| name | dev-flow |
| description | Lightweight development flow — the default track for everyday work. source-only changes, dev/staging live applies, and zero-user production (owner-declared). Evidence is a merged PR + green CI + one ledger line; no sealing, no successor tickets, no verdict ping-pong. Includes the one-question track test that routes irreversible or customer-visible work to prod-flow, hollow-green defenses, and a fix-plus-regression-gate rule for problems found live. Trigger phrases: "dev-flow", "start the flow", "which track", "can we ship this?". |
dev-flow — lightweight development flow (default track)
English | 日本語
Principle (one line)
Concentrate rigor at irreversible mutation gates; everywhere else, ship fast
and let CI + PR review + regression gates carry the discipline. Ceremony that
duplicates a runtime fail-closed check is paying for the same safety twice.
This split was distilled from a real failure mode: a flow built for one-shot
irreversible launches (sealed receipts, successor tickets, verdict documents)
leaked into everyday development and made everything slow — while adding no
safety where it was applied.
The track question (ask first, every time)
"If this fails, do we lose a non-refundable consumable (approval, signature,
publication, quota) — or does a real customer see it?"
- No → this skill (lightweight, the default).
- Yes → prod-flow — the full staged flow with
packet sealing and a human GO gate.
Examples:
- Lightweight: source-only changes (PR/CI cycle), live applies to dev/staging
estates, production with zero real users (owner-declared — see below).
- prod-flow: customer-visible production mutation, signing/publication
semantics, org-level authority (account vending, permission changes),
anything that consumes a one-shot tuple.
Never carry prod-flow equipment (packets, sealing, successor tickets,
write-ahead receipts) into the lightweight track.
Lightweight rules
- No sealing, no receipt-SHA reporting, no successor tickets. A failure
means: fix it and run again.
- Evidence = merged PR + green CI + one appended line in the project ledger
(CHANGELOG/EVENT or equivalent).
- Scout = look (read-only) and paste your notes into the issue. Read-only
discipline still applies; guessing instead of looking is still forbidden.
- PoC and certification collapse into "show it working once in an isolated
environment" — necessity at the implementer's discretion.
- After a live apply, leave a one-line readback (the fact, no ceremony).
- No state file. The issue / PR is the source of truth.
What is NOT cut (real gates, not ceremony)
- Hollow-green defenses. Test ownership separation (RED tests written and
owned by a different lane; any implementer diff to tests = instant
rejection), a held-out test at review, and real-boundary green as the merge
gate. A fake environment cannot tell an honest implementation from a hollow
one.
- 1 issue = 1 lane (issue-lane skill).
- Environment identification is measured, not assumed — estate mix-ups are
an accident class unrelated to speed, and cutting ceremony does not license
cutting this.
Zero-user production (owner-declared exception)
Production with no real users may run on this track by explicit owner ruling.
The trade is a single rule:
- A problem found live is closed only with fix + a recurrence gate
(regression test / QA walk / harness addition) in the same change.
- The same problem appearing live twice = the gate is defective → circuit
breaker: stop and redesign the gate, don't just try harder.
- The exception expires the day real users arrive. Expiry is declared by the
owner, never assumed.
This flips the discipline from pre-emptive ceremony to post-hoc hardening:
release fast, find problems live, make each one structurally unrepeatable.
Review rules (no verdict ping-pong)
- No CONFORMS / NOT-CONFORMS verdict documents on this track. Review = PR
approve / request-changes with inline comments.
- Reviewers state all blocking findings in the first pass (no
drip-feeding). A late blocking finding that was visible in round 1 is a
reviewer defect (breakage newly introduced by rework is exempt).
- Re-review checks only the delta of the requested changes.
- Two rounds without convergence = circuit breaker: stop the async
ping-pong; put reviewer and implementer in one conversation, or have the
owner adjudicate the remaining points.
Circuit breaker
Count progress in outcomes (merged PRs, readbacks) — not process artifacts. On
2 rejections of the same operation, 30 minutes of stall, or a second
hollow-green from the same lane: stop and change the method (re-slice the
issue, switch to contract-test-first, or apply a ledger-controlled gear boost).
Project adaptation
If the project has its own canonical flow document, that document wins; this
skill is the default for projects without one. Pair it with
prod-flow for the heavy path, and record the track
ruling ("what counts as consumable / customer-visible here") in the project's
canonical doc as it accumulates.