builder-essential-skills

Builder's Essential Skills
A considered collection of reusable workflows for planning, building, researching, learning, validating, writing, and reviewing with Claude Code and Codex.

Install · Pick a skill · Create a skill

Install

Install with the skills.sh CLI:

npx skills add tamdogood/skills

Install one skill from the collection:

npx skills add tamdogood/skills --skill lead

The skills.sh page for this repository is skills.sh/tamdogood/skills. It appears after the repository has been installed through the skills CLI and the skills.sh cache refreshes.

You can also use the bundled npm installer.

Install every skill for your user account:

npx @tamng0905/builder-essential-skills

Install into the repository you are currently in:

npx @tamng0905/builder-essential-skills --project

Install only one skill:

npx @tamng0905/builder-essential-skills --skill lead

Add --project to that command to install the selected skill into the current repository. Restart Claude Code or Codex after installation.

Pick a skill

Every skill lives in skills/<name>/SKILL.md. Some also ship scripts, reference material, or agent definitions that support the workflow.

Why this exists

An AI agent is much more useful when it has a repeatable way to handle the work. This collection packages practical workflows as portable skills: compact instructions, scripts, references, and agent definitions that you can install, adapt, and share.

Use the whole collection, or take only the skills your team needs.

Local install

Use the local scripts only when you are installing from a cloned checkout.

From the repository root, install every skill for the current user:

./install.sh

The npx command installs the same files as the scripts below. It supports macOS, Linux, and Windows wherever Node.js 18 or newer is available.

On Windows:

./install.ps1

Restart Claude Code or Codex, then invoke a skill in plain English:

/lead Build the smallest version of this feature and open a PR.
$async-learning-teacher Teach me this paper step by step: https://arxiv.org/abs/...
$map-the-landscape Show me how this repository fits together.
$paper-opportunity-radar Audit this topic's buried research for credible opportunities.

Want to try the collection inside one repository first? Install it locally instead:

./install.sh --project

./install.ps1 -Project

The project install uses .claude/skills and .codex/skills, so it stays with that repository rather than changing your user-level setup.

What gets installed

The installers copy each folder in skills/ to the right location for your agent runtime.

You will need Claude Code and/or Codex, depending on which agent you use. Some bundled deterministic tools, including session-profiler, require Python 3.

Go deeper

Native-agent orchestration

lead and lead-research use the subagent tools provided by whichever runtime invokes them. They probe concurrency, isolation, messaging, shell, write, and web capabilities at the start of a run, then assign work by role. There are no model IDs or provider CLI commands to configure when switching between Codex and Claude Code. A runtime without native subagent delegation stops instead of letting the orchestrator write code or perform research itself.

Profile an agent session

After installation, point SP at the wrapper for Hermes:

SP="${HERMES_HOME:-$HOME/.hermes}/skills/session-profiler/scripts/sp"

Then explore your sessions:

$SP list --provider hermes --n 20
$SP info <session-id-or-jsonl-path>
$SP parse <session-id-or-jsonl-path>
$SP brief
$SP agent-summary
$SP costs
$SP slowest-tools --n 20
$SP errors
$SP turns

Generate richer artifacts:

$SP toc
$SP review
$SP trace

brief writes a scan-friendly brief.md with headline metrics, standout slow paths or failures, a prompt trail or TOC storyline, and an agent scoreboard. The Perfetto trace adds an overview lane, labeled TOC phases, per-agent work lanes, prompt lanes, and cumulative token/cost counters.

Cost reports are estimates based on public model pricing, not billing records. Session files and traces can include prompts, code, file paths, and other sensitive project context. Review them before sharing.

Repository map

skills/                  The skill collection
  <skill>/SKILL.md        Each skill's entry point and workflow
assets/readme-hero.png    README banner
install.sh                macOS/Linux installer
install.ps1               Windows installer
package.json              npm package metadata and npx entrypoint
bin/builder-essential-skills.js     cross-platform npx installer

Publish the installer

Choose an available npm package name in package.json, authenticate with npm, then publish from the repository root:

npm login
npm publish --access public

npm requires either 2FA enabled on the account, followed by the one-time code requested by npm publish, or a granular access token with read/write access and Bypass two-factor authentication enabled. For a local interactive publish, enabling 2FA and running npm login again is the simplest option. Never commit an access token; use npm's login flow or a secure CI secret.

After publishing, users can run npx @tamng0905/builder-essential-skills without cloning the repository. Bump the version in package.json for each subsequent publish.

Bring your own workflow

Start with one folder:

skills/my-skill/
  SKILL.md

Give SKILL.md clear frontmatter and a focused workflow:

---
name: my-skill
description: Explain when an agent should use this skill.
---

# My Skill

Write the workflow, rules, examples, and expected output here.

Keep skills easy to move between projects:

• Avoid hard-coding a company, repository, branch, or tool unless the skill is intentionally private.
• Put reusable scripts beside the skill instead of relying on global shell state.
• Include request examples that show exactly when the skill should run.
• Create assets/skill-cards/<name>.svg for each new skill and add it to the README's "Pick a skill" table.
• Never put credentials, tokens, or private data in a skill folder.

Re-run the installer after adding or changing a skill:

./install.sh

License

MIT. See LICENSE.