chore: agent-context command

[notnmeyer]

generates a json doc of all commands, args, and flags. an agent can use this command to get the complete use.

then, in theory, the cli skill can focus on higher-level workflows rather than the specific cli use.

an example filtered with jq so you can get the gist,

➜ loops agent-context | jq '.commands[] | select(.name == "transactional") | .commands[] | select(.name == "list")'
{
  "name": "list",
  "path": [
    "transactional",
    "list"
  ],
  "use": "list",
  "short": "List published transactional emails",
  "long": "",
  "hidden": false,
  "runnable": true,
  "args": [],
  "flags": [
    {
      "name": "cursor",
      "shorthand": "",
      "type": "string",
      "default": "",
      "description": "Pagination cursor for a specific page",
      "required": false,
      "persistent": false
    },
    {
      "name": "per-page",
      "shorthand": "",
      "type": "string",
      "default": "",
      "description": "Results per page (10-50, default 20)",
      "required": false,
      "persistent": false
    },
    {
      "name": "pick",
      "shorthand": "",
      "type": "bool",
      "default": "false",
      "description": "Interactively pick a row with fzf",
      "required": false,
      "persistent": false
    }
  ],
  "flagGroups": {
    "oneRequired": [],
    "mutuallyExclusive": [],
    "requiredTogether": []
  },
  "commands": []
}