Auto-invokedCommand
deadeye-mute
Mute deadeye's advisories, plan-gate nags, and workflow hints for this session
@deepaksinghcs14View CommandDevelopment
/deadeye-route
Dry-run the routing decision for a task, with reasoning
Install
> /plugin marketplace add deepaksinghcs14/deadeye-cc > /plugin install deadeye@deadeye
How it fires
How this command gets triggered: by you, by Claude, or both.
- Fires itselfClaude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/deadeye-route
Context preview
What this command does when you run it.
Dry-run the routing decision for a task, with reasoning
Command definition
deadeye-route.mddescription: Dry-run the routing decision for a task, with reasoning
argument-hint: "[task description]"
allowed-tools: Bash(deadeye route:*), Bash(~/.deadeye/bin/deadeye route:*)
Run `deadeye route "$ARGUMENTS"` (or `deadeye route` with no arguments, to dry-run against the current working tree's modified/staged files instead of a task description). If that reports "command not found", it's very likely just not on PATH -- deadeye never adds itself to PATH, it only resolves its own binary internally for hook invocations. Retry with `~/.deadeye/bin/deadeye route ...` before concluding it's missing.
Present the output as-is: the evidence each signal provider contributed (or "none" if every provider had nothing to go on), and the resulting Decision (model, effort, confidence, threshold, reason). A threshold above the configured base means recorded escalations for this task shape raised the bar -- the real Agent call uses the same adjusted number.
This changes no state, but it is not free: with `mode.routing_judge=on` (the default) an unsure decision spends one real `claude -p` classification, the same one a live Agent call would. Say so if the user is counting spend.
If evidence is empty, explain that this is expected and correct, not a bug -- per INV-1, missing evidence must default to the most conservative decision (the ceiling), never a cheaper one.
This command only reports what the kernel *would* decide -- it never applies anything.
Read more
description: Dry-run the routing decision for a task, with reasoning argument-hint: "[task description]" allowed-tools: Bash(deadeye route:*), Bash(~/.deadeye/bin/deadeye route:*)
Run `deadeye route "$ARGUMENTS"` (or `deadeye route` with no arguments, to dry-run against the current working tree's modified/staged files instead of a task description). If that reports "command not found", it's very likely just not on PATH -- deadeye never adds itself to PATH, it only resolves its own binary internally for hook invocations. Retry with `~/.deadeye/bin/deadeye route ...` before concluding it's missing.
Present the output as-is: the evidence each signal provider contributed (or "none" if every provider had nothing to go on), and the resulting Decision (model, effort, confidence, threshold, reason). A threshold above the configured base means recorded escalations for this task shape raised the bar -- the real Agent call uses the same adjusted number.
This changes no state, but it is not free: with `mode.routing_judge=on` (the default) an unsure decision spends one real `claude -p` classification, the same one a live Agent call would. Say so if the user is counting spend.
If evidence is empty, explain that this is expected and correct, not a bug -- per INV-1, missing evidence must default to the most conservative decision (the ceiling), never a cheaper one.
This command only reports what the kernel *would* decide -- it never applies anything.
Ships withdeadeye
Claude Code plugin that fits the model, effort, and context to each task — fewer tokens, same quality. Deterministic policy kernel in the hooks; every number it reports is measured, not estimated.
Get the whole plugin, auto-invoked
Other commands on deadeye.
@deepaksinghcs14View CommandAuto-invokedCommand