GitHub - aiagenta2z/agtm: Official AI Agent Registry CLI Tool to manage help Agent Cli,Skills,MCP and Agent Meta

agtm: CLI Tool for AI Agent Management, Skills, Agent Registry, Benchmarks and Hints in AI Agent Marketplace

agtm (AI Agent Manager CLI) unifies skill management, agent registration, marketplace search, and provider CLI execution. Install skills from GitHub, log and rate skill runs, upload agent metadata to registries, query the public marketplace, and run agent toolchains with fuzzy hints.

Features

agtm skills: Manage Skills, Add Skills, List Skills, Log Skills Performance, Skills performance Evaluator, compare to realworld benchmarks
agtm upload: AI Agent Registry,Agent Meta,APIs to Skills/MCPs/CLIs,register local agent meta information of json or yaml format(agent.json/agent.yaml) or sync your github source meta including README.md
agtm search: Search the open source AI Agent Marketplace, including github community, huggingface community, product hunt community, deepnlp ai agent marketplace index, etc
agtm run: Run agent clis, don't need to remember, with the powerful hints and completion ability, just type a few characters and "--hint" will help you complete the command line.

Live Update

20260407: Agtm Hints now works with any npm-registered package, automatically converting CLI help output into structured, highlighted hints—e.g., for example @larksuite/cli, you can quickly find "message" related clis using "agtm run @larksuite/cli message" and it will guide you to the clis, such as "lark-cli im messages-reply"

The agent cli run hint from "play" to "playwright" complete cli.

Furthermore, agtm provides memory to track skill outputs and enables performance rating against industry job level benchmarks. This allows you to score each skill execution and assign a professional tier to your AI Agent's capabilities—for example, evaluating its performance as equivalent to that of an L3 or L5 software engineer, marketing professional, etc.

skill_id             run_times  score  level
-------------------  ---------  -----  -----
code_success_skills  5          0.9     L3(100%)    
code_fail_skills  4          0.9     L3(100%)

Quickstart

Installation

Node

npm install -g @aiagenta2z/agtm

Run a playwright Webpage cli

Setup hint and skills benchmark

npx agtm setup --levels  ## Needed before `agtm rate`, to sync the benchmarks json to local folder 
npx agtm setup --hint    ## Needed before `agtm run`

npm install -g @playwright/cli@latest
agtm run play
### hint to "playwright-cli goto" command line

Result Hint

Selected Skill/Cli is microsoft/playwright-cli

Enter command to run (leave empty to list cli hints): play

Complete the Cli with your arguments or leave blank and press Enter

Final command line [playwright-cli goto <url>]:
playwright-cli goto https://www.google.com
agtm run microsoft/playwright-cli playwright-cli goto https://www.google.com

### Browser `default` opened with pid 4603.
- default:
  - browser-type: chrome
  - user-data-dir: <in-memory>
  - headed: false
---

### Ran Playwright code
await page.goto('https://www.google.com');


### Page
- Page URL: https://www.google.com/
- Page Title: Google
### Snapshot
- [Snapshot](.playwright-cli/page-2026-03-22T03-08-05-614Z.yml)

Agtm CLI Options

CLI	Command and Options	Document
agtm skills	add, list, build, log, rate	Doc
agtm search	--q query	Doc
agtm upload	--github --config to local agent meta	Doc
agtm run	--hint agent-cli hint and auto completion	Doc

`skills`

skills add

Download and add skills to your agent directory.

Usage

npx agtm skills add <unique_id>
npx agtm skills add <github_url>
npx agtm skills add <github_url> -a <agent_id>
npx agtm skills add <owner_id/repo_id> -s <skill_id>

Example

npx agtm skills add anthropics/skills -a claude-code  ## install skills only to claude-codex
npx agtm skills add msitarzewski/agency-agents
npx agtm skills add aiagenta2z/onekey-gateway
npx agtm skills add msitarzewski/agency-agents -s academic-anthropologist -a codex
npx agtm skills add anthropics/skills -s skill-creator -a claude-code --global

skills list

Usage

npx agtm skills list
npx agtm skills list --agent <agent_id> --global

Lists installed skills with agent, skill_id, description, install path, average score, and aggregated level (if ratings exist).

Example

npx agtm skills list
npx agtm skills list --agent codex
npx agtm skills list --agent claude-code --global

skills build

Convert registered Agentic API to skills easily. Before running the build command, make sure the unique_id agent and API meta infos are registered in AI Agent Marketplace using agtm upload.

Usage Example

agtm skills build <unique_id>

# build skills for food calories app
agtm skills build fdcnal/usda-fooddata-central-agent
# or 
agtm skills build aiagenta2z/financeagent

skills log

Usage

npx agtm skills log <skill_id> --data '<json_payload>'

Persists a run record at .agtm/skills/log/<uuid>.json (or the --logDir you supply).
<json_payload> must contain at least input and output; optional fields (meta, rating, level) are accepted.

Example

npx agtm skills log <skill_id> --data '{"input":"write a website for store","output":"success"}'
npx agtm skills log code_success_skills --data '{"input":"generate sql","output":"ok","meta":{"agent":"claude-code"}}'

skills rate

Setup

To use the rate command, have to setup the benchmark levels configuration. save to ./agtm/levels/*.json files

npx agtm setup --levels

Usage

npx agtm skills rate prepare --skill_id <skill_id> --prompt "<eval_prompt>" --benchmark <path/benchmark.json>
npx agtm skills rate apply   --skill_id <skill_id> --result '<result_json>'
npx agtm skills rate show    --skill_id <skill_id>

prepare exports logs plus the top benchmark slices (e.g., Google SWE L3–L7) for an external evaluator.
apply writes evaluator outputs (rating, level) back to each log.
show summarizes run counts, average score, and level distribution.

Example

npx agtm skills rate prepare --skill_id code_success_skills --prompt "Evaluate the results" --benchmark path/customized_agent_benchmark.json
npx agtm skills rate apply --skill_id code_success_skills --result '{"results":[{"log_id":"3679a3fe-4d97-4eb1-83bc-f83d711be195","rating":0.90,"level":"L4"}]}'
npx agtm skills rate show --skill_id code_success_skills

Sample output:

skill_id             run_times  score  level
-------------------  ---------  -----  -----
code_success_skills  3          0.88   L4(100%)

Benchmarks live in data/config/levels/*.json and follow this structure:

{
  "software-engineering": {
    "Google": [
      { "level": "L3", "title": "Software Engineer II", "description": "Entry-level engineer. Delivers well-scoped tasks with guidance.", "signals": ["task execution","learning velocity","code quality basics"] },
      { "level": "L4", "title": "Software Engineer III", "description": "Independent contributor. Owns small features end-to-end.", "signals": ["ownership","code quality","debugging ability"] }
    ]
  }
}

The evaluator compares each <input,output> log to the benchmark definitions and assigns a rating. You can customize levels such as poor, fair, good, excellent in your benchmark file.

Partial List of Skills and Job Level Description

Job Category	Company	Levels Description (e.g. Google L3, Meta E4)
Marketing	Apple	marketing_sales_apple_levels.json
Software Engineer	Google	software_engineer_google_levels.json
Software Engineer	Meta	software_engineer_meta_levels.json

write your customized_agent_benchmark.json following the formats

{
  "domain": {
    "my_benchmark": [
      { "level": "poor", "description": "the skill failed or the output is meaningless"},
      { "level": "fair", "description": "the skills produces fair results, complete the task"},
      { "level": "good", "description": "the skills output a report, an image is good in real life standards.."},
    ]
  }
}

`Run`

Execution of Agent Cli with hints and auto completion.

The run command executes agent workflows with interactive hints and fuzzy CLI completion. Typing a few characters surfaces suggested commands so you can finish the full invocation quickly (for example, typing play will suggest the Playwright runner).

Let's say you want to run an agent command of Playwright to go to a URL and fetch a webpage. You don't need to remember the full command—type play, pick the provider, then pick the CLI action.

Usage

Remember to setup hint before running the agent-cli

agtm setup --hint

agtm run <provider_unique_id> <agent_cli>

Cli Examples

agtm run play
Skill ID suggestions:
  1. microsoft/playwright-cli
  2. googleworkspace/cli
Skill ID suggestions: 1
Command hints:
  1. playwright-cli goto <url> # navigate to a url
  2. playwright-cli open [url] # open browser, optionally navigate to url

Select command (number or input custom): 1
Final command [playwright-cli goto <url>]: playwright-cli goto https://www.github.com
agtm run microsoft/playwright-cli playwright-cli goto https://www.github.com

Support CLI List, Please welcome to contrib

unique_id	agent cli
microsoft/playwright-cli	playwright-cli open [url], playwright-cli goto
googleworkspace/cli	gws drive files list --params
aiagenta2z/onekey-gateway	onekey agent <unique_id> <api_id> <data_json
openai/codex-cli	codex, codex exec "[instruction]"
anthropic/claude-code	claude "[prompt]", claude --dangerously-skip-permissions
paul-gauthier/aider	aider --model [model_name], /test [command]
openinterpreter/open-interpreter	interpreter, interpreter --os
google-gemini/gemini-cli	gemini, gemini -p "[prompt] @[file/dir]", gemini --yolo, /memory add "[fact]", /mcp list, /restore

npmjs registered packages

Agtm Hints now works with any npm-registered package, automatically converting CLI help output into structured, highlighted hints—e.g., for example @larksuite/cli.

## hints on all the commands of the package
agtm run @larksuite/cli

## Search all the message related clis 
agtm run @larksuite/cli message

Result

npx agtm run @larksuite/cli message 

INFO: Updating CLI Hints for Package @larksuite/cli...

Command hints:
  1. lark-cli im messages-resources-download Download images/files from a message; user/bot; downloads image/file resources by message-id and file-key to a safe relative output path
  2. lark-cli mail user_mailbox.message.attachments user_mailbox.message.attachments operations
  3. lark-cli im messages-search # Search messages across chats (supports keyword, sender, time range filters) with user identity; user-only; filters by chat/sender/attachment/time, enriches results via mget and chats batch_query
  4. lark-cli im messages-reply # Reply to a message (supports thread replies) with bot identity; bot-only; supports text/markdown/post/media replies, reply-in-thread, idempotency key
  5. lark-cli im messages-mget # Batch get messages by IDs; user/bot; fetches up to 50 om_ message IDs, formats sender names, expands thread replies

AI Agent Registry

`search`

agtm search helps to search AI Agent MCP and skills marketplace by id or query keywords.

Example Usage

agtm search --q 'coding agent'
agtm search --id 'google-maps/google-maps'

`upload`

agtm update helps to update your local agent.json or agent.yaml meta information to DeepNLP AI Agent Marketplace Index. Example Usage

agtm upload --github https://github.com/AI-Hub-Admin/My-First-AI-Coding-Agent
agtm upload --config ./agent.json
agtm upload --config ./agent.yaml

schema.json should have two keys defined the required fields and optional fields you want to submit from the agent.json file.

Remember to keep the access_key in safe place, the post request will post the access_key as well as schema to the endpoint.

Please visit the command line github package agtm and DOC detailed usage.

Use the test account and access

export AI_AGENT_MARKETPLACE_ACCESS_KEY="TEST_KEY_AI_AGENT_REGISTRY"

agtm upload --config ./agent.json --endpoint https://www.deepnlp.org/api/ai_agent_marketplace/registry --schema ./schema.json

agtm upload --config ./agent.json --endpoint https://www.aiagenta2z.com/api/ai_agent_marketplace/registry --schema ./schema.json

Skills Agtm-Cli

We provide Skills repo to use in various agents to evaluate skills and run agent hints. The skills can be found in ./skills/ folder

skill	description
agent-cli-hint-completion	This skill uses `agtm run --mode agent` to help hint agents clis usage
agent-skills-evaluator	This skill use `agtm skills log` and `agtm skills rate` to track other skills performance from LLM-based evaluator, match to professional Job Level Benchmarks, such as Google L3 level software engineers/ Apple M3 level marketing specialist performance.

npx agtm skills add aiagenta2z/agtm ## install all the skill evaluation and skill cli-hints
npx agtm skills add aiagenta2z/agtm -s agent-skills-evaluator

npx skills add aiagenta2z/agtm ## install all the skill evaluation and skill cli-hints
npx skills add aiagenta2z/agtm -s agent-skills-evaluator

Contributing

Agent CLI List

You are welcome to contrib your cli list to the agent cli hints json file to folder hints or table README.md

Skill Run Benchmark

You are welcome to contrib your own customized benchmarks of skills and unique levels system in folder levels.

Name		Name	Last commit message	Last commit date
Latest commit History 15 Commits
data		data
docs		docs
lib		lib
skills		skills
.gitignore		.gitignore
README.md		README.md
agent.json		agent.json
agent.yaml		agent.yaml
agent_api.json		agent_api.json
agent_api.yaml		agent_api.yaml
agtm-cli.ts		agtm-cli.ts
package-lock.json		package-lock.json
package.json		package.json
schema.json		schema.json
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

agtm: CLI Tool for AI Agent Management, Skills, Agent Registry, Benchmarks and Hints in AI Agent Marketplace

Features

Live Update

Quickstart

Installation

Run a playwright Webpage cli

Agtm CLI Options

skills

skills add

Usage

Example

skills list

Usage

Example

skills build

Usage Example

skills log

Usage

Example

skills rate

Setup

Usage

Example

Run

Usage

Cli Examples

npmjs registered packages

AI Agent Registry

search

upload

Skills Agtm-Cli

Contributing

Agent CLI List

Skill Run Benchmark

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`skills`

`Run`

`search`

`upload`

Packages