Hinata Wiki

Appearance

CLAUDE.md Routing Flow ​

How the CLAUDE.md file controls information flow through the Hinata system. This is the central routing mechanism — every session reads CLAUDE.md and follows its routing tables to load context, execute work, and write outputs.

The Pipeline ​

mermaid
graph TD
    A[Session Start] --> B[Load CLAUDE.md]
    B --> C{Task Type?}
    
    C -->|Runtime/Architecture| D[Must-Read Routes]
    C -->|Domain Work| E[Dynamic Federation Routing]
    C -->|Cross-cutting| F[Captain Routing]
    
    D --> D1[supreme-court/runtime/*]
    D --> D2[supreme-court/format-design/*]
    D --> D3[supreme-court/deployment/*]
    
    E --> E1[Match domain to federation table]
    E1 --> E2[Load federation/{rank}_{pillar}_{agent}_context.md]
    E2 --> E3{≥2 commanders in same pillar?}
    E3 -->|Yes| E4[Load colonel context first]
    E3 -->|No| E5[Execute via commander subagent]
    E4 --> E5
    
    F --> F1[Canary: inbox routing]
    F --> F2[Orochimaru: pipeline health]
    F --> F3[Jimmy Neutron: infrastructure]
    F --> F4[Meruem: evolution assessment]
    
    E5 --> G[Task Execution]
    G --> H{Diataxis Classification}
    
    H -->|Action + Application| I[the-government/how_to_guides/]
    H -->|Action + Acquisition| J[the-government/tutorials_learning/]
    H -->|Cognition + Application| K[the-government/information_reference/]
    H -->|Cognition + Acquisition| L[the-government/understanding_explanation/]

Three Routing Tables in CLAUDE.md ​

1. Must-Read Routes (context loading) ​

Maps task categories to doctrine files. Loaded before execution.

mermaid
graph LR
    CLAUDE[CLAUDE.md Must-Read Table] --> |Runtime delegation| RW[runtime_workflow]
    CLAUDE --> |Federation hierarchy| DR[domain-registry]
    CLAUDE --> |Architecture| HA[hinata-architecture]
    CLAUDE --> |Naming standards| NC[naming-conventions]
    CLAUDE --> |Write paths| AFP[approved-file-paths]
    CLAUDE --> |Network addresses| AIP[approved-ip-addresses]
    CLAUDE --> |Deployment| DP[deploy-policy]
    CLAUDE --> |Cross-harness sync| UHB[universal-hinata-bridge]
    CLAUDE --> |Diataxis framework| DX[diataxis]

2. Dynamic Federation Routing (domain dispatch) ​

Maps task domain to the correct commander context file. This is how work gets delegated.

Task arrives → Identify domain → Match to federation table → Load context → Dispatch subagent

The federation table in CLAUDE.md maps 5 pillars × ~40 commanders. Each context file at federation/{name}_context.md contains: personality, scope, active initiatives, operating instructions, infrastructure references.

3. Must-Write Routes (output classification) ​

Every output is classified per Diataxis and written to the correct the-government/ subfolder.

mermaid
graph TD
    OUT[Output Ready] --> Q1{Informs action or cognition?}
    Q1 -->|Action| Q2{Serves acquisition or application?}
    Q1 -->|Cognition| Q3{Serves acquisition or application?}
    Q2 -->|Acquisition: learning| TUT[tutorials_learning/]
    Q2 -->|Application: doing| HOW[how_to_guides/]
    Q3 -->|Acquisition: understanding| EXP[understanding_explanation/]
    Q3 -->|Application: looking up| REF[information_reference/]

How Replication Works ​

The routing is deterministic because:

  1. CLAUDE.md is the single control surface — every harness (Claude Code, Cloud, Antigravity, Codex) reads it
  2. Tables are explicit — no inference needed, just table lookup
  3. Federation context files are self-contained — each has scope, personality, instructions
  4. Diataxis classification is a decision tree — 2 binary questions, no ambiguity

To replicate for a new domain:

  1. Add a commander context file to federation/
  2. Add the commander to CLAUDE.md's federation table
  3. The routing tables automatically include it

Cross-references ​