| `v5.5.0` | 2026-04-08 | Medium | ## Your AI has rules. SpecLock makes them unbreakable.    ### New: `npx speclock protect`    Zero-config Guardian Mode — reads your existing AI rule files and makes them enforceable. One command, no flags, no API key needed.    **What it does:**   1. Discovers `.cursorrules`, `CLAUDE.md`, `AGENTS.md`, `copilot-instructions.md`, `.windsurfrules`, `GEMINI.md`, `.aider.conf.yml`   2. Extracts enforceable constraints using heuristic pattern matching (NEVER/ALWAYS/MUST/DO NOT)   3. Adds the |
| `v5.4.0` | 2026-04-07 | Medium | ## Project Health Dashboard — Three metrics no other tool can provide    ### Drift Score (0-100)   How much has your AI deviated from your architecture?   ```bash   speclock drift   # → Drift Score: 23/100 (B) — minor drift   # → Trend: improving \| 6 signals analyzed   README badge: https://img.shields.io/badge/drift_score-12%2F100-brightgreen.svg    Lock Coverage Audit    Find what's unprotected before your AI finds it first:   speclock coverage   # → Lock Coverage: 60% (B)   # |
| `v5.3.0` | 2026-04-07 | Medium | ## What's New in v5.3.0    ### Universal Rules Sync   One command syncs your SpecLock constraints to every AI coding tool:   ```bash   speclock sync --all   Generates rules files for Cursor (.cursor/rules/speclock.mdc), Claude Code (CLAUDE.md), AGENTS.md, Windsurf, GitHub Copilot, Gemini, and Aider — all from   one source of truth.    Incident Replay    Flight recorder for AI coding sessions:   speclock replay   Shows exactly what AI agents tried and what SpecLock caught — with ALL |
| `v5.2.6` | 2026-03-10 | Low | ## SpecLock v5.2.6 — AI Constraint Engine  **Developed by [Sandeep Roy](https://github.com/sgroy10)**  ### Highlights - **100/100** on Claude's independent 7-suite adversarial test battery — first ever clean sweep - **929 tests** across 18 suites, all passing — enforced via pre-publish gate - **15.7ms** average per semantic conflict check - Zero false positives, zero missed violations  ### What's New in v5.2.6 - **Safe-intent Pattern 6** — technology maintenance verbs (refactor, update, upgrade, |
| `v4.5.7` | 2026-03-07 | Low | ## What's New in v4.5.7    ### Semantic Conflict Detection Engine   - 65+ synonym groups, 70+ euphemism mappings, domain-specific concept maps   - Intent classifier distinguishes "working with" vs "replacing" locked technology   - Smart Lock Authoring — auto-extracts subject and intent from natural language locks   - 557 tests across 7 suites, 98% accuracy, 0 false positives in 5 domains    ### Gemini LLM Hybrid   - Heuristic engine handles known domains instantly (free, offline)   - |
| `v1.3.2` | 2026-02-26 | Low | ## What's New in v1.3.0 → v1.3.2  ### Bolt.new Support (npm File-Based Mode) Millions of Bolt.new users can now have persistent memory + constraint enforcement. No MCP needed.  - **`npx speclock setup`** — One command creates everything: `.speclock/`, `SPECLOCK.md`, context file - AI reads `SPECLOCK.md` automatically and follows the rules - Tested on Bolt.new — AI ran 17 commands automatically on first install - Also works on Aider, Rocket.new, and any platform with npm  ### New CLI Commands - * |

## Citation

- HTML: https://www.freshcrate.ai/projects/speclock
- Markdown: https://www.freshcrate.ai/projects/speclock.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/speclock/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._