golem15/msd-core
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(#1133): honor model_policy on the claude runtime (forward-port to next) (#1144)

Browse Source 
Forward-port of the #1133 hotfix (commit 22f237a5, 1.4.5 hotfix line) onto
next. The hotfix was authored against src/core.cts (v1.4.4); on next the
resolver logic lives in src/model-resolver.cts (ADR-457 extraction, #888),
so the patch is re-applied there rather than cherry-picked.

resolveModelInternal step 2.5 now honors model_policy on the claude runtime:
the policy-resolved full model ID is mapped back to a Claude Code agent alias
via CLAUDE_POLICY_ID_TO_ALIAS (reverse of MODEL_ALIAS_MAP + claude-fable-5 ->
fable). Bare aliases (opus/sonnet/haiku/fable) pass through; an ID with no
Claude alias warns once to stderr (deduped by agentType::policyModel::tier)
and falls back to the configured tier alias. Non-claude runtimes return full
IDs verbatim (unchanged). resolveModelForTier is intentionally unchanged.

The warn-dedupe cache lives in model-resolver.cts; core.cts composes the
exported _resetRuntimeWarningCacheForTests to clear both that cache and the
config-loader warning cache (config-loader cannot import model-resolver --
circular dependency).

Ports the 6 #1133 tests (rewriting the old claude-no-op test that asserted
the bug) plus one added test covering the MODEL_ALIAS_MAP reverse-map path.

Forward-port of #1133

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Tom Boucher
2026-06-12 19:56:54 -04:00
committed by GitHub
parent 4ab5c7b3f2
commit 44024aa535
7 changed files with 136 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/CONFIGURATION.md
View File

@@ -1294,7 +1294,7 @@ Choose a provider and budget level via the settings workflow; GSD writes the can
}
```
Known providers: `openai`, `anthropic`, `anthropic-fable`, `google`, `qwen`. Budget levels: `high`, `medium`, `low`. Use `anthropic` to keep the Opus 4.8-backed Claude preset, or `anthropic-fable` to opt into Claude Fable 5 for high-budget top-tier routing.
Known providers: `openai`, `anthropic`, `anthropic-fable`, `google`, `qwen`. Budget levels: `high`, `medium`, `low`. Use `anthropic` to keep the Opus 4.8-backed Claude preset, or `anthropic-fable` to opt into Claude Fable 5 for high-budget top-tier routing. On the default `claude` runtime, policy-resolved model IDs are mapped to Claude Code agent aliases (for example `claude-fable-5` → `fable`); an ID with no corresponding Claude alias emits a warning and falls back to the configured tier.
For advanced per-runtime control, `runtime_tiers` accepts explicit entries using the internal profile tier names (`opus`, `sonnet`, `haiku`):