OmniCell

OmniCell transforms the Operating System into a structured data source. Instead of parsing raw text from stdout, you receive JSON objects.

Tools

nuShell

Universal shell tool – use for ANY terminal command.

Parameters:

• command (string): Any terminal command (auto-detects observe vs mutate)
  • Examples: ls -la, cargo test, git status, npm run build
  • Read: open config.json (Returns parsed JSON/Dict directly)
  • List: ls **/*.py | sort-by size (Returns List[Dict])
  • Query: ps | where cpu > 10
• intent (string, optional): Explicitly force observe or mutate
• chunked (bool, default false): Enable chunked delivery for very large payloads
• action (string, optional): start or batch
• session_id (string): Required for action=batch
• batch_index (int, default 0): Required batch index for action=batch
• batch_size (int, default 28000): Character window per batch

Usage:

<tool_call>{"name": "omniCell.nuShell", "arguments": {"command": "cargo test"}}</tool_call>

Chunked mode (large output):

<tool_call>{"name":"omniCell.nuShell","arguments":{"command":"ls **/*","chunked":true}}</tool_call>

<tool_call>{"name":"omniCell.nuShell","arguments":{"action":"batch","session_id":"abc123","batch_index":1}}</tool_call>

Best Practices

1. Structured Data First: Always prefer open over cat. OmniCell automatically parses JSON, YAML, TOML, XML, and CSV into Python dictionaries.
2. Pipelines: Use Nu pipes (|) to filter data before it reaches the LLM context.
   • Bad: ls -R (returns huge text block)
   • Good: ls **/* | where size > 1mb | to json (returns clean data)
3. Large Output: For huge results, prefer chunked=true and read all batch_index values instead of truncating content.