v1.1.0 — Skill for Designing & Reviewing CLIs

agent-native-cli

Design CLIs that AI agents can actually use. Stable stdout contracts, deterministic exit codes, dry-run, schema introspection, and delegated auth — as a review and refactor skill.

git clone https://github.com/Agents365-ai/agent-native-cli.git ~/.claude/skills/agent-native-cli

GitHub SkillsMP ClawHub

Why This Skill

Turn any CLI into an interface that serves humans, agents, and orchestrators at the same time.

📦

Stable stdout envelope

Every command returns the same shape — ok, data, error, meta. Agents parse once and never guess.

🚦

Deterministic exit codes

Each failure class maps to a documented exit code. No more treating every non-zero the same.

🔍

Schema introspection & dry-run

Full self-description layer. Preview the request shape without executing anything. Agents learn the tool on the fly.

🛡️

Safety tiers

Open / warned / hidden. Graduated command visibility keeps destructive operations off the default surface for agents.

🔑

Delegated authentication

The human owns the auth lifecycle. The agent just uses a token. Directional trust model across env vs. CLI args.

📊

14-criterion rubric

Score any CLI across 14 criteria (0–2 each) and get a prioritized P0/P1/P2 refactor plan with concrete interface examples.

The Stdout Contract

Same command, same outcome — one side is guesswork, the other is parseable.

Native CLI

$ weatherctl today --city Paris
Fetching weather for Paris...
Temp: 14°C (feels like 12°C)
Oops: network flaky, retrying (1/3)
Done.
$ echo $?
0

Agent-native CLI

$ weatherctl today --city Paris --json
{
  "ok": true,
  "data": {
    "city": "Paris",
    "temp_c": 14,
    "feels_like_c": 12
  },
  "error": null,
  "meta": { "retries": 1, "elapsed_ms": 412 }
}
$ echo $?
0

Stderr stays human. Stdout stays machine. Exit codes map to failure classes — not just success/failure.

vs Native Agent

What you get with the skill vs prompting an LLM directly.

Capability	Native Agent	This Skill
Evaluate whether a CLI is agent-native	✗	✓ 7-principle structured diagnosis
Design stdout JSON contract	Inconsistent	✓ Stable `ok` / `data` / `error` envelope
Define exit code semantics	Ad hoc	✓ Documented, deterministic per failure class
Layered `--help` & schema introspection	✗	✓ Full self-description pattern
Design dry-run previews	Rarely	✓ Always — request shape without execution
Define safety tiers for commands	✗	✓ Open / warned / hidden
Design delegated authentication	✗	✓ Human owns lifecycle; agent uses token
Separate trust: env vs. CLI args	✗	✓ Directional trust model
Produce prioritized refactor plan	Rarely	✓ P0 / P1 / P2 with examples
Score CLI across 14-criterion rubric	✗	✓ 0–2 per criterion with verdict

Install

Pick your platform. Or just ask any coding agent to clone the repo for you.

# Global install (available in all projects)
git clone https://github.com/Agents365-ai/agent-native-cli.git ~/.claude/skills/agent-native-cli

# Project-level install
git clone https://github.com/Agents365-ai/agent-native-cli.git .claude/skills/agent-native-cli

# Via ClawHub
clawhub install agent-native-cli

# Manual install
git clone https://github.com/Agents365-ai/agent-native-cli.git ~/.openclaw/skills/agent-native-cli

# Project-level install
git clone https://github.com/Agents365-ai/agent-native-cli.git skills/agent-native-cli

# Install into Hermes engineering skills
git clone https://github.com/Agents365-ai/agent-native-cli.git ~/.hermes/skills/engineering/agent-native-cli

# pi-mono reads skills from any directory you register
git clone https://github.com/Agents365-ai/agent-native-cli.git ~/skills/agent-native-cli

# User-level install
git clone https://github.com/Agents365-ai/agent-native-cli.git ~/.agents/skills/agent-native-cli

# Project-level install
git clone https://github.com/Agents365-ai/agent-native-cli.git .agents/skills/agent-native-cli