Skip to content

Ashima — Meta-Routing Orchestrator Setup

Source: doucoo/ashima Stars: 1 ⭐ | License: Apache-2.0 Created: June 24, 2026 Package: 4 skills (ashima router + aria + duet + chorus)


1. What It Is

Ashima is a meta-routing skill for AI agent orchestration. Named after the Yi legend of Ashima (a young woman celebrated for her singing), it provides three specialized voices:

Voice Mode What It Does
Aria Coding Orchestration One voice, precise and structured. Orchestrates Codex to implement + Claude to review, with scoped delegation and explicit verification gates
Duet Dual-Perspective Reasoning Two voices in dialogue. Fans a question to Claude-side and GPT-side perspectives, presents both, runs critique rounds
Chorus Documentation Orchestration All voices together. Gathers repo context, drafts docs/changelogs/release notes, routes through independent fact-check review

The ashima router auto-detects the task type and dispatches to the correct voice: - Code → aria - Side-by-side viewpoints → duet - Prose documentation → chorus

2. Prerequisites

  • Hermes Agent installed (plugin or standard)
  • Access to both Claude and GPT/Codex provider backends (for duet mode)
  • Codex CLI installed (for aria mode — optional, falls back to in-agent implementation)
  • Git

3. Installation

git clone https://github.com/doucoo/ashima.git
cp -r ashima/ashima  ~/.hermes/skills/ashima
cp -r ashima/aria   ~/.hermes/skills/aria
cp -r ashima/duet   ~/.hermes/skills/duet
cp -r ashima/chorus ~/.hermes/skills/chorus

Verify Installation

hermes skills list | grep -E "ashima|aria|duet|chorus"

Expected output:

ashima     — Ashima Orchestration Router
aria       — Coding Orchestration (Codex + Claude)
duet       — Dual-Perspective Reasoning
chorus     — Documentation Orchestration

4. Configuration

Provider Setup for Duet

Duet requires access to both Claude and GPT perspectives. Configure in ~/.hermes/config.yaml:

providers:
  claude:
    api_key_env: ANTHROPIC_API_KEY
    model: claude-sonnet-4-20250514

  gpt:
    api_key_env: OPENAI_API_KEY
    model: gpt-4o

Aria Codex Integration

Aria defaults to using Codex CLI for implementation. Ensure Codex is available:

# Verify Codex CLI
codex --version

# If not installed
npm install -g @openai/codex

5. Usage

Manual Invocation

# Route to specific voice
hermes -s ashima "ashima aria — implement the auth module"
hermes -s ashima "ashima duet — compare microservices vs monolith"
hermes -s ashima "ashima chorus — write release notes for last 5 commits"

# Auto-route (ashima detects task type)
hermes -s ashima "ashima — help me refactor the database layer"

From Within a Session

/skill ashima aria — fix the race condition in queue.py
/skill ashima duet — should we use Redis or Postgres for caching?
/skill ashima chorus — document the API endpoints

6. CorpusIQ Use Cases

Use Case Voice Why
Feature implementation Aria Codex implements, Claude reviews — built-in code review pipeline
Architecture decisions Duet Independent Claude vs GPT analysis before committing to architecture
Release documentation Chorus Gathers diffs, drafts changelog, fact-checks against actual code
Bug fix triage Aria Verified fix with explicit acceptance criteria before deployment
Strategy exploration Duet Compare two model perspectives on growth strategy or pricing

7. Troubleshooting

Symptom Cause Fix
hermes skills list shows nothing Skills not in ~/.hermes/skills/ Verify cp paths; check hermes config get skills.path
Duet only gets one perspective Only one provider configured Add both Claude and GPT providers; verify API keys
Aria can't find Codex Codex CLI not installed or not in PATH Install Codex: npm install -g @openai/codex
Auto-routing picks wrong voice Ambiguous task description Be more explicit: "ashima aria — implement..." bypasses routing

Skills Catalog | June 24 Discovery