release: 0.8.15 (gated model works from the CLI with an API key)
0.8.15. Fixes the launch blocker found testing 0.8.14 against production, and bumps the version.
0.8.14 published the three harnesses (shannon, deepsec, vvah), but with only WALLY_CLOUD_KEY they refused deepseek-v4.1-flash-1017-uncensored ("not available on this account") even though the same key calls the model (200) and the catalogue lists it.
Cause: the CLI built its model list from GET /v1/models alone; the gateway strips every private (listed = false) model from that list for every key, so the granted account could never select it.
Fix: account_models_with also adds non-decisions-only models the grant-aware catalogue names that /v1/models omitted. No grant → never listed. Done in the CLI, not the gateway filter (no caller identity there; a mistake would leak the private id to every key).
Tests: new red→green test + no-grant and listed-once guards; cargo test --lib 621 passed, clippy -D warnings + fmt clean, golden suite passes, check-versions consistent at 0.8.15.
Verified live on production with an API key: deepsec and vvah resolve the model via https://inference.runanywhere.ai/v1 and hand off the right config; Shannon launched a real scan on it.
Label release:patch → merging tags v0.8.15 and publishes.
🤖 Generated with Claude Code
Summary by CodeRabbit
- New Features
- Models listed in your account’s catalog but missing from the model list are now included, with catalog pricing. Models marked as decisions-only and models already in the list are excluded.
- Release
- Updated the product version to 0.8.15.