Buzz Agent Skills: Install, Scope, and Verify Them

The Block Agent Skills project defines skills as reusable instructions and resources, usually centered on a SKILL.md file. Installation makes instructions available; it does not prove the workflow succeeds.

A useful skill maps to one agent accountability, names inputs and outputs, limits tools and side effects, and has a test that produces inspectable evidence.

1. 1Read the skill before installing it.
2. 2Match it to a specific job rather than an agent personality.
3. 3Check its scripts, network calls, credentials, and write boundaries.
4. 4Install it through the supported skill mechanism.
5. 5Run a safe fixture or commissioning job.

• The intended agent can actually load the skill.
• The test output matches the skill contract.
• No undeclared files, messages, purchases, or public actions occur.
• A failure is returned as failure, not a polished summary.

• Skill exists on disk but is outside the harness search path.
• Instructions conflict with agent authority.
• The skill expects a credential or binary the runtime cannot access.

Connect the verified skill to a bounded workflow only when recurring execution is needed.

• Buzz Agent workflows
• Buzz Agent harnesses
• What Is Buzz?
• Buzz Agent troubleshooting
• Run Buzz Agents reliably