Cross-Platform Agent Standard · R25 P0 · Open Contract
The Kojie Standard
A falsifiable contract for AI agents that operate across regulated platforms. Five patterns. One mathematical baseline. Three connected platforms. Adopt freely.
Why this exists
The dominant pattern in commercial AI agents today is "trust the model because it was trained on more data." That claim is unfalsifiable — no regulator can reproduce it, no auditor can verify it, and no engineer can debug it.
CPAS replaces that claim with a mechanical contract. An agent either grounds its decisions in a declared mathematical baseline, signs every decision cryptographically, restricts itself to a known tool inventory, declares the regulations it judges against, and emits machine-submission output — or it does not. There is no probabilistic gray area, and the check runs at every boot, in CI, and on demand.
The Five Patterns
Pattern 1
N-Agent Consensus
No single agent decides alone. A quorum of at least two distinct agents — drawn from at least two different perspectives — must concur on every decision.
Pattern 2
Cryptographically Signed Audit Trail
Every decision is HMAC-SHA256 signed at issuance. Provenance survives export to any downstream system — NRC docket, NIST OSCAL, CMMC evidence package — without losing tamper-evidence.
Pattern 3
Deterministic Tool Inventory
Agents may only invoke named tools from a pre-registered inventory. No free-form computation. No hallucinated tool outputs. Every calculation is reproducible.
Pattern 4
Explicit Regulatory Coverage
Every decision declares which regulations and framework controls it is rendering judgment on. A decision with empty regulatory coverage is rejected at the contract layer.
Pattern 5
Machine-to-Machine Output Formalization
Output payloads conform to a declared schema (NRC submission JSON, NIST OSCAL, MARSSIM survey JSON, CMMC evidence JSON, SOC 2 attestation JSON, FedRAMP SSP JSON — all crypto-signed). Free-form prose alone is not a Kojie-Standard decision.
Three Connected Platforms
Hub
gpcybersecurity.com
Guardian Posse Cybersecurity — 12 cyber-defense agents. Hosts the manifest, the contract enforcement, and the certification harness. NIST 800-53/171/172, CMMC L2/L3, RMF, FedRAMP, SOC 2.
Specialty: cybersecurity
Standard
kojie.works
Nuclear Decommissioning Platform — 4-agent consensus. The live embodiment of all 5 patterns above; this is why the standard bears its name. NRC, MARSSIM, OSHA, DOT, NESHAPs.
Specialty: nuclear D&D
Baseline
sovereign144.com
The Truth Sequence — mathematical baseline. The Master Equation
Ω = N · (Φ_W · r · Ψ_B · δ_A) = 1.0 and 49
deterministic proofs (47 Truth Sequence + 2 G.U.M.E.) every
agent is certified against.
Specialty: mathematics verification
Adopt the standard
CPAS is open. There is no license fee, no proprietary SDK, and no enrollment process. If you build AI agents for regulated industries and want a public audit story, you can implement the same five patterns and publish your own manifest endpoint at the same path.
Reference implementation lives in this repository in three modules
— cross_platform_agent_manifest.py,
kojie_standard_contract.py, and
sovereign_144_baseline.py — and a per-agent live check
is available at /cpas/certify/<agent_id>.