菜单
Implements a feature end-to-end — spec, plan, build, review, simplify, ship — fully autonomous from a feature description file. Use when you have a feature described in a file and want the full pipeline to run without pausing for input.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | mb-implement |
| description | Implements a feature end-to-end — spec, plan, build, review, simplify, ship — fully autonomous from a feature description file. Use when you have a feature described in a file and want the full pipeline to run without pausing for input. |
Takes a path to a feature description file and drives the full implementation pipeline autonomously. At every decision point, apply your best judgment and default to the most conservative, idiomatic choice for this codebase. Do not pause for human input between phases.
The feature file path is passed as the skill argument ($ARGUMENTS). Read the file first.
If the file does not exist, stop and report the error.
Invoke the agent-skills:spec-driven-development skill.
Derive the spec from the feature file — do not ask clarifying questions. Where ambiguous, default to the minimal scope that satisfies the described goal. Cover all six areas: objective, commands, project structure, code style, testing strategy, and boundaries.
Save to tasks/spec.md (not SPEC.md in root). Announce ✓ Spec → tasks/spec.md and continue.
Invoke the agent-skills:planning-and-task-breakdown skill.
Read tasks/spec.md and relevant codebase sections. Slice work vertically — one complete path per
task. Write each task with acceptance criteria and a verification step. Save the plan to
tasks/plan.md and the ordered task list to tasks/todo.md.
Do not wait for human review. Announce ✓ Plan → tasks/todo.md and continue.
Invoke agent-skills:incremental-implementation alongside agent-skills:test-driven-development.
Work through every task in tasks/todo.md in order. For each task:
bun run check, bun run lint, bun run test — fix all errors before moving ontasks/todo.mdIf any check fails, apply agent-skills:debugging-and-error-recovery and resolve before continuing. Never skip a task or leave failures unresolved.
Announce ✓ Build complete when all tasks pass all checks.
Invoke agent-skills:code-review-and-quality.
Review all changes introduced in this session across five axes: correctness, readability, architecture, security, performance. Categorize findings as Critical, Important, or Suggestion.
Fix all Critical and Important findings. Apply Suggestions only when they are unambiguous
improvements with no behavior risk. Run bun run preflight after fixes; resolve any new failures.
Announce ✓ Review complete and continue.
Invoke agent-skills:code-simplification.
Target code introduced in this session. Apply simplifications incrementally — run bun run test
after each one. Revert any simplification that breaks tests. Run bun run preflight at the end.
Announce ✓ Simplification complete and continue.
Invoke agent-skills:shipping-and-launch.
Spawn code-reviewer, security-auditor, and test-engineer subagents in parallel (all three Agent tool calls in a single turn). Merge their reports. Fix all blockers. If a Critical finding cannot be fixed, halt and report — do not ship.
Once GO: confirm bun run preflight passes, then output the ship decision with a rollback plan.
Feature: [name from feature file]
Tasks completed: N
Commits: [list]
Review findings fixed: N critical, N important
Final status: GO / NO-GO
Long video production. Composes a video from a sequence of distinct shots. Trigger ONLY when the user explicitly asks to generate a video longer than 10 seconds (duration > 10s). DO NOT trigger if the requested video is 10 seconds or less.
Cinematography vocabulary for shot framing, camera angles, subject relationship, and camera movement. Load when the user's request involves shot-type choices or when constructing ask_user options for framing/movement settings.
Image generation. Produces a still image from text, edits an existing image, or composites multiple images. Use for any output that is a single still frame — concept art, character references, product shots, scene keyframes, style explorations, background plates, virtual try-on, or image editing.
Video generation. Animates a source image into a short video clip, or generates video from text. Use for any output that requires motion — bringing images to life, camera moves, atmospheric animation, or full scene videos.
Character generation and identity board planning. Creates an artistic, high-quality Character Identity Board to establish and lock a character's visual identity (face, clothing, body type, expression range, and silhouette) for future image and video generation.
Cinematic storyboard generation. Produces a compact multi-panel storyboard sheet (16:9 image) with monochrome rough-sketch panels, a designed masthead, and a director strip animatic track board. Use for any sequence that needs a shot-by-shot visual breakdown — action, performance, cooking montage, fantasy, or any genre.