hoangnb24/skills
hoangnb24/skills
Marketplace Khuym is a focused Codex plugin containing five independent meta-skills and one tool-automation skill. It improves how an agent understands, researches, sequences, lands, and visualizes work without imposing a repository workflow.
No marketplace file This repo has no .claude-plugin/marketplace.json at its root, so /plugin marketplace add stops with an error. It ships plugins for OpenClaw or another agent instead.
Install
The repo has no one-line install. Follow its README.
Files
Khuym Skills
Khuym is a focused Codex plugin containing five independent meta-skills and one tool-automation skill. It improves how an agent understands, researches, sequences, lands, and visualizes work without imposing a repository workflow.
There is no required chain, onboarding step, plugin-specific state directory, issue graph, handoff file, or approval gate. Invoke one skill when its behavior helps; combine several only when the task genuinely needs them.
Skills
| Skill | Use it when | Concrete result |
|---|---|---|
prompt-leverage |
A prompt is vague, noisy, or missing execution rules | An execution-ready prompt with objective, context, verification, and done criteria |
goal-griller |
An idea is too fuzzy for autonomous /goal work |
A verifiable goal prompt produced through one focused question at a time |
xia |
A feature is unfamiliar, ambiguous, version-sensitive, or risky | An evidence-labeled research brief based on local code, upstream patterns, and current official docs |
sequence-execution-plan |
Priority order conflicts with dependency or risk order | A dependency-aware Now/Next/Later plan with explicit cause and effect |
smart-commits |
Existing changes need to become a clean commit stack | Intent-based commits, appropriate validation, and a push when a destination exists |
tldraw-api |
A tldraw canvas should be created, edited, scripted, or verified without mouse automation | A saved .tldraw diagram built through the authenticated local Canvas API |
Concrete Examples
Improve a weak prompt
Input:
Make the API faster.
prompt-leverage preserves the intent, then adds the missing execution contract. The result identifies which API evidence to inspect, how to measure latency, what behavior must not regress, and what proves completion.
Cause and effect:
- The original prompt has no measurable target.
- Without a target, an agent can make arbitrary “performance” edits.
- Adding a benchmark and regression boundary makes the work verifiable.
Research before building
Input:
Add passkeys to this app.
xia first reads the repository's auth packages and versions, then searches for reusable local seams, upstream patterns, and current official docs. It returns a recommendation before code is changed.
Cause and effect:
- Passkey APIs differ by framework and version.
- Guessing the stack can produce an incompatible design.
- Mapping local reality first makes the recommendation fit the actual repository.
Sequence urgent work honestly
Input:
P0: stop duplicate charges. P2: add idempotency storage.
sequence-execution-plan can place the P2 prerequisite before the durable P0 fix without lowering the P0 outcome. If duplicate charges are active, it also places a small mitigation first.
Cause and effect:
- A priority says what matters most; it does not prove what can execute first.
- Durable duplicate-charge prevention may require idempotency storage.
- Therefore the safe order can be mitigation → storage → durable fix → cleanup while the P0 remains open.
Automate a tldraw diagram
Input:
Draw this architecture in tldraw without Computer Use.
tldraw-api discovers the running desktop app, creates stable shape records, connects nodes with real bindings, saves the document, and verifies it with the canvas linter and a screenshot.
Cause and effect:
- Mouse automation positions pixels but does not prove connector semantics.
- API-created bindings keep arrows attached when nodes move.
- Saving through tldraw plus lint and screenshot checks proves both structural and visual correctness.
Optional Composition
The skills can compose, but none requires another:
rough idea
├─ goal-griller → make the outcome verifiable
├─ xia → reduce implementation uncertainty
├─ sequence-execution-plan → order the resulting work
├─ smart-commits → land completed changes cleanly
└─ tldraw-api → create or edit a canvas directly
For a simple request, use only the one relevant skill. For example, “group these changes into commits” should invoke smart-commits directly and should not trigger research or planning ceremony.
Install In Codex
- Clone this repository.
- Add
.agents/plugins/marketplace.jsonas a local Codex marketplace. - Install the
khuymplugin. - Restart Codex if the marketplace does not appear immediately.
The canonical plugin lives at plugins/khuym/, and its skills live at plugins/khuym/skills/.
The plugin advertises Exa and DeepWiki as optional research paths for xia. Xia can continue with other available browser/search paths when those services are unavailable.
Raw Skill Mirrors
Mirror the canonical skill directories for tools that consume raw skills:
bash scripts/sync-skills.sh --target agents
bash scripts/sync-skills.sh --target claudeThe sync removes stale symlinks previously created from this repository, including links for retired skills. It never removes unrelated skills or real directories.
Preview without changing either target:
bash scripts/sync-skills.sh --target all --dry-runValidate Changes
bash scripts/check-markdown-links.sh
bash scripts/sync-skills.sh --target all --dry-run
python3 plugins/khuym/skills/prompt-leverage/scripts/test_augment_prompt.py
node scripts/test-goal-guard-hook.mjsSee CONTRIBUTING.md for the skill format and validation checklist.
Facts
- Kind
- Marketplace
- Repo
- hoangnb24/skills
- Group
- Uncategorized
- Language
- JavaScript
- Created
- 2026-01-09
- Forks
- 68
- Plugins
- 14
- 1f/prompts.chatf/prompts.chatf.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.
- 2affaan-m/everything-claude-codeaffaan-m/everything-claude-codeThe agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
- 3obra/superpowersobra/superpowersAn agentic skills framework & software development methodology that works.
- 4anthropics/skillsanthropics/skillsPublic repository for Agent Skills
- 5anthropics/claude-codeanthropics/claude-codeClaude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
- 6nextlevelbuilder/ui-ux-pro-max-skillnextlevelbuilder/ui-ux-pro-max-skillAn AI skill that provides design intelligence for building professional UI/UX across multiple platforms.