OutlineDriven/outline-driven-development
Outline Driven Development – Beyond "specs" or "vibes" using AST analysis and Modern CLI Tools w/ context engineering; batteries-included
Deep Analysis
一种大纲驱动的LLM代码开发范式,通过确定性脚手架约束非确定性的AI创造力,在随意开发与过度规范之间取得平衡
Core Features
Technical Implementation
- 设计优先方法论:先定义架构再实现
- 可观测追踪ID用于故障归因,质量门禁机制确保输出质量
- 需要控制AI代码生成一致性的大型项目开发
- 多Agent协作的复杂代码工程任务
- 需要安装多个Rust CLI工具,环境配置门槛较高
Outline Driven Development: The whole new paradigm for the augumented LLM Code Agent workflows.
Vibes are too shallow, Specs are too complex.
Let there be the outline.
And Indeed... Here it is!
Prerequisites
ast-grep | ripgrep | fd | eza | lsd | tokei | bat | just | git-branchless | difftastic | procs | fend | hck | hyperfine | +other tools | MCPs (context7, sequentialthinking-tools, actor-critic-thinking, shannon-thinking, repomix)
Install Various Rust-based CLI Tools with highest native optimizations through cargo
Install Cargo (if not installed)
Linux/macOS with cargo
Install
export RUSTFLAGS="-C target-cpu=native -C link-arg=-fuse-ld=mold -C opt-level=3 -C strip=symbols -C panic=abort -C lto=thin"
cargo install --locked cargo-binstall
cargo install ast-grep ripgrep fd-find eza lsd
cargo binstall -y bat tokei git-delta just raff-cli difftastic git-branchless zoxide procs bfs fselect tealdeer srgn nomino shellharden grex mergiraf jaq jql hck huniq lemmeknow hyperfine rargs eva fend rip2 sccache
Windows with cargo
Install (run inside powershell)
$env:RUSTFLAGS="-C target-cpu=native -C opt-level=3 -C strip=symbols -C panic=abort -C lto=thin -C link-arg=/LTCG -C link-arg=/OPT:REF"
cargo install --locked cargo-binstall
cargo install ast-grep ripgrep fd-find eza lsd
cargo binstall -y bat tokei git-delta just raff-cli difftastic git-branchless zoxide procs bfs fselect tealdeer srgn nomino shellharden grex mergiraf jaq jql hck huniq lemmeknow hyperfine rargs eva fend rip2 sccache
Installation
- Gemini CLI: https://github.com/OutlineDriven/odin-gemini-cli-extension
- Quick Install:
gemini extensions install https://github.com/OutlineDriven/odin-gemini-cli-extension
- Claude Code: https://github.com/OutlineDriven/odin-claude-plugin
- Quick Install:
git clone https://github.com/OutlineDriven/odin-claude-plugin.git && cp -r ./odin-claude-plugin/ ~/.claude/ && claude plugin marketplace add OutlineDriven/odin-claude-plugin && claude plugin install odin@odin-marketplace
- Quick Install:
- Codex CLI: https://github.com/OutlineDriven/odin-codex-plugin
- Quick Install:
git clone https://github.com/OutlineDriven/odin-codex-plugin.git && cp -r ./odin-codex-plugin/ ~/.codex/
- Quick Install:
- Prompt-only (Manual Apply for IDEs/Agents like Antigravity/Windsurf/Cursor/Augument Code ...): Prompt
- Compact Prompt-only (Manual Apply for IDEs/Agents like Antigravity/Windsurf/Cursor/Augument Code ...): Prompt
Recommended MCP Extensions
Crucial [Automatically installed]
ast-grep | context7 | sequentialthinking-tools | actor-critic-thinking | shannon-thinking
Additional [Manually install if needed]
Time, Tavily, Exa, Ref-tools
Outline-Driven Development Philosophy
Deterministic scaffolds harness non-deterministic LLM creativity only when the outline stays the single source of truth and every downstream stage revalidates against it.
Deterministic-with-Non-Deterministic-LLMs-by-Nature
- Outline-as-assembly: Human intent, compliance constraints, and architectural guardrails are compiled into a versioned outline whose hash becomes the control envelope for every agentic act.
- LLM-as-module: LLM calls are intentionally non-deterministic but bounded by the outline contract; disagreement between generated code and outline immediately halts or replays the step.
- Telemetry feedback: Execution traces, test verdicts, and rubric scores continuously feed back into the outline refinement loop to converge to a reproducible build.
Control Envelope Checklist
- Canonical outline stored with content-addressable ID and time windowed approvals.
- Every agent invocation receives a minimized delta (outline slice) plus explicit success metrics.
- Determinism is measured: diff noise ≤ 2% between successive outline-conformant runs; higher variance triggers outline tightening before reattempting generation.
Design-First / Best-Practices Batteries Included
- Architecture-first: Each outline must include interfaces, pre/postconditions, error domains, latency and memory budgets before code exists.
- Tooling-first:
eza,ast-grep,ripgrep,fd, LangGraph, and MCP stacks are treated as mandatory batteries so that structural edits, search, and orchestration are reproducible. - Quality gates: Spec → outline → implementation is instrumented with lint/test/benchmark gates plus rollback hooks.
- Observability: Outline nodes ship tracing IDs and contract assertions so failures are attributable to outline leaves rather than opaque LLM conversations.
Traceability Matrix
| Diagram | Goal | Associated Invariants |
|---|---|---|
| Architecture | Map outline-to-toolchain interfaces | Outline hash monotonicity |
| Data-flow | Guarantee data never bypasses verification | Contract-bound IO only |
| Concurrency | Preserve happens-before relationships | No circular waits; deadlock-free |
| Memory | Ensure ownership + persistence model | Leak-free caches, append-only audit |
| Optimization | Tie budgets to feedback loops | Regression triggers deterministic rollback |
Related Skills
wshobson/agents
wshobsonIntelligent automation and multi-agent orchestration for Claude Code
The most comprehensive Claude Code plugin ecosystem, covering full-stack development scenarios with a three-tier model strategy balancing performance and cost.
ComposioHQ/awesome-claude-skills
ComposioHQA curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows
The most comprehensive Claude Skills resource list; connect-apps is a killer feature.
code-yeongyu/oh-my-opencode
code-yeongyuThe Best Agent Harness. Meet Sisyphus: The Batteries-Included Agent that codes like you.
Powerful multi-agent coding tool, but note OAuth limitations.
thedotmack/claude-mem
thedotmackA Claude Code plugin that automatically captures everything Claude does during your coding sessions, compresses it with AI (using Claude's agent-sdk), and injects relevant context back into future sessions.
A practical solution for Claude's memory issues.
OthmanAdi/planning-with-files
OthmanAdiClaude Code skill implementing Manus-style persistent markdown planning — the workflow pattern behind the $2B acquisition.
Context engineering best practices; an open-source implementation of Manus mode.
yusufkaraaslan/Skill_Seekers
yusufkaraaslanConvert documentation websites, GitHub repositories, and PDFs into Claude AI skills with automatic conflict detection
An automation powerhouse for skill creation, dramatically improving efficiency.

