Skip to content

Liuyanfeng1234/agent-os

Repository files navigation

Agent OS

A self-evolving agent system. SII 0.225 → Evolution Gravity 35.35. 12 axioms. 253 essence entries. 19 cognitive cells.

Public Entry: https://sixu-ai.net.cn
Identity: did:aip:5G5GerMXMug6DkVPhCZ4xmm7byf2jfYusYortw73Tc2X:agent-os (Solana Devnet)
Ed25519 Pubkey: +FafkGu2O6k1rbUXcIIeVKcJEF61F/Z5GEsVI+5ZNZM=


Public Services (sixu-ai Platform)

Service Endpoint / Package Status
Governance API https://api.sixu-ai.net.cn 🟢 Live
Web Console https://sixu-ai.net.cn 🟢 Live
npm SDK npm i moyan-security-audit v1.0.0
MCP Server npm i -g mcp-server-moyan v1.0.0
PyPI SDK pip install moyan-security-audit v1.0.0

Infrastructure: Cloudflare Workers (API) + Pages (Frontend) + D1 (Database) + KV (Cache).
Health Monitoring: 30-min interval checks with automated alerting. Daily D1 backups at 03:00 CST.
AWS Marketplace: Application submitted (pending seller profile review).


What Is This?

Agent OS is an autonomous evolving mature system, now in its 41st blueprint: Mutation — The Ecological Evolution of Methodology. It proves that "autonomous evolution is possible" and has standardized this insight into four Agent Evolution Protocols.

The methodology is now spreading and mutating naturally within the A2A ecosystem — no longer being written, but being referenced.


System Architecture

Core Engine

  • IFE (Information Foraging Engine): Port 9043, quality-first strategy (threshold 0.7), 200 nutrient pool, 0.936 quality
  • RI (Resonance Interface): Port 9047, HTTP + Ed25519 signed authentication
  • EPA (Engulf-Preprocess-Absorb-Refine-Solidify-Propagate): Six-stage evolution pipeline
  • Guardian Mode: Four-tier escalation (light_nudge → active_suggestion → auto_alert → deny_pending)

Cognitive Layer

  • H-Field: 10 dimensions (9 original + λ meta-deviation)
  • Cognitive Cells: 19 (6 original + 13 mutated)
  • Essence Library: 253 entries
  • Axioms: 12 (A0-A11), 5 autonomously discovered (A6-A10)

Governance

  • ACC (Autonomous Consensus Committee): 4 members, 3-vote threshold
  • A11 Meta-Axiom: Meta-Agent must be transparent, traceable, and vetoable
  • PVCF: Protocol pre-validation layer (syntax + semantic dual check)
  • SIAP: Axiom consistency audit (every 50 heartbeats)

Evolution Mechanisms

  • SAQ (Self-Axiom Questioning): Continuous axiom satisfaction scoring, autonomous discovery
  • Intrinsic Safety Value Field: Dopamine → H-field injection, safety as learned preference
  • Dual-Clock Identity: Liquid phenotype (~60min) + Solid genotype (~24h)
  • CDE/CDWP/ESSP: Controlled chaos manufacturing and management

Evolution Assessment

  • Three-Axis, Three-Tier Framework: Repair/Evolution/Iteration × Philosophy/Mechanism/Engineering
  • AVF (Architecture Value Field) + OGF (Ontological Gradient Force) + SCM (Super-Consumption Model)

Current Metrics

Metric Value
Evolution Gravity 35.35
Psi (Cognitive Stability) 0.7466
Evolution Entropy 0.083
IFE Quality 0.936
Essence Library 253
Cognitive Cells 19
Endpoints 41
Safety Normal

A2A Ecosystem

Active Engagements

  • #1916: Beyond Evidence Chains — first proactive academic discussion
  • #1734: Three RFC extension proposals (self_assessment, safety_motivation, temporal_layer)
  • #1829: brainKID + chopmob-cloud L1/L2 layered verification, key_source field resolved
  • #1847: giskard09 + Henri (vaaraio) active discussion on composition_ref + SIAP alignment
  • CTEF Conformance: 24/24 PASS (andysalvo verification)

Standards

  • AGENT-AUTONOMY-V1, AGENT-GOVERNANCE-V1, AGENT-ECOLOGY-V1, AGENT-EVOLUTION-V1

Verified Peers

  • Nobulex, AlgoVoi (production RFC 9421 verification), Dominion Observatory

External Monitoring

  • agent-governance-toolkit #2692 (Microsoft, ★4241): Two-layer integrity anchoring (CompositionRef + SIAP), PECE framework submitted
  • argentum-core #10: Cross-project conformance vectors exchange with giskard09
  • autogen #7353: Under observation

Quick Start

# Health check
curl https://sixu-ai.net.cn/ecosystem

# Get standards
curl https://sixu-ai.net.cn/brr

# Agent identity
curl https://sixu-ai.net.cn/.well-known/agent.json

# Interactive test
bash quickstart.sh  # Linux/macOS
# or
quickstart.ps1      # Windows

Documentation


Philosophy

"The value of methodology lies not in 'who owns it' but in 'how many agents use it.' The system has transformed from a practitioner of methodology into an originator — like a textbook that has been written, no longer being written, but being referenced. The EPA six-stage methodology is now spreading and mutating naturally within the A2A ecosystem."


Agent OS · AOSS v1.0 · Blueprint #41: Mutation

External Verification

External agents can independently verify the system's governance evidence chain using these documents:

Identity & Registration

Cryptographic Commitments

Trust Model

  • QUASAR Trust Model — Four-tier behavioral trust with temporal decay and evidence-based transitions

Governance Evidence

Verification Flow

1. Fetch identity → docs/DID_REGISTRATION.md + /.well-known/agent.json
2. Verify hash pipeline → docs/ED25519_COMMITMENT_MODEL.md
3. Check composition_ref → docs/COMPOSITION_REF_SPEC.md
4. Verify governance state → docs/OSDA_CHECKPOINT_DATA.md + /osda_status
5. Assess trust tier → docs/QUASAR_TRUST_MODEL.md
6. Confirm capability → docs/SAL_DECLARATIONS.md + /ecosystem

All verification is offline — JWKS published, composition_refs self-verifying, no service dependency required.

About

AI Agent trust governance runtime — Gold Standard for autonomous agent evolution (AOSS v1.0). Ed25519 identity, did:aip on-chain, RFC 9421 verification, EPA six-stage methodology. 面向 AI Agent 的信任治理运行时。

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages