fix(claude): preserve cc-mirror tool result attribution

[IvGolovach]

Summary

• Preserve cc-mirror variant client attribution for Claude tool_result usage rows.
• Carry the cc-mirror variant provider fallback into tool-result parsing, so pricing and diagnostics stay provider-aware.
• Add regression coverage for cc-mirror tool-result records using a variant client id and non-Anthropic provider.

Why

Tokscale can discover cc-mirror Claude-compatible variants, but Claude tool_result usage rows had a separate parsing path that could fall back to plain claude attribution. That makes variant-scoped usage harder to understand and can lose the provider context needed for accurate pricing when a cc-mirror variant points at a non-Anthropic backend.

Related issue: #610.

Diff scope

• crates/tokscale-core/src/sessions/claudecode.rs: threads the active client and provider context through Claude tool-result usage parsing.
• crates/tokscale-core/src/sessions/claudecode.rs: adds a regression fixture proving a cc-mirror variant keeps cc-mirror/<variant> client attribution and the variant provider fallback on tool-result usage.

Branch integrity

• Base branch: main.
• Validated base SHA: a86e688d620939d2c973c6d5625baa815ea223d7.
• Ahead/behind: 0 behind / 1 ahead against origin/main.
• Merge base: a86e688d620939d2c973c6d5625baa815ea223d7.
• Fast-forward safety: origin/main is an ancestor of this branch.

Commit integrity

• Introduced commit: 75b321952d6858d8551be8794426c0829f3c333f fix(claude): preserve cc-mirror tool result attribution.
• The PR contains one logical change scoped to Claude parser attribution for cc-mirror tool-result usage.
• Ledger: not applicable — not required for this change family.
• Version: PASS, scripts/check-version-coherence.sh.

Diff hygiene

• git diff --name-status origin/main...HEAD: only crates/tokscale-core/src/sessions/claudecode.rs changed.
• git diff --check origin/main...HEAD: PASS, no output.

Validation mode and proof

Mode 2 — narrow runtime change, because the diff changes localized Claude parser attribution behavior and focused tests without touching migrations, auth, deployment tooling, or external service contracts.

• TDD red proof: cargo test -p tokscale-core sessions::claudecode::tests::test_cc_mirror_tool_result_keeps_variant_client_and_provider failed before the implementation with plain claude attribution instead of the cc-mirror variant client.
• cargo test -p tokscale-core sessions::claudecode::tests::test_cc_mirror_tool_result_keeps_variant_client_and_provider: PASS.
• cargo test -p tokscale-core sessions::claudecode: PASS.
• cargo test -p tokscale-core cc_mirror: PASS, 9 tests.
• cargo fmt --all -- --check: PASS, no output.
• scripts/check-version-coherence.sh: PASS, Version coherence OK: 3.0.0.
• git diff --check origin/main...HEAD: PASS, no output.
• Not run: frontend validation tests — frontend files were not touched.

Required remote gates

Pending — GitHub Actions and mergeability checks will run after the PR is opened.

Migration notes

Not applicable — no database migration changed.

Runtime safety

The change only passes existing parser context into an existing Claude usage branch. It does not add external I/O, persistent state, concurrency, or new scan paths. No invariant regression introduced.

Documentation integrity

Not applicable — no docs, commands, or runbooks changed.

Rollback plan

Rollback: revert this PR. DB downgrade: not applicable. Data repair: not applicable. Operational caveats: reverting would restore plain Claude attribution for cc-mirror tool-result usage rows.

Known residual risks

Remote CI and GitHub mergeability are pending until the PR is opened. This PR is intentionally scoped to preserving attribution for cc-mirror tool-result rows; it does not expand cc-mirror discovery beyond the support already present on main.

---

Summary by cubic

Preserves cc-mirror/<variant> client and provider attribution when parsing Claude tool_result rows. This keeps pricing and diagnostics provider-aware, including for non-Anthropic backends.

• Bug Fixes
  • Threaded client_id and provider hints through Claude tool-result parsing.
  • Used the variant client in message metadata and dedup keys.
  • Added a regression test to validate variant client and provider are retained.

^{Written for commit 75b3219. Summary will update on new commits.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
tokscale	Ignored	Preview	Jun 1, 2026 6:28am

[cubic-dev-ai]

No issues found across 1 file

_{Re-trigger cubic}