AI Skill Hub 强烈推荐:Plumbline 是一款优质的Claude技能。AI 综合评分 8.0 分,在同类工具中表现稳健。如果你正在寻找可靠的Claude技能解决方案,这是一个值得深入了解的选择。
Plumbline 是一款基于 Shell 开发的开源工具,专注于 claude_skill、agent-framework、agentic-ai 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
Plumbline 是一款基于 Shell 开发的开源工具,专注于 claude_skill、agent-framework、agentic-ai 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
# 克隆仓库 git clone https://github.com/DYAI2025/Plumbline cd Plumbline # 查看安装说明 cat README.md # 按 README 完成环境依赖安装后即可使用
# 查看帮助 plumbline --help # 基本运行 plumbline [options] <input> # 详细使用说明请查阅文档 # https://github.com/DYAI2025/Plumbline
# plumbline 配置说明 # 查看配置选项 plumbline --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export PLUMBLINE_CONFIG="/path/to/config.yml"
| Feature | What it means | Honest status |
|---|---|---|
| **True-Line Governance** | Product Vision and Product Canvas alignment are treated as first-class gates. Requirements must stay linked to the confirmed user problem, value promise, success signal, non-goals, and risks. | Partly machine-backed, partly model-judgment. Human confirmation remains final. |
| **Runtime Integrity / Reality Layer** | PRIL-style checks for reality evidence, context artifacts, scope containment, run ledgers, redaction, and fail-closed hook behavior. | The narrow PRIL core has executable tests and CLIs; not every governance rule is mechanically enforced. |
| **Plumbline Watcher** | Independent value gate that can return pass, review-required, pause, or blocked. It may pause when tests are green but value remains unproven or contradicted. | Watcher judgment is model-mediated. Benchmarks show specific real cases, not universal correctness. |
| **Multi-model Council** | /concilium can stress-test ideas through market, technical, skeptical, distribution, and character-based lenses. OpenRouter-backed foreign-model paths exist for real boundary crossings. | Capability smokes exist. Council quality lift is not proven; underpowered runs are reported as underpowered. |
| **Reality Ledger** | Requirements and claims carry evidence classes such as fake/mock, integration, real-boundary-smoke, or stronger observed evidence. Fake-only evidence cannot silently become “done.” | Useful for preventing evidence laundering; only as strong as the recorded artifacts and invoked gates. |
| **Continuous Learning** | Retrospectives can become persistent rules only through guarded persistence, diff preview, tests, and explicit approval. | Designed to reduce repeated process errors without blind self-modification. |
| **Honest update/install boundary** | Plumbline installs through scripts and local Claude Code config. It is not a Claude Code plugin and ships no MCP server. | See [DEPENDENCIES.md](DEPENDENCIES.md). |
---
| Area | Purpose |
|---|---|
core/ | Base roles such as coder, planner, researcher, reviewer, tester. |
agileteam/ | Product-governed workflow roles: requirements, Product Owner, Watcher, security, spec audit, context. |
concilium/ | Council bodies, reports, and character-based critique structures. |
config/claude/commands/ | Slash commands: /agileteam, /concilium, /honest-status, /bench-oracle, /merge-when-true, /openrouter-live-smoke, /persist-learning, and reflection/update commands. |
config/claude/lib/ | Runtime integrity, council, update, scope, context, redaction, and run-ledger libraries. |
config/claude/tests/ | Shell/Python regression checks for commands, gates, update layer, council paths, GUI, and docs honesty. |
config/claude/skills/ | 16 vendored skills for portable workflows. |
docs/benchmarks/ | Captured benchmark and live-smoke write-ups with evidence ceilings. |
docs/canvas/ | Product Canvas artifacts created by the mandatory Product Canvas gate. |
docs/reality/ | Reality-ledger evidence JSONL files. |
docs/templates/ | Product Canvas, traceability, contradiction, and governance templates used by the workflow. |
metrics/ | Corpus, runs, measurement tools, and experiment artifacts. |
explorer/ | Source for the visual Agent Explorer. |
Browse the full agent set in the Agent Explorer.
---
A team starts a larger software initiative with good discipline: Product Vision, PRD, traceability, TDD, implementation, review, security, validation. The tests are green.
But the increment has a hidden semantic failure: it satisfies the written test while violating the confirmed product meaning. For example, the Product Canvas says tracking must not become surveillance, but the implemented “analytics improvement” quietly ships behavior the Canvas named a non-goal.
In a normal green-test pipeline, the work may proceed because the local evidence says “pass.” In Plumbline, the Watcher can stop the process anyway:
Tests: green
Implementation: present
Local acceptance criteria: satisfied
True-Line check: contradiction with confirmed Canvas non-goal
Watcher verdict: pause
Required decision: user must confirm reframe, remove the conflicting requirement, or change implementation
The point is not that tests are unimportant. The point is that tests are not gravity. The real solved problem is gravity. If the code does not solve the practical problem, the work does not hang true — even if every test is green.
---
Requirements: Claude Code, git, bash, python3, and jq.
git clone https://github.com/DYAI2025/Plumbline plumbline
cd plumbline
./config/claude/install.sh
Then, inside a Claude Code project:
/agileteam "add OAuth2 login with refresh-token rotation"
Useful commands:
plumbline doctor
plumbline update --check
plumbline update
bash config/claude/tests/run_all.sh
Installation boundary:
install.sh / plumbline install style local setup.--with-flow-agents.See SETUP.md and DEPENDENCIES.md.
---
高质量的自动化工作流程工具
该工具使用 NOASSERTION 协议,商用场景请仔细阅读协议条款,必要时咨询法律意见。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
📄 NOASSERTION — 请查阅原始协议条款了解具体使用限制。
总体来看,Plumbline 是一款质量优秀的Claude技能,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | Plumbline |
| 原始描述 | 开源Claude技能:Plumbline — a self-learning, customer-value-governed agile AI agent team for Cla。⭐6 · Shell |
| Topics | claude_skillagent-frameworkagentic-aiagentic-workflowagile-aiai-agentsshell |
| GitHub | https://github.com/DYAI2025/Plumbline |
| License | NOASSERTION |
| 语言 | Shell |
收录时间:2026-06-19 · 更新时间:2026-06-19 · License:NOASSERTION · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端