ワンクリックでManusで任意のスキルを実行

journey-builder

スター0

フォーク0

更新日2026年7月8日 10:56

Build the real live-animals journey prototype (prototypes/standalone/live-animals in trade-imports-animals-frontend) on the obligations-v2 page-owned-spine model. Digest mode distils the requirement sources — Confluence "Live Animals Data Fields V4", the src/server skeleton journey, the interaction-design canvas — into a canonical machine-readable spec (journey-spec.json + conflicts.json) reviewed at a spec gate. Backlog mode derives ordered increments from the spec; build mode runs the serial implementor loop (one INCREMENT_IMPLEMENTOR subagent per increment, parent re-verifies, commit-or-rollback), halting at model-extension gates and milestone walk-throughs. Use when the user asks to digest journey requirements, regenerate the backlog, run/resume the build loop, or verify the prototype (triggers: "digest journey requirements", "journey spec", "journey-builder", "build the live-animals prototype", "run the loop"). NOT for the car-insurance spike itself or for generic ticket work.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

DEFRA

DEFRA/trade-imports-animals-workspace

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

journey-builder

Run-id: the EUDPA ticket (currently EUDPA-249 — no separate ticket for this programme). State lives in workareas/journey-builder/<run-id>/; the canonical spec lives in the frontend worktree at <workarea>/frontend-worktree/prototypes/standalone/live-animals/spec/ (branch spike/<run-id>-live-animals-spec) — never write into repos/trade-imports-animals-frontend directly: other agents work in that checkout.

Programme plan: ~/.claude/plans/so-in-the-frontend-reflective-yeti.md.

Mode: digest (Phase 1 — available now)

tools/journey-builder/prepare-digest.sh EUDPA-X — seeds workarea + worktree + cached sources + extract placeholders + spec skeleton. Idempotent; --refetch refreshes cached sources.
Fan out THREE general-purpose Task subagents in parallel, one per source (confluence-v4, skeleton, ixd-canvas), each told: "Follow ~/git/defra/trade-imports-animals-workspace/.claude/skills/journey-builder/references/SOURCE_EXTRACTOR.md for source ~~, run-id EUDPA-X."~~

Verify every extract has status: "complete" and non-trivial counts (jq .status,.fields,.pages,.behaviours per file). Re-spawn gaps — do not extract in the parent.

Spawn ONE general-purpose Task subagent: "Follow .../references/SPEC_RECONCILER.md, run-id EUDPA-X."

Parent re-runs tools/journey-builder/spec-lint.sh EUDPA-X — never trust the worker's green.

Spec gate: present to Sam — the uncommitted diff in the worktree, lint counts, conflicts, modelGap markers, and the open design questions (wipe-vs-retain, partial page completion, address copy-vs-reference, inline comments, provisional copy). Commit on the spec branch only after Sam approves.

Mode: backlog

tools/journey-builder/backlog-generate.sh EUDPA-X derives workareas/journey-builder/EUDPA-X/backlog.json from the spec: one increment per page in section order (add-page / add-collection), model-extension increments (gate: "sam", born blocked) before the first page needing each modelGap, then the car-domain removal tail (remove-car-section per baseline section + repoint-test-fixtures) — the vendored baseline ships the car domain to keep the engine-test net green; see prototypes/standalone/live-animals/PROVENANCE.md. Idempotent — re-running preserves statuses. Inspect with backlog-counts.sh / jq over the file.

Mode: build (the loop)

Serial by design — increments edit shared files (registry, flow, hub, CYA).

tools/journey-builder/next-increment.sh EUDPA-X --claim — pops the first runnable todo (deps done) and marks it inprogress; exit 3 = dry.

If the increment has gate: "sam" or closes a milestone → STOP, present to Sam (model-extension design panel / milestone walk-through).

Spawn ONE general-purpose Task subagent: "Follow ~/git/defra/trade-imports-animals-workspace/.claude/skills/journey-builder/references/INCREMENT_IMPLEMENTOR.md for run-id EUDPA-X, increment ." The persona owns the touch-lists (vendored docs/add-a-{page,field,collection}.md), the enforcedAt semantics, the never-author-gates rule (T11), and commit/rollback.

Parent re-verifies: tools/journey-builder/verify-increment.sh EUDPA-X — never trust the worker's green. Mismatch → rollback + failed.

Loop to 1. Halt early on 3 consecutive failures (systemic signal).

Per completed section run verify-increment.sh EUDPA-X --e2e; per milestone: full E2E + Sam walk-through.

Mode: verify

tools/journey-builder/verify-increment.sh EUDPA-X [--e2e] — unit suite + prettier + eslint over the prototype (log at <workarea>/.verify.log).

Tools

tools/journey-builder/: prepare-digest.sh, extract-add-item.sh, extract-finalize.sh, spec-add-field.sh, spec-add-page.sh, spec-add-conflict.sh, spec-add-behaviour.sh, spec-add-fieldgroup.sh, spec-lint.sh [--format].

このリポジトリの他の Skills

同じリポジトリ

test-stack-analysis

DEFRA/trade-imports-animals-workspace

'Given a feature description, Jira ticket, or Confluence URL, finds every related test file across the 8 EUDP Live Animals repos, classifies each by pyramid level (unit/integration/E2E) and concern type, and reports coverage gaps and cross-level duplication with file:line evidence — so a developer knows where to add or remove test coverage before writing new tests. Triggers: "test-stack-analysis EUDPA-X", "find test gaps EUDPA-X", "test pyramid analysis EUDPA-X", "analyse test coverage EUDPA-X", "check test pyramid for <feature>". One-shot, report-only — makes no code changes. NOT for correctness/style review of existing code (use `review` or `code-style`), NOT for verifying a PR author understands their own change (use `understanding-check`), NOT for scaffolding CLAUDE.md (Claude Code'\''s built-in `/init`).'

2026-07-160

govuk-upgrade

DEFRA/trade-imports-animals-workspace

Upgrade the govuk-frontend package across every EUDP Live Animals Node.js repo that consumes it. A four-stage workflow driven by canonical per-repo JSON state (`versions.{repo}.json`): Phase 1 = `start-upgrade.sh` (ticket → branch → discover repos → seed state + pre-bake CHANGELOG sections + best-practices bundle), Phase 2 = fan-out one `general-purpose` Task subagent per pending version following `references/VERSION_PLANNER.md` (uses `version-classify.sh` + `version-add-change.sh`), optional Walker = batch triage of pending plans following `references/PLAN_WALKER.md`, Phase 3 = strict-semver-order `apply-version.sh` per `todo` version (package.json + npm install + npm test + commit + state transition), then E2E once at the end. Stays inside the govuk-frontend toolbox — Nunjucks macros, govuk-* utility classes, no custom CSS or hand-rolled components. Triggers: "upgrade govuk-frontend", "govuk upgrade", "govuk-frontend upgrade", "bump govuk-frontend", "walk govuk EUDPA-XXX", "start-upgrade.sh". NOT for non-go

2026-07-090

code-style

DEFRA/trade-imports-animals-workspace

Multi-language code-style/lint review (formatting, conventions, style rules) and remediation for EUDP Live Animals PRs (EUDPA-*). Reviews Java, GDS/Nunjucks (.njk), Playwright specs, k6 and Node source files, each against its context-appropriate best-practices bundle (Java modern-java + Javadoc; GDS components/styles/patterns; Playwright; k6; Node 17-rule style guide + JSDoc). Supports fresh review, refresh review (re-review after new commits), interactive walker that triages findings one item at a time, and batched implementor that applies queued style fixes. Fans out per-file review and per-file implementation to `general-purpose` Task subagents that follow worker personas under `references/`. Use when the user asks for code style/lint review or to apply agreed style fixes (triggers: "style review EUDPA-", "code style review", "re-style review", "style refresh", "walk style EUDPA-", "triage style", "fix style EUDPA-", "implement style fixes", "lint review"). NOT for correctness/design review across language

2026-07-060

npm-upgrade

DEFRA/trade-imports-animals-workspace

Upgrade (non-govuk-frontend) npm packages across the EUDP Live Animals repos via a three-phase workflow plus interactive walker — discover outdated packages, classify each as auto (no code changes) or manual (breaking changes), run automated upgrades with rollback safety, produce a handoff manifest for the remaining manual work, then walk the manual list keystroke-by-keystroke and spawn a per-package implementor worker on demand. Per-package classification and implementation state lives in canonical JSON (`packages.{repo}.json`) — no markdown plan files on disk. Fans out per-package research in Phase 1 to `general-purpose` Task subagents following `references/PACKAGE_PLANNER.md`, and per-package implementation from the walker to subagents following `references/MANUAL_UPGRADE_IMPLEMENTOR.md`. Use when the user asks to bring npm packages up to date across repos (triggers: "upgrade npm deps", "upgrade npm dependencies", "upgrade dependencies", "run npm upgrades", "walk upgrade EUDPA-XXX", "triage upgrade EUDPA-X

2026-07-060

review

DEFRA/trade-imports-animals-workspace

Code review for correctness, security, error handling, performance, best-practices and test coverage across all languages and repos (Java, Node.js, frontend, tests) for EUDP Live Animals tickets. Handles fresh first-pass review, refresh (re-review after further work, merge conflicts or coverage gaps), interactive walker that triages findings one item at a time, and batched implementor that applies queued fixes. Fans out per-file reviews, per-repo consistency analysis and per-item fixes to `general-purpose` Task subagents that follow worker personas under `references/`. Use when the user says "review EUDPA-XXX", "code review", "re-review EUDPA-XXX", "refresh review", "check fixes", "walk review EUDPA-XXX", "triage review", "implement review EUDPA-XXX", or "apply review fixes". NOT for JS lint/format/style findings — use the code-style skill for those.

2026-07-060

skill-creator

DEFRA/trade-imports-animals-workspace

Meta-skill for authoring new workspace skills and auditing existing ones against the 8-pattern checklist. Two modes — CREATE scaffolds a new skill end-to-end (SKILL.md + references/ + tools/<name>/ + assets/ + .claude/settings.json allowlist), AUDIT walks an existing skill (or fans out across all skills) and produces a plan document under workareas/skills-audit/<name>.md. Use when the user wants to add a workspace skill or assess an existing one (triggers: "scaffold skill <name>", "skill-create <name>", "new workspace skill <name>", "audit skill <name>", "audit skills", "review skill <name> against patterns"). Disambiguated from Claude Code's built-in /init (which scaffolds CLAUDE.md, not workspace skills). NOT for editing skills you already understand — open the SKILL.md and edit directly. NOT for reviewing code correctness/security — use `review`; NOT for JS lint/style review — use `code-style` (AUDIT mode assesses a skill's own structure against the 8-pattern checklist, not the code a skill produces).

2026-07-060

name

journey-builder

description

journey-builder

Programme plan: ~/.claude/plans/so-in-the-frontend-reflective-yeti.md.

Mode: digest (Phase 1 — available now)

tools/journey-builder/prepare-digest.sh EUDPA-X — seeds workarea + worktree + cached sources + extract placeholders + spec skeleton. Idempotent; --refetch refreshes cached sources.
Fan out THREE general-purpose Task subagents in parallel, one per source (confluence-v4, skeleton, ixd-canvas), each told: "Follow ~/git/defra/trade-imports-animals-workspace/.claude/skills/journey-builder/references/SOURCE_EXTRACTOR.md for source ~~, run-id EUDPA-X."~~