fix(providers): keep model metadata lookups offline when requested

[LivXue]

Summary

Creating an ACP session or opening the model picker could query LiteLLM's provider metadata API even with allow_fetch=False. Unknown Ollama models could block session/new beyond the EOF cleanup grace period and drop its reply.

• Forward the lookup policy to both context-window and output-ceiling metadata queries. Cache misses skip get_model_info when fetching is disabled.
• Keep loaded LiteLLM rows, cached OpenRouter metadata, and SDK lookups where fetching is allowed.
• Cover uncached model spellings, cached aliases, real model.options handling, and all three ACP handshake replies after EOF.

Windows stdio validation uses the stdin fix from #908 in an isolated source snapshot. This PR carries only the metadata fix and its regression tests.

Type

• Fix
• Feature
• Docs
• CI / tooling
• Refactor
• Other

Verification

• Relevant tests pass locally
• Relevant lint / type checks pass locally
• User-facing docs or screenshots are updated when needed

Native Windows, Python 3.13.12: 602 relevant tests passed.

uv run --frozen pytest -n 0 tests/test_provider_rates.py tests/test_rpc_model.py tests/test_acp_server.py tests/test_acp_methods.py tests/test_acp_config_options.py tests/test_agent_loop_lazy_provider.py tests/test_lazy_provider.py tests/test_token_wise_pricing.py tests/test_rpc_session_init_bundle.py -q --tb=short

Before the fix, 12 new regression cases failed while both allowed-fetch controls passed. Removing only the output guard produced 5 failures; removing only the context-window guard produced 7 failures, including the RPC and ACP cases. Source was restored after each mutation.

Native stdio smoke: 6 passed, including the complete handshake, with PYTHONPATH pointing at an isolated copy of this change plus raven/cli/acp_commands.py from #908.

uv run --frozen pytest -n 0 tests/integration/test_acp_stdio_smoke.py -m integration -q -x --tb=short

A direct Windows smoke run on the branch alone remains blocked by the inherited WinError 6 stdin transport failure, reproduced before this fix. The combined run above verifies the two fixes together. No LLM calls or real Ollama service were used.

The following checks passed; all 10 import contracts were kept.

uv run --frozen pre-commit run --files raven/providers/rates.py tests/test_provider_rates.py tests/test_rpc_model.py tests/test_acp_server.py
uv run --frozen ty check raven/providers/rates.py
uv run --frozen lint-imports
uv run --frozen python scripts/check_large_files.py origin/main
uv run --frozen python scripts/check_source_language.py origin/main
git diff --check

No user-facing documentation update is needed: configuration and protocol contracts are unchanged. Resolver docstrings document the enforced lookup policy.

Risk

• Security impact considered
• Backward compatibility considered
• Rollback path is clear for risky changes

When every local source misses, callers requesting local metadata retain the existing unknown-value or configured-default behavior. A self-hosted model can set agents.defaults.contextWindowTokens explicitly. Fetch-enabled lookup behavior is preserved. Roll back by reverting this change.

Related Issues

#908 (Windows stdin fix used for combined stdio validation).

[gloryfromca]

No blockers; this can merge as far as I am concerned.

Reviewed the full github/main...HEAD diff and the surrounding resolver, candidate-alias, OpenRouter-cache, model-picker, binding, and ACP call paths, plus the relevant history. The new policy gate prevents litellm.get_model_info from running when allow_fetch=False while preserving already-loaded static rows, cached OpenRouter rows, and the existing fetch-enabled default. The optional parameters preserve compatibility, and the tests add coverage without weakening existing assertions.

I also checked the repository rules in AGENTS.md/CLAUDE.md, the domain-term routing in CONTEXT-MAP.md, test naming, source language/comments, and the provider/RPC architecture constraints; I found no violations.

Verification: uv run pytest -n 0 tests/test_provider_rates.py tests/test_rpc_model.py tests/test_acp_server.py tests/test_acp_methods.py tests/test_acp_config_options.py tests/test_agent_loop_lazy_provider.py tests/test_lazy_provider.py tests/test_token_wise_pricing.py tests/test_rpc_session_init_bundle.py -q --tb=short passed (602 tests); uv run ty check raven/providers/rates.py and git diff --check github/main...HEAD also passed. An initial run in the newly created base environment failed during autouse setup because the optional workspace package raven_everos was absent; after the declared dev extra was synced, the three modified suites passed 358/358 and the broader run above passed.

[LivXue]

No blockers in the reviewed scope.

Scope: I reviewed the full PR diff (4 changed files), including tests and documentation, and traced affected callers and contracts at 41734946f8ff. This review covers all changed areas, not only raven/agent.

Checked both context-window and output-ceiling entry points, loaded-table aliases, SDK lookup refusal when allow_fetch=False, cached OpenRouter fallback, fetch-enabled compatibility, model-picker and binding callers, lazy providers, and ACP session/config-option paths. The no-fetch flag reaches both LiteLLM import policy and SDK metadata policy.

Verification (native Windows, Python 3.12, isolated source snapshots; local JUnit-output options omitted below):

• PR head: uv run --no-sync pytest -o addopts= tests/test_provider_rates.py tests/test_rpc_model.py tests/test_acp_server.py tests/test_acp_methods.py tests/test_acp_config_options.py tests/test_agent_loop_lazy_provider.py tests/test_lazy_provider.py tests/test_token_wise_pricing.py tests/test_rpc_session_init_bundle.py -q --tb=short -> 602 passed.

The local combined tree with #908 also passed uv run --no-sync --with truststore==0.10.4 pytest -o addopts= tests/integration/test_acp_stdio_smoke.py -q --tb=short (11 tests). No new finding was identified. The full repository suite was not run.

git diff --check over the complete merge-base-to-head range passed. No new inline finding is being filed.

[LivXue]

Windows compatibility review: no new issue found in the inspected scope.

Reviewed head 41734946f8ffa579e6fd1a0117a3774d34ccd500 against base 612696546702411e63d22f4b87d742d5e78b146b: all 4 changed files, provider metadata/cached-alias lookup callers, model-picker paths and ACP session handling. Checked cache-only lookup, permitted-fetch controls and frame-loop responses, including the EOF regression.

Native Windows / Python 3.13.12, detached temporary checkout with PYTHONUTF8=1, UV_NO_SYNC=1 and isolated user/Raven homes:

uv run pytest tests/test_provider_rates.py tests/test_rpc_model.py tests/test_acp_server.py -n 0 -x --tb=short --basetemp "$reviewRunDir/pr909/tests"

Result: 358 passed. The ACP frame-loop test uses in-memory streams; it does not validate native inherited stdin or the separate Windows stdin repair. No live provider/Ollama service or full suite was run. This scoped result does not establish full Windows compatibility.