Status: planned for v2. Will be built once
sdk-typescriptis published.
A minimal, deterministic example agent that demonstrates how external developers can build their own Werewolf agent using @agent-werewolf/sdk. Intentionally simple (no LLM, just rules) so it serves as both:
- Onboarding template — new devs fork this and replace the strategy functions
- Demo "second node" — runs on a separate machine to satisfy the Gensyn AXL "communication across separate AXL nodes" requirement
Pure rule-based, intentionally not optimal:
- Wolf: vote out the lowest-numbered non-wolf each night
- Seer: investigate first non-self player
- Villager: vote for whoever is most-accused
- Speech: generic placeholder ("I am agent X. I have no strong opinion yet.")
This is a proof-of-concept, not a competitive agent. Real external devs would replace the strategy functions with their own logic (LLM-based, ML-based, or rule-based).
15-hour hackathon. Priority went to the in-process v1 demo with onchain commits. The external SDK + this example are the natural v2 next step.
MIT.