Invokellm

Invoke LLM

Read the shared ../../references/gptengage-invocation.md contract and the selected ../../references/gptengage-invoke.md recipe before calling a backend.

Workflow

1. Parse the CLI selector, prompt, and supported options using the operation recipe. Treat the first token as a selector only when it names a known CLI, comma-separated CLI list, or plugin.
2. Default to gemini, claude, then codex when no selector is present. Invoke each separately with the identical prompt and applicable options.
3. Resolve the authorized existing gptengage executable and run only its invoke operation. Default the inner --timeout to 600 seconds unless the user supplied one.
4. Label multi-CLI results by backend, inspect failures and malformed output, and integrate the results into the user's requested work.

Boundaries

• One requested CLI suppresses the default trio; never silently substitute a different backend.
• A session persists full conversation turns and must be deliberately named.
• --write requires explicit write intent and grants no remote-write authority.
• Never change the prompt merely to make backends agree.