multi-llm

Multi-LLM

Call the multi_llm tool with the argument text, converted to its four properties exactly as that tool's own argument descriptions specify. The tool is the only supported entry point on this host: it appends the flags this host requires, refuses the ones it owns, and supervises the run to a terminal result. Do not reach the dispatcher any other way. The tool runs outside the exec sandbox, and on the modes that apply changes it writes to the workspace. A call can run for up to its 15-minute budget. When you call this tool from an exec cell, start the cell with // @exec: {"yield_time_ms": 900000}, so the cell waits for the result. If a cell hands back "Script running" anyway, call wait until this call returns, and never end your turn while it is running.

Read the result's exit and next the way the dispatcher's own contract reads them; on state: partial, name the failed models and their reasons from progress.models; when the result carries validation_counts (a review phase's does), report them, and if needs_human_decision or validation_failed is non-zero point at report.md for those groups; they are decided later, when the suggestions are applied. A result without that key has no counts to report, so do not mention them. When the result carries a notice (result.notice, a finished run's: what it left out or could not check, such as items it did not apply or that nothing ran this repository's tests), report what it says, every part of it, as plain text. Put it above the closing section, and repeat nothing that section says. On this host next arrives as arguments for this same tool, not as a command: "running" it means calling the tool again with exactly those arguments, never a shell. Run next verbatim when it is not null on 0 and 2. On 3 stop: report to the user which phase stopped, on which plan or commit range, the last error line it printed, whole (on a stop it says what stopped, what it reported, what did not run and whether anything ran the repository's tests), and the options its output names (next included); do not run a different mode. Do not perform the work this run was going to review, plan, or implement: doing it yourself is never a substitute for a failed run. Run an option only after the user has picked it. Never edit driver.write_paths, or any other multi-llm setting in .multi-llm/providers.yaml, yourself. At a stop, offer only the ways forward the stop itself names, as it words them: when it names a change to driver.write_paths that lets a refused file be written (adding it to allow, or removing or narrowing a deny rule that covers it), that is one of them and the user's call; add no advice about the write policy of your own. user_next_step is the phase after this one, for the user to type: never run it yourself; end your report with its markdown, copied as it is (its words and order, each command alone in its fenced block, each why and note plain text, the manual tasks it lists included, no caption of your own). Only that markdown says "Workflow complete.": never write those words yourself. Copy it word for word, never wrapped in a fenced block of your own (its commands are fenced already), and say none of it twice: its whys, notes, manual tasks and "Workflow complete." appear only in it, never above it. It is never a tool call, unlike next, which is one. At a stop, resume_markdown is the stop's options, already written: end your report with it the same way, in place of any list, recap or summary of the options or ways forward of your own.

If that tool is not listed, the plugin's server did not start: codex mcp get multi_llm names the command it tried, and python3 >= 3.9 must be on PATH.