glancelisted
Install: claude install-skill hir4ta/mumei
# Glance — one-line reliability summary
Surface the active feature's reliability in a single line so the user can check stability without context switching.
## Trigger
User invokes `/mumei:glance` (no args) or `/mumei:glance <feature>`. The skill is `disable-model-invocation: true`; it never fires on the model's own initiative.
## What it does
1. Resolves the target feature:
- no args → reads `.mumei/current` for the active feature key
- with args → uses the given feature key directly
2. Computes pass^3 over the most recent 10 trials via `hooks/_lib/reliability.sh`.
3. Renders **exactly one line** to stdout:
```text
<feature> | pass^3: <value-or-N/A> (n=<n_trials>, window=10, k=3)
```
4. Outputs `no active feature` to stdout (not stderr) and exits 0 when `.mumei/current` is missing or stale (REQ-25.2.3).
## How to invoke
Run the CLI implementation in `scripts/mumei-glance.sh` and print its stdout verbatim:
```bash
bash "$CLAUDE_PLUGIN_ROOT/scripts/mumei-glance.sh" "${1:-}"
```
The script always exits 0 (missing active feature is not an error). Do not reformat the output.
## Don'ts
- Don't add headers, blank lines, or trailing commentary — the contract is one line.
- Don't emit `no active feature` to stderr — REQ-25.2.3 specifies stdout.
- Don't accept flags or multiple feature args. A single optional positional arg only.
- Don't re-interpret the numeric value — it is a window pass rate, not a geometric Pass^k.