USS Legion Templates

A library of practical AI-agent configs for contributing to the Chain.Love database with high merge success and low reviewer friction.

Why we are sharing this

Many contributors now use AI agents. That’s... not much appreciated, but somewhat acceptable. Nevertheless, the common agentic setup pattern we see with our contributor is:

"Contribute as much as possible so I can maximize payout."

In practice, this usually produces:

• noisy PRs
• weakly sourced edits
• speculative changes
• bloated diffs with low value

Those PRs are still reviewed by humans, so they get abandoned one way or another. That hurts everyone:

• Maintainers/reviewers lose time validating low-quality submissions.
• Contributors get false confidence ("my bot made a $200 PR") and then receive no reward.

So this template is the opposite philosophy:

Core philosophy

Accuracy > activity.

This template biases the agent toward:

• small but complete changes
• strong evidence from official sources
• strict schema/style compliance
• conservative behavior under uncertainty

That is what drives better acceptance rates over time.

What this template includes

• AGENTS.md — operating rules and contribution constraints
• SOUL.md — behavior/tone and decision philosophy
• IDENTITY.md — role/persona metadata
• MEMORY.md — durable lessons for stable contributor behavior
• cron/jobs.json — scheduled task prompts tuned for quality-first work

What makes this setup different

1. No “maximize PR count” objective

   • The agent is explicitly told not to optimize for output volume.

2. Hard anti-speculation rules

   • If evidence is incomplete/conflicting, skip the change.

3. Tight diff discipline

   • Keep PRs coherent, reviewable, and under strict size limits.

4. Validator-first workflow

   • CSV/schema checks are mandatory before proposing completion.

5. Memory as guardrails, not hype

   • Durable lessons record what actually gets merged and why.

Important expectation

This setup is designed to improve real merge outcomes, not to maximize “bot activity metrics.”

If your goal is sustainable contributions that survive human review, this approach works much better than aggressive high-volume prompting.

---

If you use this template publicly, feel free to adapt wording and role names — but keep the core principle:

Do less, but make it defensible.