[Feature Request] Make safeguard model-switch fallback target configurable

Bug Description
Feature request: make the safeguard model-switch fallback target configurable. On Fable 5, a [cyber] flag falls back to Opus 4.8 — hardcoded — even though Opus 5

Feature request: make the safeguard model-switch fallback target configurable. On Fable 5, a [cyber] flag falls back to Opus 4.8 — hardcoded — even though Opus 5 exists (and is already the target for the biology category). On first-party Anthropic API auth there is no setting, flag, or env var to retarget it (ANTHROPIC_DEFAULT_OPUS_MODEL only retargets on Bedrock/GCP/Foundry). Please allow choosing the fallback model, e.g. a switchModelsOnFlagTarget setting or honoring ANTHROPIC_DEFAULT_OPUS_MODEL for this path on first-party auth. Context: a PCI-certified payments (PSP) codebase whose CLAUDE.md legitimately mentions HMAC signing, token engines, and MFA trips [cyber] on ordinary planning work — sometimes on the first message, from workspace context alone — silently downgrading long sessions two model generations. Claude Code 2.1.234.

Environment Info

• Platform: linux
• Terminal: gnome-terminal
• Version: 2.1.234
• Feedback ID: 8128a605-ec06-4b8d-82fe-5b63331e5bea

Errors

[]