"KLOC is not here to help ideas succeed. It is here to help bad ideas die early."
KLOC (Kernel-Level Oversight Council) is an operational epistemic firewall and forced disagreement engine. It is designed to stress-test technical and strategic proposals using a simulated council of five independent adversarial AI agents.
Most AI assistants are biased toward helpfulness and compliance. KLOC is biased toward rejection and skepticism.
Each seat operates in isolation to prevent consensus drift:
- 🛡 SECURITY: Identifies exploit surfaces, privilege escalation, and data poisoning.
- ⚙️ SYSTEMS INTEGRITY: Detects non-determinism, hidden state, and observability gaps.
- 💰 ECONOMIC RISK: Identifies cost blowouts, runaway spend, and tail risks.
- 👺 FOUNDER-ADVERSARY: Assumes the proposer is wrong, biased, or euphoric.
- 🌐 EXTERNAL REALITY: Identifies legal unknowns, dependency fragility, and vendor lock-in.
- No Approval Votes: The only valid outputs are
NO_OBJECTION,SOFT_REJECT,HARD_REJECT, orESCALATE. - Deterministic Resolution:
- Any
HARD_REJECT= BLOCK - Any
ESCALATE= MANUAL REVIEW - 2+
SOFT_REJECT= DEFER
- Any
- Hypothesis Verification: Stated "Success Criteria" are treated as attack surfaces, not facts.
- Unknown Enforcement: If information is missing, the engine is programmed to increase friction, not bypass it.
- A Google Gemini API Key.
- A modern browser.
- Clone the repository:
git clone https://github.com/your-org/kloc.git
- Set your environment variable:
export API_KEY='your-gemini-api-key'
- Serve the project root (e.g., using
npx serveor simply openingindex.htmlin an environment that supports ES modules).
- Core: React 19 (via ESM)
- Styling: Tailwind CSS
- Intelligence: @google/genai (Gemini 2.5 Flash)
- Icons: Lucide React
MIT License. See LICENSE for details.
Created for the Shards Inc Ecosystem. Internal use only. No execution authority.