Agentic Task Framing

The four-component structure for writing Claude Code task prompts that enable autonomous multi-step execution: Goal (what should be true when done), Scope (which files are in play), Verification (how to confirm success), and Escalation (when to stop and ask).

Syntax

claude-code

GOAL: [What should be true when the task is complete]

SCOPE: [Files or directories involved]

VERIFICATION: [Commands to run to confirm success]

ESCALATION: [Conditions that require stopping and asking before proceeding]

Example

claude-code

GOAL: Add email validation to the registration form.
Check format, normalize to lowercase, reject the 10 most common
disposable email domains (hardcoded list).

SCOPE: src/lib/validators.ts for the function,
src/services/auth.service.ts for integration,
src/services/auth.service.test.ts for tests.

VERIFICATION: Run npm run typecheck and npm test.
Both must pass before reporting complete.

ESCALATION: If existing auth tests need more than minor
updates to accommodate validation, stop and describe
what you found before proceeding.