Conduit v0.5.2: reliable refinement handoffs

Conduit v0.5.2 makes the refinement workflow more durable and easier to diagnose. Stories, research, and architect decisions now survive the transitions where they are needed instead of disappearing behind a restart.

Complete stories from draft to preview

Multiline Markdown sections, acceptance criteria, and implementation guidance are preserved when Conduit parses a story. The packet preview now renders the complete story alongside the generated specification artifacts, while approved packet files remain the committed source of truth.

Draft state and research reports stay in Conduit’s local project database. Draft writes are serialized to prevent a rapid edit or submit from being mistaken for a conflicting external update.

Research that survives retries

Completed research is restored after an architect startup failure and shown before another run begins. Conduit reuses that report by default; rerunning research is now a separate, explicit action.

Codex-specific final-output capture also lives in the Codex runner adapter, keeping provider behavior out of the refinement workflow and making captured reports consistent with the selected runner.

Errors and questions with enough context

Role launch failures now expose their underlying cause, bounded process output, remediation, and run identifier. Worktree hook failures and missing runners are reported distinctly, making configuration problems actionable without requiring a separate log hunt.

Architect questions use a stable Markdown contract and preserve the question, context, options, and smallest decision needed to continue. The clarification screen gives controls their own rows so instructions remain readable at supported terminal sizes.

Read the documentation to configure Conduit and start a refinement workflow.