[Bug] Unable to process empty bug report

Bug Description

Environment Info

• Platform: linux
• Terminal: xterm-256color
• Version: 2.1.170
• Feedback ID: 3116ec8c-6b19-4e26-b4c2-9229ccda2147

Subject: Fable 5 — cross-model prompt cache (a fun launch-day observation + a feature wish)

Shipped Fable 5 today and I immediately took it for a spin on a defensive
security audit of my own secrets vault. The safety classifier flagged it and
fell back to Opus 4.8 — fair enough, the warning even says it over-triggers on
benign security work, no complaint.

What caught my eye: that automatic Fable→Opus fallback preserved my prompt
cache (first Opus turn read the full cached prefix, ~0 re-cache). But when I
then switched models myself via /model in the same session, it triggered a full
cache rebuild — even though it bounced right back to the same model. Same model
pair, opposite cache behavior depending on who triggers the switch.

Big caveat: this is one session, read off the CLI transcript usage numbers, not
reproduced or certified — I could be misreading it.

But it made me wonder: if cross-model cache reuse is feasible (your fallback
seems to do it), exposing it for user-initiated switches would be huge. Token
cost, energy footprint, and it'd let people right-size the model to the task —
I know far too many folks running a flagship just to draft emails. Letting me
adapt the model to the question mid-session, cache intact, would be genuinely
great. (Bonus: surfacing the automatic switches transparently, like you already
do, is a nice touch.)

Anyway — no gotcha, just found it fascinating on day one. Thanks for Fable.