harzva Everything Agent-CLI to Claude Code
Overview
This skill collection bridges CLI-first model access back into a Claude Code-centered workflow.
It solves the problem where:
- access is granted through web login
- usage is tied to an official CLI
- free or subscription quotas exist
- but a normal public API is missing, incomplete, or does not match the paid entitlement
Key idea: Claude Code or LikeCode stays the control plane, while external CLIs become specialized workers.
Modes
1. Single-provider mode
Use this when you want one external model for a specific task.
Pattern:
- Prepare prompt text
- Invoke provider wrapper (e.g.,
bin/usecli-gemini.sh) - Capture stdout as result
- Hand result back to Claude Code
2. Multi-model review mode
Use this when you want multiple perspectives in one task.
Pattern:
- Prepare shared prompt
- Call multiple provider wrappers in sequence or parallel
- Collect all outputs
- Merge findings in Claude Code
See references/workflow-multi-model-review.md for the full workflow.
Sub-skills
This collection contains four provider-specific sub-skills that can be installed independently:
| Sub-skill | Purpose | Best for |
|---|---|---|
usecli-codex |
OpenAI Codex integration | implementation-oriented feedback, code-aware review |
usecli-gemini |
Google Gemini integration | broad review, first-pass critique, structured feedback |
usecli-qwen |
Alibaba Qwen integration | secondary critique, math reasoning, multilingual |
usecli-grok |
xAI Grok integration | critical review, alternative perspectives |
Each sub-skill has its own SKILL.md, README.md, scripts/, and tests/ - fully independently installable and testable.
Recommended workspace layout
For multi-model workflows, prefer:
- prompts in
examples/prompts/ - outputs in
examples/outputs/ - workflow scripts in
examples/workflows/ - reference docs in
references/
Guardrails
- Do not assume external CLIs are available without checking first
- Do not expose API keys or auth tokens in command arguments
- Keep each provider call bounded to one task
- Prefer prompt files over inline prompts for long or sensitive content
- Use timeout environment variables for long-running calls
Resources
bin/usecli-lib.sh
Shared library for all provider wrappers. Provides argument parsing, prompt handling, and execution helpers.
bin/usecli-codex.sh
Codex CLI wrapper. See usecli-codex/SKILL.md.
bin/usecli-gemini.sh
Gemini CLI wrapper. See usecli-gemini/SKILL.md.
bin/usecli-qwen.sh
Qwen CLI wrapper. See usecli-qwen/SKILL.md.
bin/usecli-grok.sh
Grok CLI wrapper. See usecli-grok/SKILL.md.
examples/workflows/multi-model-review.sh
Runnable multi-model review workflow. Demonstrates calling Gemini and Qwen in sequence.
tests/test_wrappers.sh
Smoke test for all umbrella wrappers. Proves command construction works without provider login.
references/skill-naming.md
Naming strategy for unified usecli:* entrypoints and provider-specific <provider>:<action> skills.
references/command-skill-contract.md
Minimal executable contract for command-oriented skills.
references/architecture.md
Layering model: auth, mapping, translation, integration, UX.