Top-level Workflows¶

Two slash commands in this plugin are orchestrator-only: they do not implement, review, or merge anything themselves — they sequence other skills and agents through their phases, holding the human gates between them.

/ship — spec → plan → build → review → PR, end-to-end.
/test-improve — seven-phase consolidated analyze-then-improve orchestrator for legacy or in-flight test suites.

Both follow the same orchestrator contract: delegate every phase to the owning skill or agent, honor human gates, surface ambiguous inputs in one batch up front, and report outcomes concisely.

`/ship`¶

File: skills/ship/SKILL.md Role: orchestrator. Use when: the user says "ship this", "take this feature end to end", or wants the spec → plan → build → PR flow without re-assembling it each time.

Steps¶

#	Step	Delegates to	Human gate after?
1	Approach contract — screen request against `knowledge/decision-defaults.md`; resolve ambiguous high-reversal-cost axes in one batch.	(orchestrator only)	yes, if a blocker remains
2	Spec (skipped with `--skip-spec`) — produce Intent, Architecture, Acceptance Criteria.	`/specs`	yes — operator approves the spec
3	Plan — decompose into vertical slices with Gherkin scenarios; a tier-scaled set of plan-review personas (1–5, by plan complexity) runs in parallel before the gate.	`/plan`	yes — operator approves the plan
4	Build — RED-GREEN-REFACTOR per slice, inline review checkpoints, verification evidence. Do not proceed until the suite is green.	`/build`	no
5	Review — run quality-review agents and let the auto-fix loop converge. Only judgment-call findings escalate to the operator.	`/code-review`	no
6	PR — pre-PR quality gate, open PR, arm auto-merge by default (`--no-auto-merge` to opt out).	`/pr`	yes — the PR is the final review artifact
7	Report — PR URL, quality-gate result, whether auto-merge is armed.	(orchestrator only)	—

Agents involved (dispatched by the delegated skills)¶

/build's inline review checkpoints dispatch the review agents listed in team-structure.md → Review Agent Dispatch. /plan dispatches a tier-scaled subset of the five plan-review personas (prompts/plan-review-*.md) — the Acceptance Test Critic always runs; the rest are added as the plan's tier (trivial/standard/complex) warrants — and the progress-guardian gate-keeper. /code-review re-runs the same review agents over the full changeset.

Arguments¶

/ship <feature-description> [--skip-spec] [--no-auto-merge]

Notes¶

Sequencing only — every gate, fix loop, and evidence requirement comes from the underlying skills. If any phase stops at a gate, /ship stops with it.
For a plan-only pass, use /plan; for build-only, use /build.
Resume across sessions with /continue.

`/test-improve`¶

File: skills/test-improve/SKILL.md Role: orchestrator.

Consolidated analyze-then-improve test orchestrator. Defaults to lightweight ceremony, prompts for heavier capabilities on demand, and always baselines coverage (and mutation, when enabled) before any test change.

Phases¶

Each phase writes a progress file to memory/test-improve/<slug>/phase-<n>.md so /continue (and --from-phase) can resume.

Phase 0 — Approach contract. Batched prompt (Enter accepts all defaults): mutation [off], BDD rubric [none], refactor [no-refactor], quality targets, sink (--parent <url> vs local files). Go stack shows the alpha go-mutesting advisory before the mutation prompt. Answers are immutable for the run.
Phase 1 — Analyze. Delegate to /test-health (sole worker). No separate calls to /cd-test-architecture, /test-design, /mutation-testing. Mutation section respects Phase-0 setting.
Phase 2 — Baseline (before any test edit). /coverage-baseline --workflow test-improve unconditionally; /mutation-testing --baseline --workflow test-improve when mutation is on. Go = advisory-only marker. Honest score = hard kills, timeouts separate.
Phase 2b — Derive Gherkin (conditional). none skips entirely; xunit-with-annotations writes .feature files without a runner; bdd-runner wires the native parser.
Phase 3 — Triage. /issues-from-assessment --workflow test-improve partitions findings into NO_REFACTOR (Phase-4 Stories) / REFACTOR_REQUIRED (deferred to Phase 5) / LOW_VALUE (advisory-only).
Phase 4 — Improve without refactoring. Per Story: /build (no-refactor) → /coverage-delta --workflow test-improve --story <id> → mutation-kill agent (--file <story-file> --max-rounds 3, [c/r/w/q] on residuals). End-of-phase review loop runs /test-design --since and /code-review --since in parallel, /apply-fixes then re-run, cap 2 iterations, [r/w/q] escalation. Evidence in phase-4-review.json.
Phase 4b — Refactor decision prompt. [y] enter Phase 5 / [b] backlog and skip to Phase 6 / [q] quit.
Phase 5 — Refactor-for-testability (conditional). Only when [y]. Seam-only production-code changes; existing tests are immutable. Same end-of-phase review loop; evidence in phase-5-review.json.
Phase 6 — Validate. /quality-targets-converge --workflow test-improve. Mutation off = skipped (not waived). Go = advisory-only. Coverage < 90% in no-refactor mode → [y/n] re-run-in-refactor-allowed prompt lists backlogged items.
Phase 7 — Executive-summary report. Interpolates the shipped templates/executive-summary.md from memory/test-improve/<slug>/ files to reports/test-improve/<repo-slug>-<date>.md. 10 numbered sections; empty sections render "Not applicable" (never omitted). Parent tracker (or plans/test-improve/FEATURE.md) is updated with a link to the report. Report is regeneratable from memory.

Invocation¶

/test-improve <repo-path> [--parent <url>] [--analyze-only] [--from-phase <n>] [--stack <id>]

Flow diagram: diagrams/test-improve-flow.svg.

Why these are documented together¶

/ship and /test-improve are the two multi-phase pipelines with inter-phase human gates in the plugin. Every other slash command is either a single-step worker (e.g. /coverage-baseline, /triage) or a one-shot orchestrator that returns in a single pass (e.g. /code-review, /test-design). Knowing the phase order, the owning skill or agent for each step, and where the human gates fall is the difference between operating these workflows confidently and re-reading every SKILL.md each time.