tungnt-ai-skills

tungnt-ai-skills

tungnt-ai-skills is a personal fork of obra/superpowers with custom bootstrap rules, curated workflow skills, fork-specific plugin metadata, and local agent setup adjustments.

What This Fork Provides

• using-tungnt-ai-skills bootstrap rules for this fork
• workflow skills under skills/ for investigation, quick fixes, brainstorming, planning, implementation, review, and branch finish work
• domain skills for API design, security hardening, and UI/UX evidence inside the normal workflow
• manual utility skills such as prompt-leverage for explicit prompt upgrade, clarification, templating, or apply flows
• writing-skills guidance for creating, editing, and pressure-testing skills
• optional YAML plan status tracking under docs/tungnt-ai-skills/status/
• plugin metadata for Claude Code, Codex, GitHub Copilot CLI, Gemini CLI, and Google Antigravity
• a zero-dependency npm installer that copies the package into supported agent plugin directories

The active docs root for plans, specs, investigations, and status files is docs/tungnt-ai-skills/.

Menu

• What This Fork Provides
• Repository Layout
• Core Workflow
• Install
  • Install Menu
  • 1. NPM/npx Installer
  • 2. Native Installer Mode
  • 3. Manual Setup When NPM/Npx Is Not Available
• Update
• Manual Harness Notes
• Development
• License

Repository Layout

• skills/ bundled workflow skills
• skills/using-tungnt-ai-skills/ session bootstrap and platform references
• docs/tungnt-ai-skills/ active plans and specs root
• hooks/ session bootstrap and cross-platform hook wrappers
• installer/ npm installer target map and copy logic
• bin/tungnt-ai-skills.js npm executable entrypoint
• .claude-plugin/, .codex-plugin/, .agents/, .opencode/ harness metadata
• tests/ regression and installer tests

Core Workflow

Agents should start with using-tungnt-ai-skills, then choose the smallest relevant process skill:

• bug diagnosis, incident tracing, or unfamiliar code exploration: investigation
• trivial low-risk changes that can be completed quickly: quick-dev
• fuzzy idea or design work: brainstorming
• approved design that needs an implementation plan: writing-plans
• substantial work that needs isolation: using-git-worktrees
• plan execution with subagents: subagent-driven-development
• plan execution without subagents: executing-plans
• review before handoff: requesting-code-review
• final merge, cleanup, or handoff: finishing-a-development-branch
• creating or updating reusable skills: writing-skills

Domain skills are supporting references, not workflow selectors:

• REST/HTTP contract design evidence: api-design
• application security and DevSecOps evidence: security-and-hardening
• UI/UX design evidence: ui-ux-pro-max

Domain skills must be used inside the selected process workflow. They do not replace brainstorming, writing-plans, execution, review, investigation, or writing-skills gates.

Manual utility skills do not auto-trigger:

• prompt-leverage runs only when the user explicitly invokes skill:prompt-leverage or asks to improve, upgrade, clarify, template, or apply a raw prompt. skill:prompt-leverage prompt: <text> upgrades only; skill:prompt-leverage apply prompt: <text> upgrades first, then restarts the normal workflow selection.

Skill calls use the real names from each SKILL.md file, not a plugin-prefixed namespace.

Recent workflow additions:

• prompt-leverage provides a manual prompt preprocessor for explicit prompt-upgrade requests without changing the normal workflow gates.
• ui-ux-pro-max provides UI/UX design intelligence as a domain skill; use it inside the normal workflow without replacing brainstorming, writing-plans, or execution gates.
• brainstorming can emit an optional Spec Kernel with goal, users, acceptance criteria, constraints, and out-of-scope items for handoff to writing-plans.
• executing-plans and subagent-driven-development can maintain lightweight YAML status files at docs/tungnt-ai-skills/status/<plan-name>-status.yaml.
• executing-plans checks for unresolved review continuation items before starting new plan tasks.
• requesting-code-review and the subagent code-quality prompt use Blind Hunter, Edge Case Hunter, and Acceptance Auditor lenses with Must-Fix, Should-Fix, Consider, and Praise buckets.
• finishing-a-development-branch now validates Definition-of-Done before presenting merge, PR, keep, or discard options.

Install

Install Menu

Choose the path that matches your environment: