Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add yike-gunshi/forge-skills --skill forge-devgit clone --depth 1 https://github.com/yike-gunshi/forge-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/yike-gunshi/forge-skills/forge-dev)<a href="https://agentmods.dev/skills/yike-gunshi/forge-skills/forge-dev"><img src="https://agentmods.dev/badge/skills/yike-gunshi/forge-skills/forge-dev/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/yike-gunshi/forge-skills/forge-dev"><img src="https://agentmods.dev/badge/skills/yike-gunshi/forge-skills/forge-dev.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00122 | $0.04273 |
| Opus 5 | $0.00061 | $0.02136 |
| Sonnet 5 | $0.00024 | $0.00855 |
| Haiku 4.5 | $0.00012 | $0.00427 |
Grade A, and why
forge-dev scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 327 lines — stays where its author put it; the contents beside it link to each section on GitHub.
文档落地路径:遵循 forge-doc-policy 规范。完整白名单 + frontmatter schema 见
~/.claude/skills/forge-doc-policy/doc-paths.md。 当前文档加载顺序:先读项目CLAUDE.md、docs/README.md、docs/INDEX.md, 再读相关根级当前真相源和.features/{feature-id}/feature-spec.md。 详细规则见~/.claude/skills/_shared/current-doc-loading.md。
/forge-dev:开发调度器
接力 /forge-prd 或 /forge-brainstorm 的产出,调度设计、工程、QA 子技能完成开发交付。
全程中文。
前置脚本(每次先运行)
检测:当前分支与项目根目录、项目环境(Node / Python / Go / Rust / Makefile)、Playwright 可用性、
.forge/dev-state.json 未完成流水线、.features/_registry.md 注册表、brainstorm 思考文档。
脚本全文见 references/orchestration-details.md「前置检测脚本」。
流程
读取 PRD 迭代摘要
→ 项目类型判断(frontend / backend / fullstack)
→ Discussion 阶段(结构化偏好收集,识别灰区)
→ Research 阶段(技术调研,基于知识+代码扫描+按需搜索)
→ 产出 RESEARCH.md
→ 分析变更 + 调研结果,判断需要哪些子 skill
→ 列出建议的执行计划,用户确认
→ Wave 并行调度子 skill(每个子 skill 在独立上下文执行)
→ 汇总交付结果
全程中文。
上下文工程(核心设计)
问题:随着会话上下文窗口被填满,AI 输出质量会逐步劣化(context rot)。
解决方案:主调度器只做编排,真正的工作发生在子 agent 的独立上下文中。
原则
- 主上下文只做调度 — 读取 PRD、判断计划、协调子 agent,不做具体实现
- 子 skill 在独立上下文执行 — 使用 Agent 工具启动子代理,每个子 skill 拥有全新的上下文窗口
- 精准传递上下文 — 只传递子 skill 需要的文档路径和指令,不传递无关信息
- 结果汇总回主上下文 — 子 agent 返回执行摘要,主调度器汇总为交付报告
子 skill 调用方式
使用 Agent 工具,prompt 中包含:
1. 子 skill 的完整指令(从 SKILL.md 读取)
2. 项目路径和 PRD 路径
3. 本次变更的具体内容(从 PRD 迭代摘要提取)
4. 前序子 skill 的产出路径(如 DESIGN.md)
5. RESEARCH.md 路径(如有)
关键约束:
- 每个 Agent 调用只执行一个子 skill
- 模型路由(Superpowers v5 实践):机械型子任务(跑测试、批量重命名、模板填充)给 Agent 传
model: sonnet用便宜模型;判断型子任务(架构、设计、审查)保持默认模型,不降级 - 不要在主上下文中重复子 skill 的工作
- 子 agent 完成后,读取其产出文档确认结果,而非依赖其返回的文本
- 传递 Feature Spec:如果
.features/{feature-id}/feature-spec.md存在,SHALL 将其路径和关键内容(行为场景 + 验收检查表)传递给 forge-eng 和 forge-qa 子 agent - forge-eng 行为约束:告知 forge-eng 子 agent 每完成一个功能点后,SHALL 自验对应的 Given/When/Then 场景,确认 PASS 后再继续下一个功能点
模式说明
交互模式(默认)
在 3 个硬卡点暂停等待用户确认:
- Discussion 完成后 — 确认偏好收集是否完整
- Research + 执行计划确认 — 调研结果 + 建议的执行计划
- 所有子技能完成后 — 交付总结确认
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 327 lines · 122 tokens per session scan A a5bf56e36fc5
forge-dev is a skill published in the GitHub repository yike-gunshi/forge-skills (13 stars, last pushed 1mo ago), licensed MIT. It adds 122 tokens to every session and 4,273 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
vastai
Vast.ai CLI to manage GPU instances, volumes, serverless endpoints, and billing.
vastai-sdk
Vast.ai Python SDK — high-level API for GPU instances, volumes, serverless endpoints, and billing.
forge-evolution
Autonomous persona evolution pipeline for failure-driven capability growth.
docs-context
Super Base Context — project doc loader & synchronizer. Triggers even without explicit "load docs"/"sync docs." READ MODE — load architecture/modules/coding/decision docs when work references project state: code work (write/modify/remove); design (features/APIs/schemas/architecture); review (modules, past solutions)…
houtu-dependencies
Skill "houtu-dependencies" from lujiafa/houtu-project-skills, covering houtu framework — ai agent coding guide, core principles, code generation workflow (must be executed in order), step 1 — detect version & dependencies and step 2 — identify scenario & select module.
senzing-entity-resolution
Guides AI agents through Senzing entity resolution workflows using the Senzing MCP server. Covers data mapping to Senzing format, SDK code generation (Python, Java, C#, Rust, TypeScript/Node.js), documentation search, error troubleshooting, sample data access, reporting and visualization, SDK setup guides, and…