01 / DECIDE
Architecture decision
Use it whenever a model can move from recommending an action to executing it in an external system.
Classify tools by impact and pause before external messages, purchases, destructive writes, credential use, or privilege changes. Map input, output, state, and side effects as one observable path before deciding which layer owns insert approval at the side-effect boundary.
02 / BUILD
Three-step implementation
- 01
Frame the contract
Write down the caller, data classification, success condition, timeout, cancellation, and ownership. Use it whenever a model can move from recommending an action to executing it in an external system.
- 02
Build one narrow path
Implement one end-to-end path with request correlation, typed state, and reversible failure handling. Classify tools by impact and pause before external messages, purchases, destructive writes, credential use, or privilege changes.
- 03
Prove the outcome
Turn acceptance into a repeatable fixture, contract test, or browser test. Audit records connect actor, intent, approval, tool arguments, execution result, and any compensating action.
03 / BOUND
Production boundary
Approval must show exact target, payload, consequences, and reversibility; a generic confirmation is insufficient.
04 / PROVE
Acceptance evidence
Audit records connect actor, intent, approval, tool arguments, execution result, and any compensating action.
SOURCE / HTTP
Reproducible source probe
curl -fsSI 'https://developers.openai.com/api/docs/guides/tools' | sed -n '1,5p'