Use a document as the Native requirements entry point
You can provide requirements in several forms:
State what the document is for when you start:
Native turns the source into formal requirements
Native does more than preserve a summary. It splits the source by headings, paragraphs, lists, tables, code blocks, examples, constraints, and links, then records each source unit as:complete: the content was read in full and classified;partial: only part of the content was read, so valid requirements may still be missing;unavailable: the current platform or permissions cannot access it.
Every executable source unit must enter the complete target Specs and map to at least one acceptance item. Background information and non-goals keep their classifications and reasons without becoming implementation tasks. When a newer document corrects an older requirement, Native marks the old unit as
superseded and points to the active replacement.
Skip question-by-question clarification when the document is complete
Native first investigates facts it can determine from the repository, tools, and runtime environment. After mapping the source, it skips Sequential or Batch questions when all of the following are true:- all accessible content has been read in full;
- every active requirement is covered by the Specs and acceptance scope;
- the document provides one executable interpretation of the goal, scope, default behavior, and failure results;
- the document does not contain conflicting requirements;
- repository facts and project rules reveal no additional user choices;
- no
partial,unavailable, unmapped, or unconfirmed blockers remain.
“Skip clarification” means skipping unnecessary question rounds. Final
requirements confirmation remains in place so you can verify that the
generated Specs match the document before the Runtime begins long-horizon
execution.
Ask only about gaps in the document
Document length does not prove that requirements are complete. The following conditions create a[blocking] item:
- the link requires a login that the current Agent cannot use;
- the file is damaged, cannot be parsed, or was only partially read;
- two sections give different answers for the same behavior;
- wording such as “support failed-request retries” omits semantics that would change user-visible results;
- an active requirement has not been mapped to a Spec or acceptance item;
- the material may be debugging evidence, review input, or implementation reference, and its requirements purpose is unclear.
brief.md, the complete target Specs, and acceptance items before moving to final confirmation.
If a link is inaccessible, export the source as a file and deliver it to the current conversation. A summary cannot replace the complete source because omitted tables, examples, or constraints may still change the target behavior.
Use Grill me with docs for your preferred clarification flow
Native’s clarification protocol is useful when you want to develop requirements from an initial goal. If you prefer the intensive questioning style of Grill me with docs, use it first to discuss product decisions, failure paths, unstated assumptions, and non-goals, then produce a complete document. Deliver that document to Comet afterward:Use Native as a long-horizon task executor
After confirming the document, you can focus on the final result instead of manually advancing the workflow. Native continues by:- creating or binding the branch, worktree, and Change workspace;
- choosing implementation methods within the Specs and maintaining task progress;
- running required checks and submitting a verifiable Builder candidate;
- starting a fresh read-only Verifier to inspect the result against acceptance items;
- returning to Build after failed verification and rechecking the affected scope;
- recovering from project state after a session interruption, context compression, or device change;
- completing Archive and the confirmed delivery action after Verify passes.
Choose an entry point based on requirements maturity
Native provides its own clarification flow and also accepts requirements you have already developed. You can start Comet with a single goal or a mature document; both paths enter the same recoverable, acceptance-driven execution workflow.
Continue reading: Which decisions need your confirmation, Sequential and Batch clarification, Native Loop, and Continuous progression and stable recovery points.

