Spec Optimizer
Run this workflow to improve specification quality without silently changing intent.
Workflow
- Define scope before editing.
- Scan spec files and classify findings into four buckets:
- Omission
- Redundancy
- Duplication
- Conflict
- Prioritize by severity:
- P0: behavior/spec contradictions that can split implementation
- P1: missing contract fields, broken links, incomplete matrices
- P2: wording drift and style inconsistencies
- Present findings first, with file+line references.
- For each conflict, ask the user exactly one decision question at a time (A/B or A/B/C).
- Apply only user-confirmed decisions.
- Re-run consistency checks and summarize changed files.
Output Contract
When reporting findings, use this structure:
- Severity + finding title
- Why it matters
- Evidence with explicit path and line
- Recommended options
- Single decision question
When reporting completion, include:
- Confirmed decisions list
- Files edited
- Residual risks or follow-ups
Conflict Resolution Rules
- Never auto-resolve semantic conflicts.
- Never batch multiple conflict questions in one prompt unless user requests batching.
- Keep options mutually exclusive and implementation-ready.
- If one option has clear architectural alignment, recommend it explicitly.
Editing Rules
- Keep changes minimal and local to the agreed scope.
- Preserve existing section structure unless restructuring is explicitly requested.
- Update related tables/types/examples together to avoid partial drift.
- When adding new fields to schema examples, update all corresponding command/DTO matrices.
AgentDispatch Checklist
Always verify these high-risk areas after edits:
- Task state machine diagram vs
TaskStatustype. - IPC command matrix vs CLI command sections.
- Auth/role text vs endpoint restrictions.
- ACP
newSession.cwdsemantics vs artifact output directory docs. - Module count and architecture summary consistency across overview docs.
Read detailed patterns in references/conflict-patterns.md when needed.