overdo-orchestrate

Execute Overdo task graphs to completion with validation loops.

Use when:

• •Running a task tree end-to-end
• •Recovering from failed attempts or restarts

Bug Harvest Mode (Linear-first)

When running hardening/polish loops, use this strict sequence:

1. •Synthesize bugs and taste/polish feedback from current implementation.
2. •Open Linear issues for all discovered bugs in project test-loop.
   • •Apply label bug to every generated bug issue.
3. •Run linear-bug-plan on each issue.
4. •Run linear-bug-fix on each issue one at a time:
   • •create/switch to a fresh branch per issue
   • •implement + validate
   • •open PR
   • •switch back to main before starting next issue (critical)
5. •After bug PRs are open, process feature-request findings similarly and open PR(s) to base branch.
6. •Stop active loop execution after PR creation and wait for human review.

Guardrails:

• •If linear-bug-plan or issue status is unknown, do not auto-fix; label/escalate for manual triage.
• •Do not run parallel fixes on the same base branch checkout.
• •Persist issue -> branch -> PR mapping in evidence artifacts.

Workflow:

1. •Fetch next ready task(s).
2. •Dispatch worker claims with lease safety.
3. •Run required validation gates.
4. •Persist loop iteration result.
5. •Retry/escalate based on policy.
6. •Queue commit and finalize transaction.

Exit rule for PR-driven loops

Once all intended PRs are opened, do not continue looping automatically. Wait for review feedback, then resume via a new loop pass.

Rules:

• •Never mutate orchestration state outside MCP.
• •Always persist evidence for retries/escalations.
• •Resume from SQLite state after crash before starting new work.