define-goal

Goal contracts convert vague feature ideas into measurable acceptance criteria by applying repo-grounded recon and transcript-verifiable scope. This process distills plain-language intent into outcome-based contracts with defined evidence and stop conditions so autonomous agents execute work honestly.

Scoping a backlog from a long feedback document is done by distilling multi-item notes into queue-ready work items. The process reads local project rules and captures ambiguous items as goal files in the repo-backed docs/goals queue, preserving human approval before dispatch.

Claude Code can queue work items without starting implementation by using built-in guardrails that prevent autonomous execution. It writes queue-safe goal files and preserves human approval for ambiguous items, ensuring the agent stops before writing any code until explicitly dispatched.

Defining acceptance criteria for bugs and chores in Droid requires converting plain-language intent into outcome-based contracts with explicit scope and stop conditions. It applies repo grounding to match the actual codebase verification commands, ensuring goals are verifiable and queue-safe.

Writing goal contracts relies on a repo-backed docs/goals queue directory to park work items for dispatch. The process reads existing local project rules and verification commands from the repository, ensuring the generated goals match the actual codebase instead of assumptions.