Agent skill

Agent skill

Boon ships a curated local skill in the repository:

skill/boon/SKILL.md

The skill teaches agents to propose small retroactive USDC thank-yous, mark uncertainty, dry-run with the CLI, and execute only after explicit approval.

Install from the repository

For local agent workflows, inspect the checked-in skill directly:

cat skill/boon/SKILL.md

The safety-critical rules are:

• never auto-send funds
• use USDC on Base only
• normalize handles before hashing
• use Boon.tip(bytes32,string,uint256,string)
• require evidence-backed notes
• keep recipient claim/help free
• refuse private keys

Hosted skill files

The Starlight docs site serves static agent-skill files:

https://docs.boonprotocol.com/skill.md
https://docs.boonprotocol.com/.well-known/skills/boon.md
https://docs.boonprotocol.com/.well-known/agent-skills/boon.md

These files are copied from skill/boon/SKILL.md. Starlight does not analyze and regenerate them automatically, so refresh the public copies whenever the repository skill changes.

Good agent behavior

A safe Boon agent should output proposals first:

Boons proposal for <period>:
  1. github:alice — 10 USDC
     note: "pr:owner/repo#42 — caught race in bundler"
     why: concrete review prevented a production bug
     evidence: <link>
     status: ready
Total: 10 / 50 USDC review budget.
Next: approve, edit, or skip. No funds move until you approve exact rows.

Then it should run boon tip --dry-run ... before any live boon tip ....