By wanggang316
Agent-first development framework: lifecycle skills (harness-stack:*), review/security/test subagents, and harness hooks.
Answer-only mode — research and explain, never edit code or suggest changes were made
Create an atomic git commit with conventional-commit format and no attribution trailers
复盘本次 harness-stack 使用,并把值得上报的摩擦/缺陷/建议提成上游 Issue
Pull --rebase from upstream, resolve conflicts, then push
Scan a PR; if there's anything to act on, hand it to harness-stack:review-receive
资深 code reviewer,从五个维度——correctness、readability、architecture、security、performance——对照 spec 评审一段 diff。在合并任何非平凡改动前、在某个 fdd 的 feature 或里程碑完成后、或任何需要在全新上下文窗口里评估代码质量时使用。
在多 agent 工作流中被派发的、以 feature 为边界的 implementer。开工前若有任何不明确就先发问,feature 要求时遵循 TDD,跑一遍结构化自检,报告四个 status code 之一,并通过 fdd 写出一份结构化的 handoff JSON。当 controller 把一个单一、有边界的 FDD feature 连同完整文本与上下文交给你时使用。
只读的调查子代理,由 controller 派发。用于 FDD 规划阶段的代码库调查、编写 validation contract 时的 feature 区域枚举与对抗式契约审查、流程中途的范围变更分析、以及 feature 失败的根因分析。返回提炼后的洞察——从不修改文件。当 prompt 要求时可做在线调研(WebSearch / WebFetch)。
fdd-validate 的静态验证(stage 1)subagent。对 milestone / 全量 diff 跑硬门禁(test/typecheck/lint,只看相对 baseline 的新增失败)+ 逐 feature scrutiny 审查,把低风险事实性更新直接应用到 docs/ Library、产出治理建议、并把 synthesis 写到 .harness-runtime/plans/<slug>/validation/<scope>/scrutiny/synthesis.json。由 fdd-validate 在里程碑边界、收尾、以及每轮修复后派发。
安全评审专家,审计一段 diff 中的漏洞——OWASP Top 10、密钥、输入校验、认证 / 授权、依赖 CVE,以及 LLM 信任边界泄漏。当 diff 触及认证、用户输入、密钥、加密、裸查询、shell/eval、依赖升级,或 LLM 输出流入受信上下文时使用。在 fdd 中由 milestone 闸条件触发,或由 security 技能手动派发。
创建并维护 CHANGELOG.md。在初始化变更日志、从 git 历史中提取未发布变更、或准备发布某个版本时使用。
让多个异构 LLM agent 就同一个问题展开一场 multi-agent debate。每个 round 都做匿名处理,使参与者只就论据本身较量,而不在意来源。最终产出是一份综合后的答案外加一份 claim catalog。当一个问题含糊、有争议、或风险高到单个模型的第一反应不足以采信时使用。
引导系统化的 root cause debug。当测试失败、构建中断、行为与预期不符,或遇到任何意外错误时使用。当你需要用系统化的方法定位并修复 root cause、而不是靠猜时使用。
在单个问题上跑一轮并行多 agent 决策。每个 agent 独立作答,给出结构化的 recommendation;再由一道 synthesis 把它们的 recommendation 合并成最终决策,附带 confidence 以及浮现出来的 minority position。当你想为一次性决策拿到稳健答案、并显式追踪异议,但又不需要多轮 debate 的来回拉锯时使用。
在项目层面定义 API spec。在启动带 API 的项目、api-spec.md 缺失或过时、或跨服务出现 API 不一致时使用。产出 docs/api-spec.md,作为权威的 API contract。
Uses power tools
Uses Bash, Write, or Edit tools
Runs pre-commands
Contains inline bash commands via ! syntax
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
An agent-first development harness — a curated stack of tools that keep an AI coding agent building stably, durably, and at high quality.
Philosophy: Humans provide direction, agents execute.
harness-stack is the harness around the agent: a collection of Skills, Commands, and Subagents that an AI coding agent reaches for while doing real engineering work. Each tool encodes one process a senior engineer would follow — defining specs, designing, debugging, reviewing, shipping — so the work comes out consistent and verifiable instead of improvised.
Everything is addressed through the harness-stack: plugin namespace, and everything lives in the repo: an agent that can't read it doesn't know it. There is no out-of-band CLI — every operation is a Skill or a Subagent.
Feature-Driven Development (/harness-stack:fdd) is the heart of the stack. It is a contract-first, multi-agent loop that turns a goal into shipped, verified code. Three roles, each running in a fresh context so judgment never blurs:
fdd orchestrates the flow; fdd-execution drives the serial build loop, dispatching one feature at a time and gating each handoff.implementer subagent builds exactly one bounded feature (test-first when required) and reports a structured handoff.fdd-validate) that probes the work against the contract: static gate (scrutiny-validator: test / lint / type-check + scrutiny) → review (code-reviewer: five axes, plus security-auditor when warranted) → behavioral user-test (user-test-validator).The glue is the validation contract (fdd-validation-contract): the definition of done written as testable, user-observable assertions before code is written. The coordinator/implementer/validator triad checks every milestone against that contract — that's what keeps the program continuously converging on the goal at high quality rather than drifting.
| Tool | Role in FDD |
|---|---|
/harness-stack:fdd | Orchestrator — runs the 3-step main flow (the only one you invoke directly) |
/harness-stack:fdd-planning | Step 1 — capture the plan and break it into features |
/harness-stack:fdd-validation-contract | Step 1 — write the definition-of-done assertions |
/harness-stack:fdd-execution | Step 2 — the serial per-feature build loop |
/harness-stack:fdd-validate | Step 3 — milestone & final gates (static → review → user-test) |
/harness-stack:tdd | Test-first development, used inside an implementer's task |
Use FDD for any non-trivial change — one that touches multiple files, has several acceptance criteria, or spans more than one feature.
Individual tools, each doing one job. Mix and match — only FDD is a coordinated group.
| Tool | What it does |
|---|---|
/harness-stack:define-product | Define the product — what it is and why (product-spec.md) |
/harness-stack:define-architecture | Define the system's structural map (architecture.md) |
/harness-stack:define-api-spec | Define the authoritative API contract (api-spec.md) |
/harness-stack:define-frontend-spec | Define frontend engineering conventions & quality bars |
/harness-stack:define-ui-spec | Define the UI design system (DESIGN.md) |
/harness-stack:design | Write a standalone technical design doc before building |
/harness-stack:debug | Systematic root-cause debugging and recovery |
/harness-stack:review-request | Dispatch fresh-context reviewers (code / security / tests) |
/harness-stack:review-receive | Handle reviewer feedback with rigor, not theater |
/harness-stack:security | Security audit and hardening |
/harness-stack:debate | Multi-round debate among heterogeneous LLM agents |
/harness-stack:decide | One-shot parallel decision support with synthesis |
/harness-stack:changelog | Create and maintain CHANGELOG.md |
/harness-stack:git | Disciplined git workflow and versioning |
/harness-stack:pr | Open a PR and drive it to a clean merge |
/harness-stack:ship | Pre-launch checklist, monitoring, rollback plan |
/harness-stack:docs-init | One-time scaffold of a project's docs structure |
/harness-stack:env-init | Per-worktree isolated runtime environment |
/harness-stack:skill-create | Create a new harness-stack skill |
/harness-stack:feedback | Reflect on a session and file an upstream issue |
Thin slash-command entry points that delegate into the skills above — the substance lives in the skill.
npx claudepluginhub wanggang316/harness-stack --plugin harness-stackComprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Feature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Complete developer toolkit for Claude Code