Date: 2026-04-19
xflow should feel like a guided delivery loop, not a bag of internal atoms. Operators should know what to do next even when they do not know the workflow internals.
The loop is:
shape -> choose-track -> execute -> review -> finish
Use xflow guide when you need the shortest reminder of this loop.
Goal: turn a rough request into a readable problem frame before implementation.
Use:
xflow:planExpected artifacts:
proposal.mdplan.md
Rules:
- Keep planning track-neutral.
- Do not choose yolo or corps inside plan.
- Reuse an existing current
plan.mdwhen yolo or corps starts.
Goal: select the smallest track that can safely complete the work.
Use xflow:yolo when the change is bounded:
- backend logic
- docs
- scripts
- infrastructure
- small UI changes without a design loop
Use xflow:corps when the change needs heavier coordination:
- complex UI or product work
- multi-agent execution
- visual review
- mission control and gate observability
CLI equivalents:
xflow workflow validate yolo
xflow workflow validate corpsworkflow validate is preflight only. It must not be used as proof that
xflow:corps ran.
Goal: move work through deterministic phases while preserving evidence.
Use:
xflow workflow run yolo --title "<title>" --change-type backendor:
xflow corps --title "<title>" --change-type frontend --change-id <id>
xflow proof --track corps --change-id <id>Expected artifacts:
status.jsonverify_proof.jsoncorps_proof.jsonfor heavy-track completion- workflow logs
- phase outputs under
specs/changes/<change-id>/
Rules:
- Gates are stop signs, not decoration.
- Human gates require explicit approval.
- Verification evidence should be produced before archive.
corpscompletion requires the governed built-in workflow manifest, hash-linked execution-log witnesses for every canonical phase, successful atom/gate evidence, and nostub,task_queued, orpencil_stubbedfallback.- TDD is a first-class gate: the red proof runs before implementation, the green proof runs after implementation, and quality review rejects code changes without meaningful changed tests.
Goal: make completion defensible before shipping or handing off.
Run:
npm test
npm run drift:scan
npm run skill:diff
xflow scoreReview:
docs/reviewer-guide.mdverify_proof.jsonHANDOFF.md- current git status
Goal: close the loop without losing durable learning.
Archive ownership stays with:
A5.archive.commit_push_close
Expected durable outputs:
- refreshed
HANDOFF.md - updated
AHA.mdwhen there is a durable insight - merged spec/workflow deltas
- committed and pushed implementation
xflow does not rely on a remembered chat ritual. The methodology is backed by:
- executable YAML workflows
- schema and registry tests
- local readiness checks
- drift scans
- installed skill sync checks
- handoff and archive artifacts
xflow scoreevidence
The result is a workflow that is understandable to humans and enforceable by tooling.