/code-review medium now runs the high-effort recipe on most models

What

/code-review (alias /review) picks a review recipe from a table keyed by model and effort level. Only claude-opus-5, claude-opus-4-8 and claude-sonnet-5 have their own row; every other model, Opus 5.5 included, reads the default row. In that row medium moved from the o5-bmin recipe to o48-high-v1, so it now runs the same recipe as high.

Details

• o5-bmin, the old medium: one careful pass over the diff, up to 15 findings.
• o48-high-v1, the new medium: eight finder angles run in the same context (line-by-line scan, removed behaviour, cross-file callers, cleanup, altitude, conventions), up to 6 candidates each, deduplicated with no verify step and biased toward recall, up to 10 findings unless --max-findings sets another limit.
• low, high, xhigh and max are unchanged in the default row; high was already o48-high-v1 in 2.1.289.
• The claude-opus-5 row still maps medium and high to o5-bmin. The Opus 4.8 and Sonnet 5 rows did not change.
• This is not the session's /effort setting, and the model's reasoning effort does not change: both recipes pass the level you typed through as the effort.

Why

On Opus 5.5 or any other model without its own row, /code-review medium is now a longer review that surfaces more uncertain findings than it did in 2.1.289, and its default cap drops from 15 findings to 10.

Evidence

medium: { ...H("o48-high-v1"), measuredExternal: !0 }