Skip to main content

Starting new work

The sequence​

  1. Open the repository.
  2. Confirm your working tree is safe to work in (no uncommitted work you'd lose track of).
  3. Identify the work item.
  4. Run:
/asterweave:deliver <id>

Asterweave determines, on its own, whether an existing specification is sufficient, whether story analysis is needed, whether blocking ambiguity exists, whether additional human input is required, and whether a repository-specific agent should handle a stage.

Possible outcomes​

OutcomeMeaningYour response
Proceeds to plan approvalchallenge returned READY or READY_WITH_ASSUMPTIONSReview the plan, approve or ask for changes
BLOCKEDA genuine product/business decision is missingAnswer the specific question(s) asked
needs-humanSomething else requires your judgment (an outcome the graph can reach at several nodes, not only challenge)Read the reason, provide the decision
policy-deniedA proposed action would weaken a mandatory gateChoose a safer approach — the gate will not be bypassed
security-escalationA security finding needs a remediation decisionReview the finding and decide

Example: a BLOCKED outcome​

/asterweave:deliver 4821

might report something like:

Work item: ADO-4821 Analysis: BLOCKED Reason: Expected behavior for partial refunds exceeds documented business rules. Questions:

  1. Should multiple partial refunds be allowed against the same payment?
  2. What happens when total refunds equal the original payment amount?

Answer the questions in your reply. Re-invoking /asterweave:deliver 4821 picks the workflow back up from its recorded state rather than starting over — see Continuing interrupted work.

Manual vs. autonomous responsibility​

Asterweave handles automatically: repository discovery, reading the work item, relevant source/test discovery, implementation planning mechanics, branch preparation, implementation, verification, agent routing, security/code review, PR creation, pipeline monitoring, retryable fixes within budget, and reporting back.

Stays yours, and Asterweave will pause for it: ambiguous product intent, high-risk or destructive actions, major architectural change, breaking API behavior without your approval, business-rule decisions not present in any source of truth, and accepting risk exceptions.

Asterweave is built to reduce routine interaction, not to remove your ownership of intent.

Daily workflow, Continuing interrupted work, Architecture overview for the full set of typed stop states.