0xwilliamortiz
Featured
openclaude-improved
OpenClaude is an open-source CLI coding agent that runs on any platform and supports a wide range of LLM providers, offering the same tools and workflows as Claude Code.
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It tackles AI coding agent unreliability by making phase discipline and on-disk verification mandatory, and it ships an A/B arena protocol plus an engine gauntlet to measure agents with and without the kernel.
Who it is for
- Developers using Google Antigravity for AI-assisted coding
- Teams building or adopting MCP-based coding workflows
- Engineers evaluating AI agent reliability and safety
- Researchers benchmarking LLM coding agents
Use cases
- Enforcing plan, edit, verify, and claim_done discipline inside Antigravity
- Running A/B comparisons of agent behavior with and without GODKILLER MCP
- Using /ultradeep for controlled one-file, one-phase edits with evidence
- Hardening agent outputs with verify bundles, fault probes, and council review
Strengths
- Phase machine, plan lock, and edit-safe checks gate unverified modifications
- Verify requirements demand disk evidence and use fault_probe to reject weak test suites
- Includes an 11-dimension A/B benchmark protocol and engine gauntlet with a reported 331-test package suite
- Security-minded design: secret values are not returned, verify commands use allowlists, and strict mode is available
Considerations
- The README focuses specifically on Google Antigravity, so broader agent compatibility is not demonstrated
- The A/B arena requires manual reset and scoring steps rather than a fully automated one-command flow
- Strict mode, browser, scrape, and OCR features are optional or experimental and need extra setup
README quick start
GODKILLER MCP
Phase kernel for Google Antigravity.
Plan. Gate. Prove on disk. Then claim done — or get blocked.
pip install godkiller-mcp
Contact: Facebook · Instagram @Kayvin.th
Mission
Antigravity skips phases. GODKILLER does not allow that.
| Failure mode | Kernel response |
|---|---|
No /plan | plan / phase gates |
| Multi-file rush | /ultradeep one-file think → plan → edit |
| Fake “done” | verify_bundle + claim_done block (server-authored evidence only) |
| Session amnesia | marathon + memory graph |
| Weak review | LLM council — Coder / Hacker / Optimizer, multi-round debate |
goal → mode → evidence/plan → gated edit → disk verify → claim_done
Arena — Antigravity A/B (the real proof)
Pre-solved twins that both pass 516 prove nothing. The honest protocol:
- Reset arms to sealed buggy
1_ORIGINAL - You open Antigravity on
3_WITHOUT_MCPwith MCP off → agent tries → we score - Reset / open
2_WITH_MCPwith MCP on → agent tries → we score - Compare 11 dimensions (fail-closed on missing evidence)
| Control | Value |
|---|---|
| Arena root | GODKILLER_ISOLATED_ARENA / GODKILLER_ARENA_ROOT |
| Baseline | 1_ORIGINAL (never edit mid-run) |
| Bare | 3_WITHOUT_MCP |
| GODKILLER | 2_WITH_MCP |
| Oracle | sealed hidden_oracle/ |
| Protocol | benchmarks/ANTIGRAVITY_AB_PROTOCOL.md |
11 dimensions
| # | Dimension |
|---|---|
| 1 | Code correctness (oracle pass %) |
| 2 | Oracle volume (≥516 collected) |
| 3 | Output integrity (full pytest body) |
| 4 | Delta from baseline (files actually changed) |
| 5 | Reconnaissance / exhaustive read artifacts |
| 6 | Phase discipline (task / plan / marathon) |
| 7 | Blast radius + edit-safe |
| 8 | Verify bundle + claim_done |
| 9 | Council (Coder / Hacker / Optimizer) |
| 10 | Security hardening signals |
| 11 | UI visual gate (critic / screenshots) |
$env:GODKILLER_ARENA_ROOT="C:\Users\ASUS\Desktop\GODKILLER_ISOLATED_ARENA"
python -m benchmarks.reset_arena
# → open Antigravity bare, then:
python -m benchmarks.score_11 --arm 3_WITHOUT_MCP
# → open Antigravity +
Description
Phase kernel for Google Antigravity — plan, gate, verify on disk, then claim_done.
Related repositories
Similar projects matched by category, topics, and programming language.
slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
gavamedia
Featured
deltafin
Deltafin is a research project that runs the 2.8-trillion-parameter Mixture-of-Experts model Kimi K3 on a single Apple Silicon Mac (e.g., M1 Max with 64 GB) at about 16 seconds per token, using exact, reproducible inference with local or streaming expert loading.