Skip to content
nexus-agents

Concepts

Why it works the way it does.

  • What a vote result means

    How nexus-agents turns individual voter decisions into approved, rejected or no_quorum: strategies and their bars, who counts in the denominator, error policies, option votes and the stored record.

    • Explanation
    • For users
  • Agent harness compatibility matrix

    Survey of agent-config standards across the agentic-coding harness ecosystem. AGENTS.md is the canonical surface for this repo; everything else is a thin redirect.

    • Explanation
    • For users
  • Architecture overview

    System design, component relationships, pipeline, adapters, and orchestration

    • Explanation
    • For users
  • Consensus protocols architecture

    Multi-agent voting algorithms, catfish contrarian role, and structured rejection workflow

    • Explanation
    • For users
  • Context load balancing strategy

    Strategy for distributing work across Claude, Codex, and Gemini CLIs

    • Explanation
    • For users
  • Decorated gateway model names & pricing

    How the model registry resolves decorated model names from OpenAI-compatible gateways to canonical pricing, when a manifest alias is still required, and how unpriced calls surface as UNMEASURED in cost output

    • Explanation
    • For users
  • MCP protocol architecture

    MCP 2025-11-25 server implementation for Claude Desktop integration

    • Explanation
    • For users
  • Memory system architecture

    Seven-backend memory system with session, belief, agentic, adaptive, typed, MobiMem, and decay management

    • Explanation
    • For users
  • Routing system architecture

    Multi-stage pipeline for intelligent model selection and task routing

    • Explanation
    • For users
  • Security architecture

    Security pipeline, threat model, sandboxing, trust classification, and input hardening

    • Explanation
    • For users
  • Threat model: Audit hash chain

    Adversarial analysis of the audit-logger hash chain and what verify_audit_chain does and does not detect

    • Explanation
    • For users
  • Untrusted input hardening architecture

    Defense-in-depth architecture for prompt injection and hostile input protection

    • Explanation
    • For users
  • What does a governed decision cost?

    How a governed decision (consensus_vote / pr_review) accrues cost — per-voter rollup to per-decision summary, the measured-vs-unmeasured floor, plan vs api billing, reading the weather_report cost section, and what each strategy's costProfile means. Grounded in the shipped DecisionCostStore (#3855), the weather_report cost section (#3856), and the manifest cost profiles.

    • Explanation
    • For users