hashgraph-online/awesome-codex-plugins / plugins/beastlabai/multi-llm-plugin/skills/multi-llm
multi-llm
FlaggedOrchestrate plan reviews, task generation, implementation and code reviews across several LLM CLIs in parallel, or ask every model one free-text question about a plan. Use when the user says: review my plan, multi-llm, run a code review, generate tasks from a plan, implement this plan, ask the models about my plan.
references/interrupted-run-recovery.md
View on GitHub# Interrupted-run Recovery (review fan-outs)
Read this only when a review core sends you here: its detached orchestrator was
interrupted (the session ended, the background process was killed), or the poll
loop ended and `{plan}/{phase}/orchestrator-run.log` shows neither a validation
marker nor a clean completion. `{phase}` is the phase directory the core names
(`review-plan`, `review-tasks`). A run that was not interrupted and whose
`orchestrator-run.log` shows a validation marker or a clean completion never
needs this file.
1. Read `{plan}/{phase}/.status.json`.
2. Determine progress from its `state` field:
- `models_running`: check which `{model}.json` files exist in `{plan}/{phase}/`
(those models completed).
- `models_complete` / `grouping_complete`: check for `grouped.json` and
`validation_tasks.json`.
- `validation_pending`: the orchestrator completed normally; go on to the core's
validation step using `validation_tasks_path` from the status file, then
reaggregate. Both the status file and the tasks file record the mode and plan
path (`reaggregate_mode`, `plan_file`) rather than a resolved command, so a
plugin update cannot leave a stale skill-dir path behind; `multi_llm.py status
<plan> --view run` prints the command resolved against the running skill dir.
- `complete`: report results.
3. With no `.status.json`, fall back to the file-based detection in
`references/resume-detection.md` (the core's resume-detection step).
4. **With `--yes` (or `--non-interactive`): always resume, never ask** (D6 row 3).
Re-run with `--force` and the original run's full `--models` selection
(`models_requested` in `.status.json`). Only the missing models actually run:
the retry prints `[SKIP] <model> - already has results` for each model it keeps,
so tell the user which model re-ran and which kept its result. Never
`--rerun-all`: it discards work that succeeded. If a model fails a second time
in the same run, reaggregate with what exists: the orchestrator (the fan-out or
`--reaggregate` alike) prints `[PARTIAL_RESULTS: <model>,…]`, names the failed
models and their reasons on the report's `**Models failed:**` line and in
`.status.json` (`models_failed`, and the `failed` rows of `progress.models`),
sets `MULTI_LLM_RESULT.state: "partial"` and exits **0**: a partial aggregate is
usable output, the phase is recorded complete and `next` stays `null`. The one
exception: if *no* model produced a result there is nothing to aggregate, so it
exits **3** naming the retry command (a `--reaggregate` does so when the
checkpoint records failed models) and records no completed phase.
5. Without `--yes`, **ask the user**, showing:
- which models completed successfully;
- which models failed or were interrupted;
- the options: (a) proceed with partial results (`--reaggregate`), (b) resume,
re-running only the missing models (`--force` with the original full
`--models` selection), (c) re-run everything from scratch
(`--force --rerun-all`).
Then return to the core's next step (salvage, then validation).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |