跳到正文
cccskills
免费在 GitHub 公开

nw-fast-forward

Fast-forwards through remaining waves end-to-end without stopping for review between waves.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md5.4 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

NW-FAST-FORWARD: Fast-Forward

Wave: CROSS_WAVE (entry point) | Agent: Main Instance (self — orchestrator) | Command: /nw-fast-forward

Overview

Chains remaining waves end-to-end after single user confirmation. Detects current progress (like /nw-continue), shows planned sequence, runs each wave automatically — DISCUSS > DESIGN > DEVOPS > DISTILL > DELIVER — without stopping between waves.

You (main Claude instance) run this orchestration directly. Each wave invoked by reading its task file.

DISCOVER skipped by default — requires interactive customer interview data that cannot be auto-generated. Use --from=discover to include.

Behavior Flow

Step 1: Input Parsing

Accept: feature description (new project)|--from flag with optional feature ID|no arguments (auto-detect from docs/feature/)

Step 2: Project Resolution

New project (description provided): Derive feature ID per ~/.claude/nWave/skills/common/wizard-shared-rules.md (Feature ID Derivation). Show derived ID, allow override via AskUserQuestion. Create docs/feature/{feature-id}/.

Existing project (no description): Scan docs/feature/ (same as /nw-continue Step 1-2). If multiple, ask user to select.

Step 3: Detect Current Progress

Check wave artifacts using Wave Detection Rules in ~/.claude/nWave/skills/common/wizard-shared-rules.md.

Step 4: Determine Wave Sequence

Default order (DISCOVER skipped): DISCUSS > DESIGN > DEVOPS > DISTILL > DELIVER

With --from flag: Validate prerequisite artifacts exist. If missing: "Cannot start from {wave} — {missing} artifacts missing. Run /nw-continue to fill the gap." Start from specified wave.

Without --from: Find first incomplete wave, start from there.

Step 5: Show Plan and Confirm

Feature: {feature-id}

  Fast-forward plan:
    1. DISCUSS  — Define requirements and user stories
    2. DESIGN   — Architecture and technology selection
    3. DEVOPS   — Platform and infrastructure readiness
    4. DISTILL  — Acceptance tests (Given-When-Then)
    5. DELIVER  — TDD implementation

  This will run all 5 waves without stopping for review.

If some waves complete, show as skipped:

  Fast-forward plan:
    ✓ DISCUSS  — complete
    ✓ DESIGN   — complete
    1. DEVOPS   — Platform and infrastructure readiness
    2. DISTILL  — Acceptance tests (Given-When-Then)
    3. DELIVER  — TDD implementation

  3 waves will execute without stopping.

One-time confirmation via AskUserQuestion.

Step 6: Sequential Execution

For each wave:

  1. Read task file (nWave/tasks/nw/{wave}.md)
  2. Follow instructions — invoke appropriate agent via Task tool
  3. Wait for completion
  4. Verify output artifacts exist (wave detection rules)
  5. Missing artifacts after wave = failure
  6. Proceed to next wave

Between waves show brief status:

✓ DISCUSS complete — user-stories.md, story-map.md created
  Starting DESIGN...

Step 7: Failure Handling

If any wave fails:

  1. Stop immediately — do not proceed
  2. Show error:
    ✗ {WAVE} failed
      Error: [details]
      Progress saved. Run /nw-continue to resume from {WAVE}.
    
  3. Suggest /nw-continue for manual resume
  4. Do NOT retry automatically

Step 8: Completion

✓ Fast-forward complete for {feature-id}

  DISCUSS   ● complete
  DESIGN    ● complete
  DEVOPS    ● complete
  DISTILL   ● complete
  DELIVER   ● complete

  All acceptance tests pass. Feature is ready.

Error Handling

ErrorResponse
No description and no existing projectsSuggest /nw-new
--from with missing prerequisitesList missing artifacts, refuse
Wave failure mid-pipelineStop, show error, suggest /nw-continue
Artifact verification fails after waveTreat as wave failure
Name conflict on new projectSame as /nw-new conflict handling

Success Criteria

  • Project resolved (new or existing)
  • Current progress detected accurately
  • Planned wave sequence shown to user
  • User confirmed one-time before execution
  • Each wave executed in sequence
  • Output artifacts verified between waves
  • Failure stops pipeline with clear error and /nw-continue suggestion
  • Completion summary shown

Examples

Example 1: Full fast-forward from scratch

/nw-fast-forward "Upgrade authentication to OAuth2"

Wizard derives oauth2-upgrade, detects no prior artifacts, shows plan: DISCUSS > DESIGN > DEVOPS > DISTILL > DELIVER. User confirms. All 5 waves execute in sequence.

Example 2: Fast-forward from mid-pipeline

/nw-fast-forward

Wizard finds notification-service with DISCUSS complete. Shows plan: DESIGN > DEVOPS > DISTILL > DELIVER. User confirms. 4 waves execute.

Example 3: Fast-forward with --from flag

/nw-fast-forward --from=distill rate-limiting

Wizard validates DESIGN artifacts exist for rate-limiting. Shows plan: DISTILL > DELIVER. User confirms. 2 waves execute.

Example 4: Fast-forward with failure

/nw-fast-forward "Add payment processing"

DISCUSS succeeds, DESIGN succeeds, DEVOPS fails. Pipeline stops. Shows error and suggests /nw-continue to resume from DEVOPS.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation

中文简介正在准备中。当前显示原文说明。

nWave-ai/nWave6222026年9月16日 更新

Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation

中文简介正在准备中。当前显示原文说明。

nWave-ai/nWave6222026年9月16日 更新

Review dimensions for acceptance test quality - happy path bias, GWT compliance, business language purity, coverage completeness, walking skeleton user-centricity, priority validation, observable behavior assertions, traceability coverage, and walking skeleton boundary proof

中文简介正在准备中。当前显示原文说明。

nWave-ai/nWave6222026年9月16日 更新

Detailed 5-phase workflow for creating agents - from requirements analysis through validation and iterative refinement

中文简介正在准备中。当前显示原文说明。

nWave-ai/nWave6222026年9月16日 更新

5-layer testing approach for agent validation including adversarial testing, security validation, and prompt injection resistance

中文简介正在准备中。当前显示原文说明。

nWave-ai/nWave6222026年9月16日 更新

Architectural style selection decision matrices, trade-off analysis, structural enforcement rules, and combination patterns. Load when choosing or evaluating architecture styles.

中文简介正在准备中。当前显示原文说明。

nWave-ai/nWave6222026年9月16日 更新

查看 nWave-ai 的全部技能

报告此技能的问题