By hujianbest
Orchestrate a structured, gate-based product development workflow from idea to release, with formal spec, design, task, TDD implementation, multi-phase review, regression, and closeout stages enforced by automated routing and evidence verification.
Implement the current active HarnessFlow task with TDD and fresh evidence (hf-test-driven-dev). Only valid when exactly one Current Active Task is locked.
Route-first entry. When you are not sure which HarnessFlow node should run next, use this. The router will recover the right next node from on-disk artifacts.
Bias toward the HarnessFlow planning stage — design (and UI design when the spec declares a UI surface) plus task breakdown. The router decides whether to enter hf-design, hf-ui-design, or hf-tasks based on artifact evidence.
Cut a vX.Y.Z engineer-level release. Aggregates multiple workflow-closeout features into a release scope ADR, runs release-wide regression, syncs CHANGELOG / release notes / ADR statuses, and produces a tag-ready pack. Decoupled from the HF workflow router.
Request a HarnessFlow review. The router dispatches hf-reviewer to the correct hf-*-review skill (discovery / spec / design / ui / tasks / test / code / traceability) based on artifact context.
`hf-implementer` is the implementation subagent used by HarnessFlow when `hf-subagent-driven-dev` delegates one locked `Current Active Task` to a fresh context.
`hf-reviewer` is the independent review subagent used by HarnessFlow for every canonical review node, not only reviews after TDD implementation.
Use when hf-test-driven-dev finishes GREEN on a frontend-touching active task whose spec declares a UI surface, and fresh browser runtime evidence (screenshot / console log / network trace) is required for downstream gates. Not for issuing verdicts (gates do that), not for replacing hf-test-review's test-quality review, not for backend-only tasks.
适用于 test review 通过后评审代码质量、用户要求 code review 的场景。不适用于评审测试(→ hf-test-review)、写/修代码(→ hf-test-driven-dev)、阶段不清(→ hf-workflow-router)。
适用于 regression gate 之后需判断任务完成走向、确认任务可宣告完成、用户要求\"能不能算完成\"的场景。不适用于缺回归记录/实现交接块需补齐(→ hf-regression-gate)、需状态收尾(→ hf-finalize)、阶段不清(→ hf-workflow-router)。
Use when a host project (vendoring HF) wants per-directory hierarchical context (project root / mid-directory / leaf-directory) auto-generated for any AI agent reading the codebase. Generates `AGENTS.md` (OpenCode), `.cursor/rules/*.mdc` (Cursor), or `CLAUDE.md` (Claude Code) skeletons; architect fills in conventions. Not for HF's own `docs/principles/` (untouched); not for spec / design / tasks artifacts (use upstream skills).
适用于设计草稿已完成需要正式 review verdict、或被指定为 reviewer subagent 执行设计评审的场景。不适用于需继续写或修设计(→ hf-design)、需拆任务或编码(→ hf-tasks)、阶段不清或证据冲突(→ hf-workflow-router)。
Uses power tools
Uses Bash, Write, or Edit tools
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.
Spec-anchored SDD, gated TDD, and evidence-based workflow control for AI coding agents.
HarnessFlow packages the practices behind disciplined AI-assisted engineering into self-contained Markdown skills: discovery, specs, design, tasks, TDD implementation, independent reviews, gates, and closeout.

Claude Code gets 7 slash commands. OpenCode and Cursor use the same intents through natural language plus using-hf-workflow.
| What you're doing | Command | Key principle |
|---|---|---|
| Enter or resume HF | /hf | Route from artifacts |
| Define what to build | /spec | Spec before code |
| Plan how to build it | /plan | Design then tasks |
| Build one task | /build | One active task |
| Review before moving on | /review | Author/reviewer separation |
| Close engineering work | /ship | Gates before closeout |
| Cut a release pack | /release | Version docs, not deploys |
Every command is a bias, not a bypass. The router still checks repository evidence before choosing the next node, except /release, which directly invokes the standalone release skill.
Install from the marketplace:
/plugin marketplace add https://github.com/hujianbest/harness-flow.git
/plugin install harness-flow@hujianbest-harness-flow
Vendor HarnessFlow into your project with the install script:
git clone https://github.com/hujianbest/harness-flow.git /path/to/harness-flow
# OpenCode
bash /path/to/harness-flow/install.sh --target opencode --host /path/to/your/project
# Cursor
bash /path/to/harness-flow/install.sh --target cursor --host /path/to/your/project
# Both
bash /path/to/harness-flow/install.sh --target both --host /path/to/your/project
The script copies or symlinks skills/, agents/, OpenCode command assets, and Cursor agent mirrors, places client-specific rules, and writes .harnessflow-install-manifest.json so uninstall can remove only HF-managed files.
Use HarnessFlow from this repo. Start with `using-hf-workflow` and route me through the correct HF workflow.
I want to add rate limiting to our notifications API.
Do not jump straight to code.
More setup detail:
You: Use HarnessFlow from this repo. Start with `using-hf-workflow`.
I want to add rate limiting to our notifications API.
HF: Reads repository artifacts, chooses the right entry point, and routes
into discovery or `hf-specify` instead of jumping straight to code.
You: Use HarnessFlow to plan the approved spec.
HF: Produces architecture with `hf-design`, adds `hf-ui-design` only when
the spec declares a UI surface, then breaks the design into reviewed
tasks with `hf-tasks`.
You: Use HarnessFlow to build the current active task.
HF: Locks one `Current Active Task`, writes the test design, records the
approval, runs RED -> GREEN -> REFACTOR, updates `tasks.progress.json`,
and writes a wisdom-notebook delta for cross-task learning.
You: Use HarnessFlow to verify and close this work.
HF: Runs test, code, and traceability reviews. If the task touches a
frontend surface, `hf-browser-testing` captures DOM / console / network
evidence before the gates decide what happens next.
You: Use HarnessFlow to ship the completed workflow.
HF: Runs regression, doc-freshness, and completion gates, then `hf-finalize`
writes `closeout.md` plus the closeout HTML companion. For a vX.Y.Z
release, `hf-release` can aggregate closed workflows into a tag-ready
release pack; it still does not deploy to production.
If the architect explicitly asks for auto mode, hf-ultrawork can continue through canonical next actions, write approval records, and keep a fast-lane audit trail while preserving Fagan reviews, gate verdicts, closeout, and hard stops.
HarnessFlow currently ships 29 hf-* skills plus the using-hf-workflow entry skill.
| Skill | What it does | Use when |
|---|---|---|
| using-hf-workflow | Public entry shell for choosing direct invoke vs router handoff | Starting a session or expressing a high-level HF intent |
| hf-workflow-router | Evidence-based runtime router and recovery controller | Continuing from repository artifacts or consuming review/gate outcomes |
npx claudepluginhub hujianbest/harness-flow --plugin harness-flowLong-running agent harness with 5-layer memory architecture, GitHub integration, autonomous batch processing, Agent Teams with ATDD, 9 hooks (safety, quality gates, team coordination), and 6 Agent Skills
Persona-driven AI development team: orchestrator, team agents, review agents, skills, slash commands, and advisory hooks for Claude Code
Hypo-Workflow for Claude Code. The plugin namespace is intentionally `hw`; plugin-root commands map /hw:* to existing workflow Skills.
Helder's personal SDLC toolbelt for AI coding agents — from PRD to ship. Bundles the tracer-bullet workflow alongside TDD, code review, audits, and shipping skills.
Harness engineering for Claude Code — hook-enforced dual review, state-machine gates, and fail-closed safety where it counts.
Unified toolkit for Context-Driven Development with spec-first planning, TDD workflow, and Beads integration