An agentic skills framework & software development methodology that works.
Claude Code skill for AI filmmaking plans
This skill turns a script, paragraph, or keyframe into a structured production pass for AI video. `SKILL.md` loads only the needed references and templates for each request, then shapes the result into beats, shots, prompts, sound, edits, and QC.
Builders who use Claude Code to plan AI video scenes, prompts, and edits.
You can turn one idea into a repeatable filmmaking workflow instead of rewriting prompts by hand.
What it does
Full production pipeline
Covers intake, script reading, beats, style overlays, director notes, blocking, shot planning, keyframe prompts, motion prompts, sound, edit assembly, and QC.
Director-style overlays
Includes 20 director modules such as Spielberg, Kubrick, Wong Kar-wai, Nolan, and others that adjust the same scene with different camera and rhythm rules.
Tool adapter routing
Maps the plan to video and image model capability classes and prompt shapes for tools like Runway, Veo, Kling, Luma, and others.
Continuity and repair system
Provides continuity bibles, failure modes, and a QC repair loop so prompts can be corrected instead of rerun blindly.
How to get it
- 1Skill files live in ~/.claude/skills/ / (user-level) or .claude/skills/ /…
git clone https://github.com/wuwangzhang1216/DirectorSKILL.git ~/.claude/skills/cinematic-director
- 2Run
git clone https://github.com/wuwangzhang1216/DirectorSKILL.git .claude/skills/cinematic-director
- 3Run
You: Turn this into a 4-shot AI video plan for Runway, 16:9. "A courier hesitates outside a Republican-era Shanghai apartment door before knocking, then slips an envelope under it and walks away." - 4Run
You: Same scene, but direct it in the style of Wong Kar-wai.
- 5Run
You: My generation looks like a slideshow — character barely moves, rain is frozen. Diagnose and rewrite this prompt: [paste prompt] - 6Run
You: I have 8 clips. Two don't match. Give me an edit plan.
README
Cinematic Director Skill
A Claude Code / Claude Agent skill that turns a script, a paragraph of prose, or a single keyframe into a complete production plan — beats, director's book, blocking, shot list with coverage, keyframe prompts, image-to-video motion prompts, a sound plan, an edit timeline, a continuity bible, and a coded QC repair loop. Optionally apply one of twenty director-style lenses as a coherent overlay across the entire pass.
Why this exists
Most "cinematic" AI prompts are adjective soup — cinematic, dramatic, masterpiece, 8K. They don't tell a model where the camera is, what the body is doing, where the action ends, or which anchors must not drift. The result is the slideshow look: pretty stills that wander between shots.
This skill replaces adjective soup with directorial reasoning:
- A shot is a story decision, not a style descriptor. If you can't say what the audience now understands that they didn't a second ago, the shot gets cut.
- Blocking comes before framing. Camera placement follows what bodies do in space. A well-blocked scene reads even shot flat.
- The keyframe is the control point. A video model mostly re-animates what the still already decided. Fixing the still is cheaper than retrying the video.
- Continuity is engineered, not hoped for. Identity strings, light direction, screen direction, and era locks are written once and repeated verbatim.
- Failures get diagnosed, not re-rolled. Nineteen coded failure modes, each with its mechanism, its ranked causes, and the cheapest fix that addresses it.
- Style is a lens, not a costume. Choosing Kubrick means wide-angle one-point perspective pressing people into architecture — not "add a zoom."
What's new in 2.0
v1 knew the vocabulary of directing. v2 knows the craft behind it.
| v1.2 | v2.0 | |
|---|---|---|
| Pipeline | 10 steps | 13 steps — adds intake/scope, sound & dialogue, edit & assembly |
| Output modes | 6 | 10 — adds beat sheet, director's book, sound plan, edit plan |
| Reference files | 4 | 13 |
| Director lenses | 14 | 20 |
| Templates | 4 | 8, all with filled worked examples |
| Context strategy | load everything | routing table — depth moved behind it, so SKILL.md loads one file set per request instead of all of them |
| Tool handling | six named adapters | capability-first routing + four prompt shapes + twelve adapter families |
| Failure handling | a checklist | coded manual F1–F19, cost ladder, three-strike rule |
Also new: lens and focal-length psychology, lighting ratios, continuity geometry (axis of action, the 30° rule, screen direction, eyeline match) with the AI-specific problem that a video model has no idea where the last shot put the camera, a staging-geometry library, per-genre playbooks, a controlled prompt lexicon with an EN↔中文 term table, and the operational side — shot difficulty scoring, retry budgets, versioning, and handoff.
See CHANGELOG.md for the full list. The step renumbering (10→13) and mode renumbering (A–F→A–J) are breaking changes.
Architecture
┌────────────────────────────────────────────────────────────────────────┐
│ SKILL.md — always loaded, routes everything │
│ │
│ routing table: request type → output mode → files to load │
│ hard rules · intake schema · defaults · self-check │
│ │
│ 1 Intake & scope 8 Keyframe strategy │
│ 2 Script & subtext 9 Tool adapter selection │
│ 3 Beat map 10 AI video prompt construction │
│ 4 Style lens (optional) 11 Sound & dialogue plan │
│ 5 Director's book 12 Edit & assembly plan │
│ 6 Blocking & staging 13 QC & repair loop │
│ 7 Shot list & coverage │
└────────────────────────────────────────────────────────────────────────┘
│ loads on demand, never all at once
┌───────────┴────────────────────────────────┐
▼ ▼
┌───────────────────────────────┐ ┌─────────────────────────────┐
│ references/ │ │ assets/ (fill-in templates)│
│ craft │ │ beat-sheet Mode B│
│ cinematic-language │ │ director-book Mode C│
│ blocking-and-staging │ │ shot-plan Mode D│
│ lighting-and-color │ │ keyframe-prompt Mode E│
│ sound-and-dialogue │ │ video-prompt Mode F│
│ editing-and-assembly │ │ sound-plan Mode H│
│ genre-playbooks │ │ edit-timeline Mode I│
│ product-and-macro │ │ qc-checklist Mode J│
│ generation │ └─────────────────────────────┘
│ prompt-lexicon │
│ ai-video-tool-adapters │
│ image-model-adapters │ ┌─────────────────────────────┐
│ failure-modes │ │ evals/evals.json │
│ production │ │ behavior contract │
│ continuity-bible Mode G │ └─────────────────────────────┘
│ production-workflow │
│ director_styles/ 20 lenses │
└───────────────────────────────┘
One skill, optional style overlay. SKILL.md is always the entry point. When the user names a director, Step 4 loads exactly one style file, and its 风格参数 YAML block overrides defaults at Steps 5, 7, 8, and 10. Style files are reference modules — they are not separately registered skills, so the host agent never has to disambiguate.
Installation
Skill files live in ~/.claude/skills/<skill-name>/ (user-level) or .claude/skills/<skill-name>/ (project-level). The folder name must match the name: field in SKILL.md (cinematic-director).
git clone https://github.com/wuwangzhang1216/DirectorSKILL.git ~/.claude/skills/cinematic-director
git clone https://github.com/wuwangzhang1216/DirectorSKILL.git .claude/skills/cinematic-director
Reload Claude Code (or your host agent) and the skill becomes available.
Updating
An install is a git checkout, so update it in place:
cd ~/.claude/skills/cinematic-director && git pull
Check which version you are running — metadata.version in the installed SKILL.md frontmatter:
grep -A2 '^metadata:' ~/.claude/skills/cinematic-director/SKILL.md
2.0.0 renamed and extended the output-mode letters (A–F became A–J) and the pipeline steps (10 became 13). An older checkout answers with the old letters, so a mode this README documents — H, I, J — looks as though the skill simply does not have it. That symptom means you are on 1.x. Pull, then reload the host agent.
Quick start
Plain directorial pass
You: Turn this into a 4-shot AI video plan for Runway, 16:9.
"A courier hesitates outside a Republican-era Shanghai apartment door
before knocking, then slips an envelope under it and walks away."
Produces: beat map → director's book with invariant clauses → shot list (story function, size, lens, angle, camera, blocking start→motion→end, light direction, continuity anchors, risk) → Runway-shaped motion prompts → QC notes.
With a director-style overlay
You: Same scene, but direct it in the style of Wong Kar-wai.
Step 4 loads references/director_styles/09_wong_kar_wai.md and its parameter block overrides the defaults: handheld in narrow corridors, neon practicals, frame-within-frame through doorways, a step-printed beat on the slip-and-leave, voice-over from the courier, a shorter average shot length. Same four shots, a different film.
Repairing a failed clip
You: My generation looks like a slideshow — character barely moves, rain is
frozen. Diagnose and rewrite this prompt: [paste prompt]
Runs the triage tree, names the failure code and its mechanism, then applies the cheapest fix on the cost ladder that actually addresses it — and tells you when the answer is to re-plan the shot rather than rewrite the prompt again.
Cutting what you already generated
You: I have 8 clips. Two don't match. Give me an edit plan.
Mode I: a timeline with in/out points, trim handles, transitions built from paired last/first frames, a continuous audio bed under the whole sequence, and the specific fix-in-edit patch for the two that don't match.
Director style overlays
| # | Director | Lens in one line |
|---|---|---|
| 01 | Steven Spielberg 斯皮尔伯格 | Emotional mainstream narrative, family stakes, awe via reaction-before-reveal |
| 02 | Alfred Hitchcock 希区柯克 | Suspense via shared knowledge, voyeurism, mistaken identity |
| 03 | Stanley Kubrick 库布里克 | Wide-angle one-point perspective, central symmetry, cold formalism |
| 04 | Akira Kurosawa 黑泽明 | Epic ensemble, humanism, weather as drama, dynamic action space |
| 05 | Martin Scorsese 斯科塞斯 | Urban energy, guilt, voice-over, kinetic camera, moral fall |
| 06 | Federico Fellini 费里尼 | Dream logic, circus parade, autobiographical fantasy |
| 07 | Ingmar Bergman 伯格曼 | Interior psychological drama, face close-ups, existential silence |
| 08 | Andrei Tarkovsky 塔可夫斯基 | Poetic time, long takes, water and fire, spiritual search |
| 09 | Wong Kar-wai 王家卫 | Urban loneliness, missed connections, neon, fragmented memory |
| 10 | Christopher Nolan 诺兰 | High-concept structure, time puzzles, large-format practical spectacle |
| 11 | Denis Villeneuve 维伦纽瓦 | Monumental scale, mist, geometric symmetry, low-rumble silence |
| 12 | David Fincher 芬奇 | Surgical control, teal-amber palette, procedural investigation, restrained dread |
| 13 | Nicolas Winding Refn 雷弗恩 | Neon color blocks, symmetric ritual framing, slow gaze, sudden violence |
| 14 | Bi Gan 毕赣 | Misty SW-China small towns, ultra-long take, dream-memory loops, poetic voice-over |
| 15 | Zhang Yimou 张艺谋 | Saturated single-color fields, mass choreography as image, folk-ritual iconography |
| 16 | Hou Hsiao-hsien 侯孝贤 | Static distant long take, doorway depth, real-time duration, ellipsis |
| 17 | Park Chan-wook 朴赞郁 | Baroque revenge, improbable camera geometry, ironic beauty over cruelty |
| 18 | Terrence Malick 泰伦斯·马力克 | Available light, wandering wide handheld at magic hour, whispered voice-over |
| 19 | Michael Mann 迈克尔·曼 | Urban night as a cold luminous field, professional competence, loneliness as architecture |
| 20 | Coen Brothers 科恩兄弟 | Deadpan geometric dark comedy, wide low-mounted moves, fate as a flat joke |
Each module carries a bilingual one-line lens, a 反例 / What this lens is NOT section naming the adjacent styles it gets confused with and the mechanical difference, a machine-readable 风格参数 YAML block (lens kit in mm, allowed and forbidden camera moves, motion budget, key ratio, palette, average shot length, sound policy, negative-prompt additions), three prompt templates, and a worked example. Every module directs the same control scene, so the twenty read as twenty answers to one question. Index and structure: references/director_styles/README.md.
See it in action. references/director_styles/example_comparisons.md renders one scene under several lenses back to back — the fastest way to feel what an overlay actually changes.
Tool adapters
The pipeline writes directorial intent; Step 9 reshapes it for the target model. v2 leads with capability-first routing so the file survives vendor churn: identify which control surfaces a tool exposes, then pick one of four prompt shapes — S1 motion-only, S2 full-description, S3 keyframe-pair, S4 multi-shot timestamped.
Video model families covered in references/ai-video-tool-adapters.md: Runway, Veo, Kling 可灵, Luma, Sora-class, Hailuo/MiniMax, Pika, Vidu, Midjourney Video, 即梦/Dreamina/Seedance, 万相 Wanxiang, and open-source/ComfyUI-class models.
Image model families covered in references/image-model-adapters.md: Midjourney, Flux, Nano Banana/Gemini-image class, Seedream/即梦 image, Qwen-Image, SDXL + ControlNet, Ideogram, Recraft — plus the character and location consistency playbooks that make keyframes hold across a scene.
Both files describe capability classes, not live feature lists. Vendors change limits constantly; verify against current docs.
Output modes
| Mode | Name | Use when |
|---|---|---|
| A | Director Analysis | Interpretation, subtext, direction rules |
| B | Beat Sheet | Structure before shots |
| C | Director's Book | The rules that govern the whole piece |
| D | Shot Plan | Shot list, storyboard plan, shooting table |
| E | Keyframe Prompt Pack | Stills, first/last frames, panels, character sheets |
| F | Video Motion Prompt Pack | Assets exist, motion prompts needed |
| G | Continuity Bible | Recurring characters, locations, multi-scene work |
| H | Sound & Dialogue Plan | Sound design, music, dialogue, voice-over |
| I | Edit & Assembly Plan | Cutting generated clips into a sequence |
| J | QC & Repair | Something failed or looks wrong |
Modes combine. A full production pass is A→B→C→D→E→F→H→I with G underneath and J at the end.
Repository structure
cinematic-director/
├── SKILL.md # Main pipeline: 13 steps, modes A-J, routing table
├── README.md # This file
├── CONTRIBUTING.md # How to add directors / adapters / genres / failure codes
├── CHANGELOG.md # Release history
├── LICENSE # MIT
├── references/
│ ├── cinematic-language.md # Shot grammar, lenses, focus, continuity geometry, coverage, format
│ ├── blocking-and-staging.md # Staging geometry, proxemics, notation, AI motion reliability
│ ├── lighting-and-color.md # Setups, ratios, practicals, time of day, palettes, grading
│ ├── sound-and-dialogue.md # Sound layers, dialogue for AI video, music brief, VO
│ ├── editing-and-assembly.md # Pacing math, cut motivation, transition engineering, timeline
│ ├── genre-playbooks.md # Per-genre directorial defaults
│ ├── product-and-macro.md # Product, packshot, cosmetics, food, macro
│ ├── prompt-lexicon.md # Verb banks, replacement table, negatives, EN/中文 terms
│ ├── failure-modes.md # Coded manual F1-F19, cost ladder, three-strike rule
│ ├── ai-video-tool-adapters.md # Video model surfaces, prompt shapes S1-S4
│ ├── image-model-adapters.md # Image model surfaces, character/location consistency
│ ├── continuity-bible.md # Character/location/prop/shot schemas + worked example
│ ├── production-workflow.md # Operating loop, difficulty rubric, versioning, handoff
│ └── director_styles/ # 20 style overlays + index + comparisons
├── assets/
│ ├── beat-sheet-template.md # Mode B
│ ├── director-book-template.md # Mode C
│ ├── shot-plan-template.md # Mode D
│ ├── keyframe-prompt-template.md # Mode E
│ ├── video-prompt-template.md # Mode F
│ ├── sound-plan-template.md # Mode H
│ ├── edit-timeline-template.md # Mode I
│ └── qc-checklist.md # Mode J
└── evals/
├── evals.json # Behavior contract
└── run.md # How to run the suite and score a result
Extending
- Add a director:
references/director_styles/NN_<slug>.mdusing the v2 section structure, with a filled风格参数block and the shared control scene; then register in four places. - Add a video or image tool: append an adapter block plus a matrix column. No
SKILL.mdchange needed unless the tool shifts a global default. - Add a genre: append a block to
references/genre-playbooks.mdusing the same fixed field set, plus a row in the comparison table. - Add a failure code: F-codes are append-only; add F20+ with symptom, ranked causes, cost-ordered fixes, and a before/after pair.
- Add an output mode: only if genuinely distinct from A–J.
- Add an eval: append a case to
evals/evals.json.
Full contributor guide: CONTRIBUTING.md.
Usage rule and disclaimer
The director-style modules describe high-level methods — narrative tendencies, camera grammar, lighting logic, editing rhythm, sound, performance, transferable checklists. They are not licensed to copy specific shots, dialogue, characters, plots, or any other copyrighted expression from any director's actual films. Use them as lenses that inform original work.
Tool adapters describe capability classes observed at the time of writing. AI video and image platforms change limits, parameters, and features constantly — always verify against the vendor's current documentation before relying on a specific behavior.
License
MIT — see LICENSE. That file is unmodified MIT text with no appended terms, so the license: MIT declaration in SKILL.md resolves to exactly what the file grants.
The MIT grant covers the skill files. The usage rule above is not a licence condition; it states what the style modules do and do not license. Copyright in any real film stays with its rights holders, and nothing here conveys an interest in it. That rule governs how you apply the directorial knowledge with respect to existing films, which is a separate concern from software licensing.
Cinematic Director Skill(中文版本)
一个 Claude Code / Claude Agent skill:把剧本、一段散文或一张关键帧,转化为完整的制作方案——节拍表、导演书、走位、含覆盖策略的分镜表、关键帧提示词、图生视频动作提示词、声音方案、剪辑时间线、连贯性圣经,以及带编码的 QC 修复流程。可叠加 20 位导演风格中的任意一位,作为贯穿全流程的统一镜头。
为什么需要它
大多数"电影感"AI 提示词只是形容词堆砌——cinematic、dramatic、masterpiece、8K。它们不告诉模型镜头在哪、身体在做什么、动作终点在哪、哪些锚点不能漂移。结果就是 PPT 感:每张静帧都好看,镜头之间却在游走。
本 skill 把形容词堆砌替换为导演式推理:
- 镜头是叙事决策。说不出"观众此刻比一秒前多懂了什么",这个镜头就删掉。
- 走位先于构图。摄影机位置取决于身体在空间中的动作;走位对了,平拍也能读懂。
- 关键帧才是控制点。视频模型基本上是在重新驱动静帧已经决定的一切。修静帧比重跑视频便宜得多。
- 连贯性是设计出来的。身份串、光位、画面方向、年代锁定写一次,然后逐字复用。
- 失败要诊断,不要重摇。19 个编码失败模式,每个都有机制、按概率排序的成因和成本最低的有效修复。
- 风格是镜头,不是外套。选库布里克意味着用广角单点透视把人压进建筑——不是"多加个推镜"。
2.0 有什么新东西
v1 掌握的是导演的词汇,v2 掌握的是词汇背后的手艺。
| v1.2 | v2.0 | |
|---|---|---|
| 流水线 | 10 步 | 13 步——新增需求梳理、声音与对白、剪辑与合成 |
| 输出模式 | 6 种 | 10 种——新增节拍表、导演书、声音方案、剪辑方案 |
| 参考文件 | 4 个 | 13 个 |
| 导演风格 | 14 位 | 20 位 |
| 模板 | 4 个 | 8 个,全部带已填好的实例 |
| 上下文策略 | 全量加载 | 路由表——深度内容移到路由表之后,SKILL.md 每次只加载一个请求需要的文件集,而不是全部 |
| 工具处理 | 6 个具名适配器 | 能力优先路由 + 4 种提示词形态 + 12 个适配器家族 |
| 失败处理 | 一张清单 | 编码手册 F1–F19、成本阶梯、三振规则 |
另外新增:焦距心理学、光比、连贯性几何(动作轴线、30 度规则、画面方向、视线匹配)以及"视频模型根本不知道上一镜机位在哪"这个 AI 专属难题的解法、调度几何库、类型片手册、受控提示词词库(含中英术语对照表),以及运营侧内容——镜头难度评分、重试预算、版本管理和交付包。
完整列表见 CHANGELOG.md。步骤重编号(10→13)与模式重编号(A–F→A–J)是破坏性变更。
架构
┌────────────────────────────────────────────────────────────────────────┐
│ SKILL.md — 始终在上下文里,负责路由分发 │
│ │
│ 路由表:需求类型 → 输出模式 → 该加载哪些文件 │
│ 硬规则 · 输入 schema · 默认值 · 自检清单 │
│ │
│ 1 需求梳理与范围 8 关键帧策略 │
│ 2 剧本与潜文本 9 工具适配器选择 │
│ 3 节拍图 10 AI 视频提示词构造 │
│ 4 导演风格镜头(可选) 11 声音与对白方案 │
│ 5 导演书 12 剪辑与合成方案 │
│ 6 走位与调度 13 QC 与修复 │
│ 7 分镜表与覆盖策略 │
└────────────────────────────────────────────────────────────────────────┘
│ 按需加载,绝不一次全读
┌───────────┴────────────────────────────────┐
▼ ▼
┌───────────────────────────────┐ ┌─────────────────────────────┐
│ references/ │ │ assets/ 可填写模板 │
│ 手艺 │ │ 节拍表 Mode B │
│ cinematic-language │ │ 导演书 Mode C │
│ blocking-and-staging │ │ 分镜表 Mode D │
│ lighting-and-color │ │ 关键帧提示词 Mode E │
│ sound-and-dialogue │ │ 视频提示词 Mode F │
│ editing-and-assembly │ │ 声音方案 Mode H │
│ genre-playbooks │ │ 剪辑时间线 Mode I │
│ product-and-macro │ │ QC 清单 Mode J │
│ 生成 │ └─────────────────────────────┘
│ prompt-lexicon │
│ ai-video-tool-adapters │
│ image-model-adapters │ ┌─────────────────────────────┐
│ failure-modes │ │ evals/evals.json │
│ 制作 │ │ 行为契约 │
│ continuity-bible Mode G │ └─────────────────────────────┘
│ production-workflow │
│ director_styles/ 20 位导演 │
└───────────────────────────────┘
一个 skill,可选风格叠加。 SKILL.md 始终是入口。用户点名导演时,Step 4 只加载一个风格文件,其 风格参数 YAML 块覆盖 Step 5 / 7 / 8 / 10 的默认值。风格文件是参考模块——它们不会被宿主 agent 单独注册为 skill,不会产生歧义。
安装
skill 文件放在 ~/.claude/skills/<skill-name>/(用户级)或 .claude/skills/<skill-name>/(项目级)。文件夹名必须与 SKILL.md 的 name: 一致(cinematic-director)。
git clone https://github.com/wuwangzhang1216/DirectorSKILL.git ~/.claude/skills/cinematic-director
git clone https://github.com/wuwangzhang1216/DirectorSKILL.git .claude/skills/cinematic-director
重新加载 Claude Code(或你的宿主 agent)即可触发。
更新
安装本身就是一个 git 检出,原地更新即可:
cd ~/.claude/skills/cinematic-director && git pull
查看当前跑的是哪个版本——已安装 SKILL.md frontmatter 里的 metadata.version:
grep -A2 '^metadata:' ~/.claude/skills/cinematic-director/SKILL.md
2.0.0 重命名并扩展了输出模式字母(A–F 变成 A–J)和流水线步骤(10 步变成 13 步)。旧检出只会用旧字母作答,于是本 README 写着的模式——H、I、J——看起来就像这个 skill 根本没有。出现这个症状就说明你还在 1.x:pull 一次,再重新加载宿主 agent。
快速开始
纯导演式流程
你:把这段做成 Runway 的 4 镜头 AI 视频方案,16:9。
"民国上海,一个信差在一户公寓门口犹豫,敲了一下门,又把信封从门缝塞进去,转身离开。"
输出:节拍图 → 含不变量条款的导演书 → 分镜表(叙事功能、景别、焦距、角度、镜头运动、起-动-止走位、光位、连贯性锚点、风险)→ Runway 形态的动作提示词 → QC 注记。
叠加导演风格
你:同一场戏,
Files in the repo
- .github
- assets
- evals
- references
- .gitignore
- .markdownlint.json
- CHANGELOG.md
- CONTRIBUTING.md
- LICENSE
- README.md
- SKILL.md
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More skills

Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.
Topic in, narrated explainer video out. A Claude Code / Codex skill that turns any topic into a black-canvas motion-graphics explainer video with TTS voiceover, subtitles and a chapter progress bar. Chinese or English; every frame drawn in code with Remotion.
Public repository for Agent Skills
Open-source AI job search: scan job portals, evaluate listings into a structured A-H report with a global 1-5 score, tailor your CV, track applications — runs locally in your AI coding CLI (Claude Code, Codex, OpenCode, Antigravity…)

Production-grade engineering skills for AI coding agents.