Skip to content

About

A curated collection of reusable AI coding skills, MCP server configs, and engineering playbooks for faster, more systematic software development.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Repository files navigation

Efficient Coding

Agent skills, MCP config and small tools for working with coding agents like Claude Code and Codex.

English · 中文

Diffusion Maps · Skills · MCP · Standards · Scripts & config

Featured: Diffusion Maps

Learn a field or shape an idea across many conversations. The agent starts from a few pillars, then grows and restructures the map as you talk. Every node remembers the Claude Code or Codex conversation it came from, and a local viewer draws it all in ink on paper.

A map about Rust ownership as a two-sided mind map: five pillars with their branches, and a timeline of ink drops along the bottom

Node panel reading the conversations linked to a node, with the relevant turns marked A project-ideation map in dark mode, with a rejected option selected
Back to the source. Open a node to reread the conversations behind it, with the relevant turns marked. Not just learning. Each map defines its own node kinds and statuses, so it can also shape a project or weigh a decision.
The same Rust map in the diffusion view: pillars on a ring with branches spreading outward The index of maps, each card with a live thumbnail
Two views. Read it as a mind map (default) or switch to the diffusion view for the overall shape. All your maps. Each card shows a live thumbnail; delete the ones you no longer need.
  • Grows with the conversation. The agent decides when to add, merge, move or retire nodes, and asks only when unsure.
  • Picks up where you left off. Maps live in ~/.efficient-coding/diffusion/. Continue in any later session, in either agent.
  • Readable at any size. The mind map folds deeper levels into "+N" so names stay legible; click to unfold.
  • Replayable. The timeline replays how the map grew, one ink drop per moment of growth.
  • Zero setup. Needs only Bun: no install, no build. The viewer stops by itself when idle.
bunx skills add https://github.com/jtsang4/efficient-coding --skill diffusion-map -g

Then invoke it explicitly with a topic, e.g. /diffusion-map Rust ownership in Claude Code or $diffusion-map in Codex. How it works →

Skills

Install any skill with bunx skills add https://github.com/jtsang4/efficient-coding --skill <name> (add -g for a global install). Name a skill in your request to force it.

Plan and build. By default, workflow comes first: brainstorming / systematic-debugging → writing-plans → executing-plans or subagent-driven-development → test-driven-development inside each task.

Skill Use it when
brainstorming A feature or its requirements are still fuzzy; settle the design first.
shape A product idea needs clear decisions and a SPEC before any code.
problem-framing A solution keeps growing exceptions or a discussion won't converge; test whether the problem itself is framed wrong.
agent-native-redesign You want to rethink a project or process assuming abundant agent and token budgets.
prove-it An idea sounds feasible but is unproven; freeze pass/kill criteria, then prove or refute it with runnable evidence.
writing-plans The approach is decided; turn it into steps with verification.
executing-plans Run a written plan in small batches with review checkpoints.
subagent-driven-development Run a plan in-session: one subagent per task, then spec and quality reviews.
test-driven-development Any feature, fix or refactor: red → green → refactor.
systematic-debugging Bugs, flakes or "unexpected behaviour": find the root cause, add a failing test, then fix.
worktree-manager Create, switch, merge or remove worktrees via Worktrunk (wt) with guardrails.
merge-and-rebase A feature branch is done: commit, PR/MR, merge without squashing, rebase onto the new trunk.
prune Periodically slim an agent-built project without changing behavior: dead code, duplicates, unused features, needless abstraction, and the structure that keeps regrowing them. Invoke with no arguments; each run is logged in docs/prune/.
harness Make a codebase agent-friendly by extracting its engineering knowledge into structured context docs.

Learn and research.

Skill Use it when
diffusion-map You want to learn a field or shape an idea across sessions as a growing, replayable map (see above).
assess-source-project-fit You need to judge whether a paper, repo or talk offers anything your project lacks ("nothing worth adopting" is a valid answer).
exa-web-search You need current web, code or company information (free Exa MCP, no API key).
cubox-research The answer should come from your Cubox collection; it searches broadly and stays read-only. Needs Bun and .env.
readwise-research The answer should come from what you saved or highlighted in Readwise/Reader; it suggests actions before changing anything.

Integrate and automate.

Skill Use it when
dev-browser Navigate, click, fill forms, take screenshots, scrape, or test authenticated flows.
memos Work with the Memos API (memos, attachments, activities). Needs Bun and .env with MEMOS_BASE_URL, MEMOS_ACCESS_TOKEN.
paseo-relay Reach other Paseo hosts through authorized Relay pairing; read-only by default (SDK reads).
see Short URLs, text sharing and file sharing through S.EE.

Author and extend.

Skill Use it when
i-diagram Draw an architecture, flow, sequence, state, mind-map or timeline diagram as one self-contained SVG with light and dark themes.
codex-skill-creator Create, improve, benchmark and package Codex skills with paired eval runs and human review.
use-remote-skill Use a skill from another repo for one conversation, from an explicit source or a YAML catalog (configuration).

External references (bookmarks, not included here): impeccable for frontend design critique and polish, and ui-ux-pro-max-skill for UI/UX prompts.

Skill sources
Skill Source repo Local changes
brainstorming obra/superpowers Worktree operations go through worktree-manager (copying the current working state).
systematic-debugging obra/superpowers Use worktree-manager when a repro needs its own worktree.
writing-plans obra/superpowers Worktree operations go through worktree-manager (copying the current working state).
executing-plans obra/superpowers
subagent-driven-development obra/superpowers
test-driven-development obra/superpowers

MCP Servers

Server Purpose Command
fetcher-mcp Fetch web pages with a headless Playwright browser (stdio). bunx -y fetcher-mcp

Add it to your agent's MCP configuration; .mcp.json is a generic example.

Standards

Default engineering conventions recommended here. New ones are added as rows.

Area Applies to Recommendation Why
Project layout Go services and apps golang-standards/project-layout A good default for larger codebases (cmd, internal, pkg); keep small projects simple.
Project layout Frontend apps Feature-Sliced Design Layers plus business slices (app, pages, features, entities, shared) scale well.
Lint Frontend / JS / TS Biome One fast formatter and linter; start from the recommended rules.
i18n React react-i18next The standard in the i18next ecosystem: hooks, namespaces, interpolation, plurals.
i18n Go go-i18n Bundles plus locale files, plurals, template variables, extract/merge CLI.

Scripts & config

Item What it does Usage
install-autojump-rs.sh Installs autojump-rs on macOS/Linux with bash/zsh/fish integration. bash scripts/install-autojump-rs.sh [--uninstall]
Worktrunk "copy from base" hook New worktrees start with the base worktree's current state, including git-ignored files like dependencies, .env and caches. Pairs with worktree-manager. .config/wt.toml, scripts/wt-copy-from-base

License

MIT. See LICENSE.

About

A curated collection of reusable AI coding skills, MCP server configs, and engineering playbooks for faster, more systematic software development.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages