Cohub Cli

Cohub CLI Skill

Overview

Cohub is a cloud platform by Neta Art for space-based agent creation, execution, and collaboration. The @neta-art/cohub-cli is the official CLI client for working with Spaces, Chats (Sessions), files, Saves (Checkpoints), Tasks, scheduled prompts, search, and multimodal generation from the terminal.

Origin: Neta Art (捏Ta) — https://www.neta.art Package: @neta-art/cohub-cli on npm Auth: cohub auth login (Logto-based, auto-refreshing) Env overrides: ENV=dev for dev environment, COHUB_EXECUTION_TOKEN for CI/sandbox

Core Concepts

| Concept | Description | |---|---| | Space | Live isolated working environment; the primary creative surface | | Session | LLM conversation context inside a space (UI calls it "Chat") | | Checkpoint | Immutable snapshot of a space state (UI calls it "Save") | | Task run | Record of an executed background task | | Cron job | Recurring scheduled prompt |

Command Patterns

Global flags

-s, --space      # Target space ID
--json               # Machine-readable output

Prompts (the primary interaction)

# Immediate send
cohub -s  prompt "message" --json

# Existing session
cohub -s  prompt --session  "message" --json

# New session with title
cohub -s  prompt --title "Chat title" "message" --json

# Pipe long content
cat prompt.md | cohub -s  prompt --json

# Schedule once
cohub -s  prompt --at "2026-05-22T09:00:00+08:00" "message" --json

# Recurring
cohub -s  prompt \
  --cron "0 9 * * 1-5" \
  --timezone "Asia/Shanghai" \
  --title "Daily reminder" \
  "message" \
  --json

Scheduling constraint: use only one of --delay-ms, --at, or --cron. --cron requires --timezone.

Sessions

cohub -s  spaces sessions ls --json
cohub -s  spaces sessions create "title" --json
cohub -s  spaces sessions get  --json
cohub -s  spaces sessions turns ls   # list messages

Files

cohub -s  spaces files ls [path] --json
cohub -s  spaces files cat 
cohub -s  spaces files write  -c "content"
cohub -s  spaces files upload  --dir 
cohub -s  spaces files mv  
cohub -s  spaces files rm 

Checkpoints

cohub -s  spaces checkpoints ls --json
cohub -s  spaces checkpoints get  --json
cohub -s  spaces checkpoints create "description" --json

Search

cohub search "query" --limit 20 --json

Models & Generation

cohub models ls --json
cohub models ls --model-type multimodal --json
cohub generate "prompt" --model  --output file.png --json
cohub generate "restyle" --model  --image ./input.png --param size=1024x1024 --json

Cron Jobs (recurring scheduled prompts)

cohub cron-jobs ls  --json
cohub cron-jobs runs  --json
cohub cron-jobs toggle  on|off
cohub cron-jobs delete 

Tasks

cohub tasks ls --space  --json
cohub tasks get  --json

Spaces management

cohub spaces ls --json
cohub spaces get  --json
cohub spaces create --name "name" --description "desc" --json
cohub spaces rename  "new name"

Available Models (as of 2026-05)

gpt-5.5, ultimate, performance, qwen3.7-max, qwen3.6-plus, deepseek-v4-pro, kimi-k2.5, glm-5.1, gpt-5.4-mini, doubao-2.0-lite, doubao-2.0-pro, qwen3.6-flash, fm/gpt-5.5, bailian/glm-5

Safety Rules

Confirm before:

• Deleting files or directories
• Creating scheduled or recurring prompts with side effects
• Enabling, disabling, or deleting recurring scheduled prompts
• Changing access policies, member roles, or membership

Efficient Usage Guidelines

1. Use --json for scripting — table output is human-friendly; JSON is for extraction and chaining.
2. Discover via help — cohub -h, cohub spaces -h, cohub spaces prompt -h. The CLI's built-in help is comprehensive and well-designed.
3. Space ID once — set -s early; most subsequent commands inherit it.
4. Session continuity — reuse --session to continue a conversation rather than creating new sessions.
5. Checkpoints for milestones — save checkpoints before risky operations or at meaningful milestones.
6. Piping — use cat file.md | cohub -s prompt for large prompts instead of inline strings.

Prompt Best Practices

Wrap text to avoid shell escaping

Multi-line prompts with code blocks, JSON, or special characters ($, backticks, quotes) get mangled by shell eval. Write to a file first:

cat > /tmp/prompt.txt  prompt --session  "$(cat /tmp/prompt.txt)" --json

Choose result handling

| Approach | When to use | Command | |---|---|---| | Fire and forget | User watches in Web UI | Send prompt, stop. No polling. | | Poll for result | Need output in this session | sleep 15 && cohub spaces sessions turns ls | | Check later | Long-running task | Note the turnId, check manually later. |

Default: After sending a prompt, ask the user "Do you want me to wait for the result, or will you check in the Web UI?" If the user says "我看就行" or similar, do not poll.

Session continuity

For multi-step tasks, always reuse the same --session so the Agent retains context across turns. Creating a new session per prompt loses state.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kjx-talesofai
• Source: kjx-talesofai/claude-skill-cohub-cli
• License: MIT

Install and usage instructions live in the source repository linked above.