Problem
The bottleneck in agent-executed development is not writing code, it is deciding what to
build. An agent executes a well-formed scoped issue reliably and produces churn against a
vague one. Today the decomposition step — brief in, milestones and scoped issues out — is
partly a CLI that expands loose prompts into gate-ready issues, always with a human review
before posting, and it is tuned on three repos by one author.
Productizing it means: it must accept a brief from someone else's head, produce plan-level
structure rather than a flat issue list, and be measurably good rather than assumed good.
Dependencies
Epic #10076 (issue-quality measurement) is the objective function. Without it, decomposer
changes cannot be evaluated — only preferred.
Sub-issues
Acceptance
A brief from an external submitter produces an approved criteria set and an executable
backlog whose issue-quality scores are comparable to hand-authored issues on existing
repos.
Problem
The bottleneck in agent-executed development is not writing code, it is deciding what to
build. An agent executes a well-formed scoped issue reliably and produces churn against a
vague one. Today the decomposition step — brief in, milestones and scoped issues out — is
partly a CLI that expands loose prompts into gate-ready issues, always with a human review
before posting, and it is tuned on three repos by one author.
Productizing it means: it must accept a brief from someone else's head, produce plan-level
structure rather than a flat issue list, and be measurably good rather than assumed good.
Dependencies
Epic #10076 (issue-quality measurement) is the objective function. Without it, decomposer
changes cannot be evaluated — only preferred.
Sub-issues
Acceptance
A brief from an external submitter produces an approved criteria set and an executable
backlog whose issue-quality scores are comparable to hand-authored issues on existing
repos.