| name | second-brain-stats |
| description | Show what the Second Brain has observed, spent and said — observed volume per segment, pass latency, window fill, token use and cost, advisories generated / gated / delivered, and uptake per detector. Use when the user asks whether the Second Brain is watching, what it has cost, or whether its advice is being taken. |
Second Brain — Stats
The Second Brain is the second-brain plugin's background observer: one cheap
model per root task, watching the task's emissions and occasionally injecting a
one-way gated advisory. Silence is its success metric.
To answer, call the plugin request stats (plugin second-brain) and relay the
result readably:
- consent / muted — whether it can run at all, and whether the user silenced it.
- per task — passes run, window/spool sizes, advisories delivered, cost in USD.
- cache — the provider's own
cacheRead/cacheWrite tokens and the hit ratio. The
fan-out shares one cached prefix across detectors, so at steady state most input
should be cached reads; a large uncached prompt with cacheRead≈0 means the sharing
is broken (see the plugin's DESIGN.md §Verifying the cache).
- debug — when on, the directory each pass's
digest.txt / pass.json /
<detector>.txt are written to.
- uptake per detector —
adopted/delivered: whether advice is being acted on.
A detector whose advice is persistently ignored is a detector to disable in
the plugin's detectors config.
The badge (🧠 in the chat toolbar) and the Second Brain sidebar panel show the
same numbers continuously, without spending a model turn.
