xobotyi/cc-foundry

最近记录的来源活动: 2026年10月31日 13:41

SkillsMP 收录数据更新: 2026年11月12日 12:52

已收集 skills: 66

GitHub 星标: 23
GitHub Forks: 1

用 AI 助手安装

复制这段 Prompt，发给你正在使用的 AI 助手。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/xobotyi/cc-foundry.

仓库 README

仓库 README 来源：README.md

查看 README 源文件

展开完整 README

<div align="center"> # cc-foundry [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE) [![Claude Code plugin marketplace](https://img.shields.io/badge/Claude%20Code-plugin%20marketplace-d97757.svg)](https://code.claude.com/docs/en/plugin-marketplaces) [![XERJ Corpus Hub contributor](https://hub.xerj.org/contributors/xobotyi/badge.svg)](https://hub.xerj.org/contributors/xobotyi) </div> Plugins that give Claude Code a working discipline (how it codes, commits, plans, and writes) and the language and platform knowledge to apply it. It is the set its author works with every day, made public, and its rules come with their reasons. Left to its defaults, Claude skips the skill that holds the answer and forgets the ones it loaded. It commits an hour of work as one unit, writes JavaScript from twenty years of the language at once, and resumes after your lunch break as though you never left. Each plugin below targets one failure like these. Some of them are measured: across four frontier models, agents opened a repository's AI policy file in 3.5% of unaided runs, and the `open-source` plugin starts from that number. Foundry is built for Claude Code, and the hooks, output styles, status line, and language-server configs work only there. Most of the skills adapt to other agents with little work: each one is a folder with a `SKILL.md` and any references it carries, and the usual changes are where it reads project rules (`CLAUDE.md`) and how it loads the skills it depends on. The exceptions are the few skills whose subject is Claude Code itself, such as `claude-md` and `output-style-engineering`. ## Start Here Install what matches your work. Each command adds the marketplace the first time it runs. - **You write code** — `the-coder` for the workflow, `git-commit` for the history, `skill-enforcer` so the skills load when the work calls for them, and your language: `golang`, `javascript`, `php`, `python`, or `rust`. - **You build prompts, skills, agents, or output styles** — `ai-helpers`. - **You plan work before anyone builds it** — `the-blueprint`. - **Your sessions run long or unattended** — `wall-clock` for the time, `the-statusline` for context and rate limits, and `the-workflow` for handoffs between sessions. ```text /plugin install the-coder --marketplace xobotyi/cc-foundry /plugin install git-commit --marketplace xobotyi/cc-foundry /plugin install skill-enforcer --marketplace xobotyi/cc-foundry /plugin install golang --marketplace xobotyi/cc-foundry ``` Swap `golang` for your language. ## Built From Daily Use Foundry grows out of its author's own work. A plugin exists because a real workflow needed it, and it is maintained as part of a working setup: fixed when it fails in use, changed when the work changes. Nothing here was built to fill a catalog or to chase attention. The repository itself is developed with these plugins, so it doubles as a sample of what they produce. [`.claude/settings.json`](.claude/settings.json) enables eight of them for every session opened here, and the commit log is `git-commit` output. A fix from the status line, cause first, then what the user saw ([ae8e706](https://github.com/xobotyi/cc-foundry/commit/ae8e706)): ```text [the-statusline] keep the root when collapsing an absolute path Collapsing cut each directory segment to its first character, and the empty segment before a leading slash became undefined, which the posix join rejects. A working directory outside home and the project crashed the renderer on a narrow pane, leaving the status line blank. ``` The decisions about how the skills are written live in [`design-docs/adr/`](design-docs/adr/), kept with the `the-blueprint` plugin's `adr` skill and indexed in [`design-docs/ADR.md`](design-docs/ADR.md) for every session. ## Workflow ### the-coder Claude writes code before it reads what exists: it guesses at APIs, skips tests, and multiplies abstractions. The `coding` skill enforces a discovery-first workflow (discover, plan, implement, verify), and the `software-engineer` output style carries the peer-engineering register and the skill queue. Every language plugin runs on top of it. [Read more](plugins/the-coder/) ### skill-enforcer Claude skips skills and forgets them mid-session. A mod makes it check which skills apply at every checkpoint: your prompt, after it reads a file, after it edits one, and after it loads a skill. When the work shifts from coding to testing, it goes back to the references it has not read. [Read more](plugins/skill-enforcer/) ### git-commit An hour of work lands as one commit under a message that says nothing. The `commit` skill splits the diff into the smallest units it judges to keep the tree working, orders them, and audits every message body paragraph by paragraph against the staged diff, so a body that narrates the code gets cut before it is committed. [Read more](plugins/git-commit/) ### the-blueprint Planning comes out too shallow or too detailed. The DRAFT pipeline (Discovery → Research → Alignment → Frame → Tasks) moves an idea through approval gates to sized work items. Standalone skills cover tracker items (`task-creation`), shared vocabulary (`glossary`), living decision records (`adr`), YouTrack behavior (`youtrack`), and diagrams (`diagramming`). [Read more](plugins/the-blueprint/) ### the-workflow The mechanics of working with Claude Code across sessions. `claude-md` owns CLAUDE.md and `.claude/rules/`, starting from which files load when and what survives compaction. `handoff` writes a 500–2000-token transfer document when work crosses a context boundary, and `debrief` turns the corrections you made into rules the agent reads next time, after you approve them. [Read more](plugins/the-workflow/) ### the-writer Agent prose carries tells, and readers discount the substance along with the style. The `humanize` skill traces them to the post-training pressures that produce them, and decides whether a text should be edited before it decides what to edit. [Read more](plugins/the-writer/) ### open-source Contributing to a project you don't maintain means following rules that live outside its code. `issue-writing` and `pr-contribution` start from the project's own files and draft reports and pull requests for a human to submit. [Read more](plugins/open-source/) ## Session ### wall-clock Claude Code gives the model today's date and nothing else about time. A mod injects the wall clock with your prompt, with how long you were away, and once a minute during a long turn. [Read more](plugins/wall-clock/) ### the-statusline A two-row status line: output style, model and effort, session cost, context left, the rate limit closest to running out, working directory, git branch, and lines changed. The color grows more urgent as context or limits run low. [Read more](plugins/the-statusline/) ### plugin-rollcall A plugin a repository enables can be missing from your sessions in a fresh clone or a new worktree until you type `/reload-plugins`. This mod runs the reload at session start and holds your first prompt until it completes. Install it at user scope. [Read more](plugins/plugin-rollcall/) ### better-memory Auto memory saves one-session requests as standing rules and keeps notes long after they stop being true. With this mod, the agent saves memory only through a tool that steers instructions to CLAUDE.md and refuses any entry that cites nothing from the session: your words, a file, a page, or a command's output. When the session goes idle, a separate Claude process removes every entry that no longer holds, and `/better-memory:sift` runs the same review with you, moving the rules you approve into CLAUDE.md. [Read more](plugins/better-memory/) ## AI Artifacts ### ai-helpers Skills for each artifact you build for an agent: `prompt-engineering` as the foundation, then `skill-engineering`, `subagent-engineering`, `output-style-engineering`, and `prompt-terser` for a retrospective audit. The `ai-engineer` output style sets the register for the work. [Read more](plugins/ai-helpers/) ## Languages Each language plugin ships a language-server configuration for semantic navigation; its README names the server to install. ### golang Conventions, error handling, interfaces, generics, concurrency, and testing, written for the version the module's `go` directive permits. Library skills: `templ`, `charm-tui` for the Charmbracelet v2 stack, and `zog`. [Read more](plugins/golang/) ### javascript Claude writes JavaScript from twenty years of the language at once. `javascript`, `typescript`, `nodejs`, `bun`, and `vitest` hold it to the baseline and the runtime version the project declares. [Read more](plugins/javascript/) ### php PHP 8.5 and later: `php` for the language, Composer, and PSR conventions, with `phpunit` and `pest` for tests. [Read more](plugins/php/) ### python Python 3.14 and later: `python` for idioms, typing, and packaging with uv and ruff, with `pytest` for tests. [Read more](plugins/python/) ### rust The 2024 edition: `rust` for ownership, error handling, traits, and concurrency (threads and rayon for CPU work, Tokio for I/O), with `rust-testing` for tests. [Read more](plugins/rust/) ## Platforms ### frontend `css`, `tailwindcss` (v4), `react`, `vue`, `svelte` (Svelte 5), and `accessibility` (WCAG 2.2), plus `frontend-design` for which visible choice to make and how to verify the render across widths and states. [Read more](plugins/frontend/) ### backend `observability` for the practice of logging, metrics, and tracing together, with `prometheus`, `statsd`, and `otel-tracing` for the tools. [Read more](plugins/backend/) ### grafana The other side of `backend`: querying, visualizing, and alerting on telemetry through Grafana. Skills: `dashboards`, `promql`, `metricsql`, `logsql`, `alerting`, `provisioning`, and `dataviz`. [Read more](plugins/grafana/) ### cli `cli` designs command-line interfaces in any language: arguments, output streams, exit codes, configuration, and signals. `shell-scripting` writes the scripts: strict mode, quoting, portability, and error handling. [Read more](plugins/cli/) ### infrastructure Without domain knowledge, agents ship configurations with insecure defaults, no idempotency, and naive networking. `devops` governs the change itself, from target identity to the evidence that closes it, and stays active beside the tool skill: `ansible`, `containers`, `proxmox`, or `networking`. [Read more](plugins/infrastructure/) ## Updates Claude Code does not update plugins from this marketplace in the background until you turn it on. Run `/plugin`, open the **Marketplaces** tab, select **foundry**, and select **Enable auto-update**. To update by hand, select **Update marketplace** on the same screen. It refreshes the plugin list and updates every plugin you installed from foundry. From the shell, `claude plugin update <plugin-name>@foundry` updates one plugin. ## License MIT

正在加载仓库 README…

这个仓库中的 skills

软件开发工程师网络与计算机系统管理员项目管理专家其他计算机职业

8 个职业分类 · 已分类 64%

软件开发工程师

Rust language conventions, idioms, error handling, concurrency (CPU-bound parallelism with threads/rayon and I/O-bound async with Tokio), the cargo/clippy/rustfmt toolchain, and rustdoc. Invoke whenever task involves any interaction with Rust code — writing,…

原文语言：英语

2026-10-28

已展示 40 / 66 个已收集 Skill。