Hush stats

Skill V-Songbird/hush/skills/hush-stats

🤫 Token-lean sessions at the harness level. An easy to grasp output style, output-shrinking hooks, and log compression cut both input and output tokens.

Install

npx -y skills add V-Songbird/hush --skill hush-stats

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

29 days oldThe repository was created 29 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.

What its author says it does

Copied from the file, not written here

Reports how much hush's compression hook actually saved this session — a per-decision byte rollup from the HUSH_DEBUG manifest plus a per-model token breakdown from the transcript. Requires HUSH_DEBUG=1 to have been set during the work being measured; without it there is nothing to report.

SKILL.md

2.0 KB, as published. Nobody here has run it

hush:hush-stats

Surfaces the value hush's compression hook already recorded but never showed anyone: a byte rollup of every decision it made this session, and — from the transcript itself — how many tokens each model actually used.

1. Run the script

node "${CLAUDE_PLUGIN_ROOT}/scripts/stats.js"

If the user gave a session id as an argument, pass it through: node "${CLAUDE_PLUGIN_ROOT}/scripts/stats.js" --session <id>.

The script auto-finds the current directory's most recent transcript when no id is given. It never modifies anything — read-only over the manifest and the transcript.

2. If it reports no manifest found

This means HUSH_DEBUG=1 wasn't set while the work being asked about happened — the manifest only exists when that flag was on (it's off by default; see the README's Settings table). Tell the user plainly: set HUSH_DEBUG=1 before the session (or turn) they want measured, then ask again. Do not estimate or infer savings from anything else — a guess dressed as a number is worse than admitting there's nothing to report.

3. Report

State the outcome in plain sentences, not a wall of labels: total bytes saved (raw and net of hush's own one-time session note, if it fired), the one or two actions that did most of the work (e.g. "cap" or "sidecar"), and the per-model token counts if a transcript was found. Skip any section the script reported as unavailable rather than padding around it.