LLM Supervisor
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill mostly matches its fallback purpose, but it can automatically and persistently change which model/provider agents use, including a hard-coded cloud profile, with incomplete confirmation safeguards.
Install only if you want this skill to control agent LLM profiles. Before using it, verify the cloud profile, local model, confirmation phrase, and reset behavior; use `/llm status` and `/llm switch cloud` after rate limits, and avoid it for provider-restricted or highly sensitive work until the hard-coded profile and persistence behavior are corrected.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
After a cloud rate-limit event, later work may run on a local model automatically, changing output quality and model behavior without a fresh opt-in at the moment of switch.
A rate-limit-like error immediately changes the active LLM mode and only notifies afterward; this is broader than the SKILL.md wording that says to offer fallback and wait for confirmation.
setState(ctx, {
mode: "local",
since: Date.now(),
lastError: event.error?.message
});Make automatic fallback an explicit opt-in, or ask before changing the active profile; preserve the previous profile and provide a clear one-step rollback.
Prompts and costs may be routed through an Anthropic default account/profile even if the user expected another cloud provider or policy-controlled profile.
In cloud mode the skill forces a specific delegated provider profile instead of preserving the user's configured cloud provider.
event.agent.setLLMProfile("anthropic:default");Use an explicit configurable `cloudProfile`, preserve and restore the user's prior profile, and display the provider before switching.
One rate-limit event or manual switch can affect later agents/tasks until someone notices and switches back.
The selected mode is saved in skill state without an expiry or reset in the provided code, and other hooks reuse that mode for future agents.
return ctx.state.get<LLMState>("llm-supervisor:state") ?? DEFAULT_STATE;
...
ctx.state.set("llm-supervisor:state", state);Scope the mode to the current session or task, implement the documented reset-to-cloud behavior, and add a cooldown/expiry for fallback mode.
The advertised local-code confirmation may be confusing or nonfunctional if the platform does not supply this missing config elsewhere.
The confirmation prompt and check depend on a config key that is not present in the provided `skill.json` defaults, which only declares `requireConfirmationForCode`.
const confirmationPhrase = ctx.config.confirmationPhrase;
...
.includes(confirmationPhrase);
...
`**${confirmationPhrase}**`Declare a default confirmation phrase, normalize casing consistently, honor the configured confirmation flag, and test the block/confirm path.