My prompt for ai coding with intent, harness and feedback
# For claude
git clone https://github.com/FFengIll/sdlc.git
# /sdlc:sdlc
cp -R sdlc ~/.claude/commands/
# or flatten /sdlc
cp -R sdlc/* ~/.claude/commands/
# For claude
claude plugin marketplace add ffengill/sdlc
claude plugin install sdlc@sdlc-marketplace
# or in claude
/plugin marketplace add ffengill/sdlc
/plugin install sdlc@sdlc-marketplace
# use
/sdlc your command
/sdlc "fix login bug" | "add user auth" | "review my changes"
│
▼
Intent Detection Engine
│
├─ fix|bug → bugfix workflow ─┐
├─ add|new → feature workflow │
├─ refactor → refactor workflow ├─→ Execute Skill → .sdlc/docs/category-feature-date.type.md
├─ review|cr → action:cr │
├─ understand → understand(cache) ─┘
└─ commit|pr → action:commit
/actions:<action>: understand | spec | coding | test | commit | cr | pr | regression
│
▼
Action Map
│
├─ understand → .sdlc/arch/overview-*.arch.md
├─ spec → .sdlc/docs/*.spec.md
├─ coding → .sdlc/docs/*.coding.md
├─ test → .sdlc/docs/*.test.md
├─ commit → .sdlc/docs/*.commit.md
└─ pr → .sdlc/docs/*.pr.md
/sdlc add user authentication
│
▼
Feature Workflow:
understand → research → spec → coding → test → commit → pr
│ │ │ │ │ │ │
.sdlc/ .sdlc/ .sdlc/ .sdlc/ .sdlc/ .sdlc/ .sdlc/
arch/ docs/ docs/ docs/ docs/ docs/ docs/
overview- auth-* auth-* auth-* auth-* auth-* auth-*
*.arch.md .research .spec .coding .test .commit .pr
.md .md .md .md .md .md
Interactive Flow: 继续 / 下一步 → Next phase 跳过测试 → Skip phase 到哪了？ → Check status
-
Specifications matter.
-
Trust in the power of process.
The Software Development Life Cycle (SDLC) may be classical, but it still has much to teach us.
-
Do not attempt to keep every specification constantly updated as the code evolves.
High-level specifications will inevitably lag behind.
-
Validation specifications are critical.
We refer to this layer as the
Harness. -
Embrace the practice of maintaining and sharing the
Harness:- It strengthens code reviews
- It accelerates debugging
- It supports safe refactoring
- ……
-
REMEMBER There is no silver bullet.
The only constant is change.
-
规范（SPEC）很重要。
-
相信流程的力量。
Software Development Life Cycle (SDLC) 虽然古典，甚至略显传统，但它能教会我们的，远不止流程本身。
-
不要试图让所有 SPEC 随代码实时同步。
一般性的 SPEC，天然会滞后，这是常态，而非问题。
-
验证型 SPEC 更关键。
我们可以称之为
Harness。 -
主动建设并沉淀
Harness：- 让 Code Review 更有依据
- 让 Debug 更高效
- 让 Refactoring 更安全
- ……
-
不存在银弹。
变化常在。
-
反馈驱动演进。
-
工作流结束，学习才刚开始。
每一次通过
/sdlc完成的任务，都会产生新的洞察。 -
不要满足于静态流程。
今天有效的方法，明天可能需要改进。
-
打分、反思、迭代。
-
持续改进：
- 评估哪些有效，哪些无效
- 基于真实经验更新工作流
- 让每次迭代为下次提供依据
- ……
-
不存在银弹。
系统在每一次反馈循环中自我完善，人的反馈非常重要且简单。
-
Feedback drives evolution.
-
The workflow completes, but the learning continues.
Every task executed through
/sdlcgenerates insights. -
Do not settle for static processes.
What works today may need refinement tomorrow.
-
Score, reflect, and iterate.
# tasks done with /sdlc
# now we score them and update them
/feedback
-
Embrace continuous improvement:
- Evaluate what worked and what didn't
- Update workflows based on real experience
- Let each iteration inform the next
- ……
-
REMEMBER There is no silver bullet.
The system improves itself, but feedback loop is important at a time.
- Inspiration from https://github.com/karpathy/autoresearch
- Implemented as
/feedback
- A markdown writing plugin (support *.md) for coding task spec writing - https://github.com/FFengIll/vscode-tingly-spec.git
- AI Orchestrator - https://github.com/tingly-dev/tingly-box