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
