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 agentmods add skills/xsxdot/super-dev/superdevnpx skills add Xsxdot/super-dev --skill superdevgit clone --depth 1 https://github.com/Xsxdot/super-devWhat 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 | $0.00250 | $0.04639 |
| Opus 5 | $0.00125 | $0.02320 |
| Sonnet 5 | $0.00050 | $0.00928 |
| Haiku 4.5 | $0.00025 | $0.00464 |
Grade A, and why
superdev 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 yesterday.
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 — 151 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SuperDev MCP 使用指南
核心理念
工具给证据，AI 下根因
diagnose_service、analyze_trace_logs、summarize_error_window 只收集确定性证据。不要把工具输出当成根因照搬；先采集运行状态、日志、trace、错误聚合，再由 AI 明确写出推理链和置信度。
读写分离 + 双层安全门
只读工具可放心使用。配置写入必须走 preview_config_change → apply_config_change。所有写工具采用统一审批模型：直接调用（不传 approval_token），需要审批时 MCP 默认在 SuperDev 桌面端等待用户批准并自动带 token 续跑，对你无感。是否真正审批由用户配置的开关决定，用户也可在批准时开启「项目级免审窗口」让后续同项目操作自动通过。详见 references/safe-operations.md。
第一步：永远先建立全局视野
开始任何 SuperDev MCP 任务前，先用 get_runtime_snapshot 或 list_services 摸清项目、服务、deployment、环境和状态。用户已经指定项目时用 list_services，用户只说“看看 SuperDev 怎么了”时用 get_runtime_snapshot。
不要一上来就调用具体写操作，也不要在没有证据时猜测根因。
配置远程主机前必须先调用 list_hosts。host_ids 只能填写 list_hosts 返回的非本机主机 hosts[].id（is_self=false），不能填写 hosts[].name、SSH Host、机器名或用户口头描述。
总决策树
| 用户意图 | 先做什么 | 继续阅读 |
|---|---|---|
| 把服务跑起来 / 重启 / 停掉（哪怕没提 SuperDev） | 先 list_services 看该项目是否已被 SuperDev 接管；已接管则走 start_service/restart_service/stop_service，不要 shell 自己起 |
本页「第零纪律」 |
| 我（AI）刚改了被接管服务的代码，且改的是不热更新的部分 | 改动落盘后主动 restart_service（编译型先确认 deployment 会重新编译），再让用户验证 |
本页「第零·五纪律」 |
| 服务挂了、报错、为什么慢 | list_services 定位 deployment，然后 diagnose_service 采证 |
references/debugging-workflow.md |
| 看日志、查某个错误 | 按已知信息选择 tail_logs / follow_logs / search_logs / get_log_context |
references/log-tools.md |
| AI 调试遇到登录/鉴权墙 | 先 get_debug_credentials 取项目/服务授信的测试账号或 api-key，再正常填登录框或带请求头 |
本页「AI 调试凭据纪律」 |
| 调试本机前端页面（点击/输入/截图/读 console，哪怕没提 SuperDev） | list_browser_targets 找 deployment → open_browser_debug_session（走审批）→ browser_snapshot 取 selector → 控制工具 |
references/browser-debug.md |
| 日志/诊断都定位不到，要停在某行源码看栈和变量（Go/Node/Python/Rust/C++/Java/Kotlin） | 确认服务在跑 → list_code_debug_targets → debug_capture_at（attach 运行中进程，不重启） |
references/code-debug.md |
| 新建受管语言运行时服务（Go/Node/Python/Java/Kotlin/Rust/C++）、不知配置填什么 | list_language_runtime_providers → describe_language_runtime_schema 照字段填，别猜命令串，再 preview→apply |
references/safe-operations.md |
改项目、服务、deployment、pipeline 配置（含 start_on_boot / depends_on / readiness） |
先读现状，再 preview，再直接 apply（需审批时自动等待续跑） | references/safe-operations.md |
| 启动、停止、重启服务、部署/回滚 pipeline、导入模板 | 可选 preview_operation，直接调用写工具并等待审批自动续跑 |
references/safe-operations.md |
| 部署、上线、回滚、查看 pipeline 运行 | 区分模板、配置、执行、观测四段 | references/pipeline.md |
| 记录一次排查过程 | 建立 debug session，过程中追加分析和观察 | references/debugging-workflow.md |
What ships with it
10 files 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.
- hooks/bootstrap.md 1.6 KB
- hooks/hermes-session-context 1.8 KB
- hooks/run-hook.cmd 1.6 KB runs code
- hooks/session-start 2.2 KB
- references/browser-debug.md 7.8 KB
- references/code-debug.md 5.4 KB
- references/debugging-workflow.md 2.5 KB
- references/log-tools.md 3.4 KB
- references/pipeline.md 5.1 KB
- references/safe-operations.md 12 KB
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.
- yesterday First seen · 151 lines · 250 tokens per session scan A 73c62bdcbaa7
superdev is a skill published in the GitHub repository Xsxdot/super-dev (1 stars, last pushed 11d ago), licensed Apache-2.0. It adds 250 tokens to every session and 4,639 once invoked, about $0.0013 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 2026-08-31.
Other skills, from other repositories
chrome-devtools
Use Chrome DevTools MCP to control and inspect a live Chrome instance for network, console, performance, rendering, and Deep debugging. Pairs with playwright-cli (deterministic interaction/E2E) — complements, not duplicates.
cmux
End-user control of cmux topology and routing (windows, workspaces, panes/surfaces, focus, moves, reorder, identify, trigger flash). Use when automation needs deterministic placement and navigation in a multi-pane cmux layout.
docs
Scale up scraping and drive interactive pages. Use batchscrape for many URLs, getbatchresults to page through async output, scrapewithactions to interact before scraping, and generatellmstxt to produce a site's AI policy file.
crawlforge-getting-started
Orientation and tool-selection guide for the CrawlForge MCP server's 28 web tools. Use when the user is getting started with CrawlForge, asks which CrawlForge tool to use, how to set up the API key, how skills or the CLI work, what a tool costs in credits, or when one tool fails and a fallback is needed. Routes…
crawlforge-web-scraping
Scrapes web pages and returns clean Markdown, HTML, plain text, links, or metadata using CrawlForge's scrape, fetchurl, extractcontent, extracttext, extractlinks, extractmetadata, mapsite, and crawldeep tools. Use when the user wants to scrape a URL, fetch a page, get the markdown or text of a website, extract links…
crawlforge-batch-automation
Automates large scraping jobs and browser interactions with CrawlForge's batchscrape, getbatchresults, scrapewithactions, and generatellmstxt tools. Use when the user wants to scrape many URLs at once, batch-scrape a list of pages, collect dozens of product, news, or competitor pages, run browser actions (click, type…