Efficient Coding

English | 中文

This repo contains:

• Optional MCP server configuration.
• Reusable agent skills/playbooks you can install via bunx skills add.

MCP Servers

Server	Purpose	Transport	Command	Source
fetcher (fetcher-mcp)	Fetch web page content using a Playwright headless browser.	stdio	bunx -y fetcher-mcp	https://www.npmjs.com/package/fetcher-mcp

To use an MCP server, add it to your coding agent’s MCP configuration. A generic example config lives at .mcp.json.

Skills

Skills are reusable capability/workflow/methodology playbooks you can invoke to approach a task (e.g. planning, debugging, TDD).

Skill	Category	When to use
brainstorming	Workflow	New feature / unclear requirements; produce a design/spec first.
codex-skill-creator	Meta	Create, improve, benchmark, review, and package Codex skills, including paired eval runs and eval-viewer/generate_review.py review loops.
cubox-research	Research	Search a user's Cubox collection deeply by topic, vary keywords intentionally, fetch strong article matches, and analyze exported Markdown locally. Requires Bun and a configured .env.
harness	Workflow	Transform any code project into one optimized for AI agent collaboration through harness engineering. Scans codebase, extracts engineering knowledge, and generates structured context documents.
memos	Integration	Work with the Memos REST API for memos, attachments, and activities: create/read/query/update/delete memos, manage comments/reactions/relations/attachments, and inspect activities. Requires Bun and a configured .env.
dev-browser	Automation	Browser/web automation: navigate pages, click/fill forms, take screenshots, scrape data, or test authenticated flows.
exa-web-search	Research	Free AI web/code/company search via Exa MCP (no API key) when you need current information or code examples.
readwise-research	Research	Build a topic memo from the user's Readwise/Reader documents and highlights, then suggest shortlist/tag/archive actions before any mutation.
see	Integration	Integrate with S.EE APIs for short URLs, text sharing, and file sharing.
shape	Product	Shape fuzzy ideas into clear product decisions and SPEC docs before coding.
impeccable (external bookmark)	Reference (External)	Frontend design reference bundle (frontend-design skill + 17 commands + anti-patterns) for critique, polish, motion, color, and responsive refinement; bookmark only (not included locally in this repo).
ui-ux-pro-max-skill (external bookmark)	Reference (External)	UI/UX prompts/workflows reference; bookmark only (not included locally in this repo).
systematic-debugging	Workflow	Bugs, flakes, or “unexpected behavior”; find root cause before fixing.
writing-plans	Workflow	Approach is decided; turn it into an executable plan with steps + verification.
executing-plans	Execution	Run a written plan in small batches with review checkpoints.
subagent-driven-development	Execution	Run a plan in-session: one subagent per task + spec/quality review loops.
test-driven-development	Implementation	Any feature/bugfix/refactor: Red → Green → Refactor (no code without a failing test).
worktree-manager	Workflow	Worktree management via Worktrunk (wt): switch/create/list/merge/remove with safety guardrails.

Install

• bunx skills add http://github.com/jtsang4/efficient-coding --skill brainstorming
• bunx skills add http://github.com/jtsang4/efficient-coding --skill codex-skill-creator
• bunx skills add http://github.com/jtsang4/efficient-coding --skill memos
• Template: bunx skills add http://github.com/jtsang4/efficient-coding --skill <skill>

Use

• If you want a specific skill, say so explicitly (named skill wins).
• For creating a new skill, revising an existing one, improving trigger descriptions, or running a human-review eval loop for skills, use codex-skill-creator.
• If multiple skills apply, default to workflow first: brainstorming/systematic-debugging → writing-plans → execute (executing-plans | subagent-driven-development) → test-driven-development inside each task.
• For bugs: systematic-debugging → add a failing test → fix with test-driven-development.
• For browser interaction tasks (navigate/click/fill/screenshot/scrape), use dev-browser.
• For questions that should be answered from the user's Cubox library, use cubox-research first; it searches broadly, fetches the strongest article details, and stays read-only unless the user explicitly asks for a mutation.
• For Memos API tasks around memos, attachments, or activities, use memos; it provides a Bun CLI plus bundled API guidance and expects a local .env with MEMOS_BASE_URL and MEMOS_ACCESS_TOKEN.
• For web/code/company research tasks, use exa-web-search.
• For questions about what the user has already read, saved, or highlighted in Readwise/Reader, use readwise-research first; it treats the Readwise library as the source of truth and stays read-only until explicit confirmation.
• For frontend visual design, UI critique/polish, motion, color, or responsive refinement, keep impeccable handy as an external design reference.

Skill sources (optional)

This table lists the source repositories for skills installed/updated from external repos. The Notes column summarizes local customizations in this repo (if any).

Skill	Source repo	Notes
brainstorming	obra/superpowers	Worktree ops are delegated to worktree-manager (copy current working state).
systematic-debugging	obra/superpowers	If you need a dedicated worktree to isolate a repro, use worktree-manager.
writing-plans	obra/superpowers	Worktree ops are delegated to worktree-manager (copy current working state).
executing-plans	obra/superpowers
subagent-driven-development	obra/superpowers
test-driven-development	obra/superpowers

Standards

These are the default engineering conventions recommended in this repo. The table stays flat on purpose so future specs can be added as new rows without changing the structure.

Category	Applies to	Recommendation	Quick note	Reference
Project layout	Go services/apps	golang-standards/project-layout	Good default for larger Go codebases; common directories include cmd, internal, and pkg. Keep small projects simpler when the extra structure is unnecessary.	https://github.com/golang-standards/project-layout
Project layout	Frontend applications	Feature-Sliced Design (FSD)	Organize by layers and business slices (for example app, pages, features, entities, shared) so frontend code stays easier to scale and evolve.	https://fsd.how/docs/get-started/overview/
Lint	Frontend / JS / TS	Biome	Unified formatter and linter for frontend projects with fast performance and sensible defaults; start with the recommended rules, then tighten conventions as needed.	https://biomejs.dev/linter/
i18n	React / frontend	react-i18next	Standard choice for React apps in the i18next ecosystem; supports hooks/components, namespaces, interpolation, and pluralization.	https://github.com/i18next/react-i18next
i18n	Go services/apps	go-i18n	Use bundles plus locale files to manage translations in Go; supports plural forms, template variables, and CLI-based extract/merge workflows.	https://github.com/nicksnyder/go-i18n

Scripts

Script	Purpose	Usage
install-autojump-rs.sh	One-click installer for autojump-rs on macOS/Linux with bash/zsh/fish integration and an uninstall mode.	bash scripts/install-autojump-rs.sh or bash scripts/install-autojump-rs.sh --uninstall

Config

Recommended repo configuration:

Config	Files	What it does	Notes
Worktrunk “copy from base” hook	.config/wt.toml, scripts/wt-copy-from-base	When Worktrunk creates a new worktree, it copies the current workspace state from the base worktree into the new one (instead of a completely clean worktree).	Makes it easy to carry along git-ignored files like local dependencies, .env, caches, etc. Pairs well with the worktree-manager skill.

License

MIT. See LICENSE.