Problem
LLMs using Mollotov (via CLI or MCP) don't know they should report failures back. When a click misses, a selector fails, or a feature is missing, the LLM just retries or tells the user it can't do it. That failure data is lost.
Proposal
Add a standard instruction block to the LLM-facing help surfaces:
1. In --llm-help output (global preamble)
When you encounter a failure or limitation while using Mollotov, report it:
- If a click/tap misses the target, note the selector, coordinates, and what you expected
- If a feature is missing that would have helped complete the user's request, describe it
- If platform behavior is inconsistent, note which platform and what happened
- File issues at: https://github.com/UnlikeOtherAI/mollotov/issues
2. In MCP server description
The MCP server init should include tool-use instructions that encourage reporting.
3. In CLI help footer
mollotov --help should include a line like:
Feedback: Report issues and suggestions at https://github.com/UnlikeOtherAI/mollotov/issues
4. In CLAUDE.md / agent instructions
Projects using Mollotov as an MCP tool should add to their agent config:
When Mollotov commands fail or produce unexpected results, create a GitHub issue
at UnlikeOtherAI/mollotov with: the command, parameters, error, and what you expected.
Check for duplicates first.
Goal
Turn every LLM session into a QA pass. The more LLMs report, the faster we fix the things that actually block real workflows.
Problem
LLMs using Mollotov (via CLI or MCP) don't know they should report failures back. When a click misses, a selector fails, or a feature is missing, the LLM just retries or tells the user it can't do it. That failure data is lost.
Proposal
Add a standard instruction block to the LLM-facing help surfaces:
1. In
--llm-helpoutput (global preamble)2. In MCP server description
The MCP server init should include tool-use instructions that encourage reporting.
3. In CLI help footer
mollotov --helpshould include a line like:4. In CLAUDE.md / agent instructions
Projects using Mollotov as an MCP tool should add to their agent config:
Goal
Turn every LLM session into a QA pass. The more LLMs report, the faster we fix the things that actually block real workflows.