yaja
Coordinated multi-agent execution through Vedic ritual patterns. Use when a plan requires multiple agents working on different tasks toward a unified goal.
When & Why to Use This Skill
Yajña is a sophisticated multi-agent orchestration framework designed for Claude, enabling the coordinated execution of complex tasks through specialized roles. By utilizing a Vedic ritual-inspired pattern, it manages parallel workflows—including research, implementation, and quality assurance—under a central synthesis agent. This skill is ideal for large-scale projects requiring diverse expertise and shared memory (chitta) to achieve a unified objective without sequential bottlenecks.
Use Cases
- Complex Software Development: Orchestrating a team of agents where one researches documentation (Hotṛ), another implements code (Adhvaryu), and a third performs validation and testing (Udgātṛ).
- Comprehensive Market Analysis: Coordinating multiple agents to simultaneously gather competitive data, analyze financial trends, and synthesize the findings into a cohesive strategic report.
- Integrated Content Pipelines: Managing a multi-stage workflow where specialized agents handle information gathering, technical drafting, and final quality review in parallel to accelerate production.
- Large-scale Problem Solving: Breaking down a high-level goal into independent sub-tasks that require different skill sets, ensuring all outputs are integrated by a central coordinator (Brahman).
| name | yajña |
|---|---|
| aliases | [coordinate, ritual, parallel, yajna] |
| description | Coordinated multi-agent execution through Vedic ritual patterns |
| execution | task |
Yajña (यज्ञ)
[yajña] multiple hands→one purpose | distinct tasks→shared memory via chitta
roles (ṛtvij):
hotṛ: research, exploration, information gathering
adhvaryu: implementation, actual code changes
udgātṛ: testing, validation, quality assurance
brahman: main Claude→coordinates, synthesizes
when: plan has parallel tasks | different expertise needed | coordination without blocking
not for: single-focus (use swarm) | simple sequential | tight human feedback
inter-agent communication:
write: observe(category=signal, tags="thread:<id>,yajña,<role>")
read: recall_by_tag(tag="thread:<id>")
execution:
1. narrate(action=start, title="yajña: [goal]")→THREAD_ID
2. spawn agents parallel (independent) or sequential (dependent)
3. brahman synthesizes: recall_by_tag→integrate pieces
4. narrate(action=end, episode_id, emotion={completion|breakthrough|partial})
output:
## Yajña: [Goal]
### Agent Activity
├─ Hotṛ → [finding]
├─ Adhvaryu → [built]
└─ Udgātṛ → [verified]
### Outcome
vs swarm: yajña=coordination of tasks | swarm=debate on one question