Playbook

1. Name the Capability

Define the exact external action or context the agentAgentA worker you delegate to: brief it, and it takes steps on its own. A chatbot answers; an agent acts.An LLM that runs tools in a loop toward a goal — it acts, checks the result, decides the next step, and repeats until done.Full definitionClose needs. Avoid installing a broad server when two narrow tools would suffice.

2. Check the Surface Matrix

Verify required primitives, transport, authentication, and approval behavior against current official documentation for the exact client: CLI, IDE, web, or cloud.

3. Scope Authority

Use a dedicated identity with least-privilege credentials. Separate read and write operations. Constrain tenant, repository, environment, and network egress at the service layer.

4. Bound the Interface

Use narrow schemas, validate all inputs, cap response size, paginate large datasets, and return stable identifiers plus provenance.

5. Verify Failure

Test unavailable server, expired credentials, denied action, timeout, malformed result, promptPromptThe brief. Closer to writing instructions for a new contractor than to programming.The text you give the model — question, instructions, context. Its quality shapes the quality of the output.Full definitionClose injection in returned content, and duplicate invocation.

6. Record External Effects

Capture who requested the action, which tool and server version ran, normalized inputs, resulting external identifier, and final status.

7. Recheck Before Promotion

Local connection success does not prove cloud compatibility. Qualify the deployment surface with its actual identity, network path, and policy configuration.