Land-ready proof for #68066:

Behavior addressed: completed OpenRouter streaming chat completions now reconcile message.usage.cost.total with OpenRouter generation metadata before the final assistant message is emitted.

Real environment tested: yes, with stored OPENROUTER_API_KEY and OpenRouter live API. A streamed google/gemini-2.5-flash call returned HTTP 200, a gen-... response id, finish=stop, and streamed cost 0.0000049. A delayed GET https://openrouter.ai/api/v1/generation?id=<id> returned HTTP 200 with matching id, streamed=true, cancelled=false, and total_cost=0.0000049.

Exact commands run after this patch:

node scripts/run-vitest.mjs extensions/openrouter/index.test.ts extensions/openrouter/provider-runtime.contract.test.ts
node scripts/run-vitest.mjs extensions/openrouter
pnpm exec oxfmt --check extensions/openrouter/stream.ts extensions/openrouter/index.test.ts
git diff --check
/Users/steipete/Projects/agent-scripts/skills/autoreview/scripts/autoreview --mode local --prompt-file /tmp/openrouter-autoreview-evidence.md

Observed result after fix: focused OpenRouter provider tests passed, full OpenRouter extension tests passed (8 files / 79 tests), formatting and diff whitespace checks passed, and autoreview reported no accepted/actionable findings after the accepted abort-path fix.

What was not tested: I did not rerun the expensive long-context tiered-pricing repro from #68066; the live test covered the real response-id to generation-metadata contract and the patch covers the persisted final message path.

ClawSweeper status: review started.

I am starting a fresh review of this pull request: fix(openrouter): reconcile streamed generation cost This is item 1/1 in the current shard. Shard 0/1.

This placeholder means the worker is alive and reading the current context. I will edit this same comment with the actual review when the claws are done clicking.

Crustacean status: shell secured, claws on keyboard, evidence pebbles being sorted.

Found 2 test failures on Blacksmith runners:

Failures

CI note before landing:

Relevant proof for this PR passed:

• node scripts/run-vitest.mjs extensions/openrouter/index.test.ts extensions/openrouter/provider-runtime.contract.test.ts
• node scripts/run-vitest.mjs extensions/openrouter
• pnpm exec oxfmt --check extensions/openrouter/stream.ts extensions/openrouter/index.test.ts
• git diff --check
• live OpenRouter API proof: streamed google/gemini-2.5-flash returned a gen-... response id and streamed usage cost; delayed GET /api/v1/generation?id=... returned the matching id with streamed=true, cancelled=false, and total_cost.

CI failures inspected from run 27080990032; they are unrelated to the OpenRouter patch:

• check-test-types: existing extensions/memory-core/doctor-contract-api.test.ts config type errors.
• check-lint / check-additional-extension-bundled: annotations in extensions/memory-core, extensions/device-pair, scripts/test-live-media.ts, and src/cli/exec-approvals-cli.ts.
• check-additional-runtime-topology-architecture: existing src/agents / src/plugins import cycle.
• build-artifacts: gateway-watch did not report ready before the idle CPU window; channels/core-support-boundary tests in that job passed.