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/apply-write-executor.md
View on GitHub# The Python write executor (apply modes)
Read this only when an apply core's command for the phase carries `--exec-subagents`,
`--detach` or `--dry-run-subagents`; a command with none of `--exec-subagents`,
`--detach` and `--dry-run-subagents` never needs this file.
## The loop it runs
The executor runs the core's batch loop in Python, in one process: batches in
ascending `batch_index`, each one's snapshot/diff classification completed and
recorded **before** the next is dispatched, the prior-batch context bound late per
dispatch (the literal `{prior_changes_context}` token never reaches a runner),
`prior_changes.py append` and `metrics.py record` per batch, and the phase marked
completed only after every batch verified. The runner is `driver.subagent_runner` in
`providers.yaml`; write concurrency is `driver.max_concurrent_write_subagents`
(default **1**), pinned to 1 on the two suggestion modes, whose batches all edit one
artifact. apply-code-fixes' batches are partitioned by file, so it may be raised
there, and even then two batches co-schedule only when their declared file sets are
provably disjoint.
It also does the core's closing steps: the outcome block names every item the judge
decided, `{phase}/summary.md` gets the sections of the core's *Generate Summary
Report*, the decisions are overlaid onto the review report, and the run prints step
3's breakdown itself (batches ok/failed with duration and runner, the count applied,
each skipped item with its status, group id and reason). On apply-code-fixes it runs
`driver.verify_command` once after the last batch (the core's step 7), records the
result in `summary.md` and `.status.json`, and ends the phase at exit 3 when it fails.
## Its flags
- `--dry-run-subagents`: full preflight, **zero dispatch**. It renders the exact
command per batch, applies the invocation profile, the co-scheduling admission
check and the `driver.write_paths` policy, writes `{phase}/subagent_dry_run.json`
and exits 0, having started no child and changed nothing else: report what that
file holds. The recommended first unattended run on a new host.
- `--detach`: the phase self-daemonizes; it is how step 2 runs on this path.
Per-batch progress lands in `{phase}/.status.json` under
`progress.write_subagents`, so an unattended run is observable with no chat
context. A host that resolves *run detached* to its own background mechanism uses
that and never passes `--detach`.
- Rollback, when a batch half-applies: `multi_llm.py util restore-batch --phase-dir
<dir> --task-id <id>` (or `--all-failed`), through the dispatcher like every mode.
It restores exactly the files `{phase}/subagent_audit.jsonl` recorded as touched,
from `{phase}/pre_batch/{task_id}/`, and **refuses** a file whose hash no longer
matches what the audit recorded at completion unless `--force` is passed. A partial
application never reverts itself: the run exits 3 naming this command, since a
revert would destroy the half-applied state a human needs to inspect.
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: 319k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 319k | 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: 319k | 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: 319k | Updated | Claude CodeCodex | Verified | Install |