/flow-orchestrate
Phase: Any
Usage: /flow-orchestrate
Processes decomposed issues sequentially overnight via flow-start --auto. Fetches open issues labeled “Decomposed”, filters out in-progress issues, and runs the full Start-Plan-Code-Review-Learn-Complete lifecycle for each one. Generates a morning report with results.
What It Does
- Fetches open issues with the “Decomposed” label, excludes those with “Flow In-Progress”
- Displays a prioritized queue table (same format as
flow-issues) and creates an orchestration state file at .flow-states/orchestrate.json
- For each issue in the queue:
- Invokes
flow-start --auto with the issue title and number
- The full 6-phase lifecycle runs autonomously
- Detects the outcome from GitHub PR state (merged = completed, closed = failed)
- Cleans up stuck features via
flow-abort --auto if needed
- Marks the orchestration complete
- Generates a summary report at
.flow-states/orchestrate-summary.md
Morning Report
The report is delivered in two ways:
- End of session: Rendered inline after the last issue completes
- Next session start: The session-start hook detects the completed orchestration, presents the report, and cleans up the state files
Compaction Survival
The orchestrator state file (.flow-states/orchestrate.json) tracks the queue position and per-issue outcomes. Self-invocation after each feature keeps the working context bounded. The session-start hook detects in-progress orchestrations and injects resume instructions after compaction.
Multi-Run Lifecycle
The “Decomposed” label is the queue:
- Completed issues are closed by
flow-complete and excluded from the next run
- Failed issues retain the label and re-enter the queue on subsequent runs
- New issues decomposed during the day enter automatically
No configuration or manual queue management needed.
Gates
- One orchestration per machine at a time (state file acts as lock)
- No parallel issue processing — sequential only
- No retries for failed issues in V1