Problem
mollotov --llm-help outputs JSON metadata for each command, but several pieces of information are missing that LLMs need to work effectively:
Missing: Error code descriptions
Currently only error codes are listed (e.g., ELEMENT_NOT_FOUND). The descriptions exist in docs/api/README.md but aren't integrated into --llm-help output.
Missing: Default values for optional parameters
Optional params like quality (screenshot), timeout (wait), format (screenshot) have defaults that aren't shown. LLMs don't know what happens if they omit a param.
Missing: Enum values
Params like dialog defaultAction show type as "enum" but don't list the valid values (accept, dismiss, queue). LLMs have to guess.
Missing: Response structure
LLMs don't know what fields to expect in a successful response. For example, screenshot returns { image, format, width, height } but this isn't documented in help.
Missing: Complex parameter shapes
Array-of-object parameters (like request interception rules) show type as "array" with no indication of the nested object structure.
Proposed changes to llm-help.ts
{
"params": [
{
"name": "format",
"type": "enum",
"values": ["png", "jpeg"],
"default": "png",
"required": false,
"description": "Image format"
}
],
"errors": [
{ "code": "NO_WEBVIEW", "description": "Browser not loaded yet" }
],
"response": {
"image": "base64-encoded image data",
"format": "png or jpeg",
"width": "image width in pixels",
"height": "image height in pixels"
}
}
Problem
mollotov --llm-helpoutputs JSON metadata for each command, but several pieces of information are missing that LLMs need to work effectively:Missing: Error code descriptions
Currently only error codes are listed (e.g.,
ELEMENT_NOT_FOUND). The descriptions exist indocs/api/README.mdbut aren't integrated into--llm-helpoutput.Missing: Default values for optional parameters
Optional params like
quality(screenshot),timeout(wait),format(screenshot) have defaults that aren't shown. LLMs don't know what happens if they omit a param.Missing: Enum values
Params like dialog
defaultActionshow type as "enum" but don't list the valid values (accept,dismiss,queue). LLMs have to guess.Missing: Response structure
LLMs don't know what fields to expect in a successful response. For example,
screenshotreturns{ image, format, width, height }but this isn't documented in help.Missing: Complex parameter shapes
Array-of-object parameters (like request interception rules) show type as "array" with no indication of the nested object structure.
Proposed changes to
llm-help.ts{ "params": [ { "name": "format", "type": "enum", "values": ["png", "jpeg"], "default": "png", "required": false, "description": "Image format" } ], "errors": [ { "code": "NO_WEBVIEW", "description": "Browser not loaded yet" } ], "response": { "image": "base64-encoded image data", "format": "png or jpeg", "width": "image width in pixels", "height": "image height in pixels" } }