fix(cost): derive local costs from token pricing

[makoMakoGo]

Summary

• Add ADR 0011 for token-derived local report cost semantics.
• Ignore app/vendor cost fields in local parsers and always derive UnifiedMessage.cost through Tokscale pricing.
• Drop cost-only rows and disable both Crush and Warp usage rows until token-level sources exist.
• Bump source-message cache schema to rebuild cached app-reported costs.

Verification

• cargo test -p tokscale-core
• cargo test -p tokscale-cli
• cargo run -p tokscale-cli -- --no-spinner --json --client crush
• cargo run -p tokscale-cli -- --no-spinner --json --client opencode

---

Summary by cubic

Local report costs now derive only from token pricing. Parsers ignore app/vendor spend, rows without positive tokens are dropped, and cost-only clients (Crush, Warp) are fully disabled and removed from scanning.

• Refactors

  • Token-only pricing: compute UnifiedMessage.cost from tokens via crate::finalize_token_priced_messages; strip embedded costs in parsers; normalize token buckets; skip zero-token rows; clean totals (including -0.0) in reports/TUI; offline totals are 0.0 without pricing.
  • Remove cost-only clients: delete crush/warp adapters and session parsers, drop their local scan paths, mark them as not locally parsed, update the scanner to ignore them, remove Warp cache checks and leftover scan/submit paths, and trim related CLI/tests.

• Migration

  • Bump source-message cache schema to 26; cache rebuilds on first run.

^{Written for commit 9ea9f30. Summary will update on new commits.}

Summary by CodeRabbit

发布说明

• Bug Fixes

  • 统一成本口径：本地报告的费用仅从 token 使用量派生计算；当缺少正 token 时会丢弃/归零仅费用类信息，且聚合总成本会进行净化。
  • 相关会话/适配流程不再依赖供应商上报的费用字段；改进集成结果一致性。
  • 命令行离线模式下的成本输出断言更严格（缺少定价缓存场景 totalCost 现在精确为 0.0）。
  • 报表中对无 token 的 Warp/Crush 数据不再纳入，并更新本地告警文案。

• Chores

  • 更新缓存架构版本，触发缓存重建以适配新的定价语义。

[coderabbitai]

Walkthrough

将所有本地 session 解析器中对应用/vendor 上报的 cost/credits 字段的读取全部移除，改为在解析后通过新的 apply_token_pricing 函数统一从 token buckets 派生 UnifiedMessage.cost（无定价时为 0.0）；聚合层新增 clean_total_cost 归一化；缓存版本升至 26；新增 ADR 0011 文档记录该架构决策。

Changes

Token 派生本地成本重构

Layer / File(s)	Summary
核心定价入口与缓存版本升级 crates/tokscale-core/src/lib.rs, crates/tokscale-core/src/message_cache.rs, docs/adr/0011-token-derived-local-cost.md	新增 apply_token_pricing 与 finalize_token_priced_messages：进入时强制清零 message.cost，无定价时直接返回，有定价时按 token 计算并仅在结果 >0 时写回；finalize 先做 token 归一化与派生字段刷新，过滤无正 token 记录，再调用 pricing。CACHE_SCHEMA_VERSION 升至 26。ADR 0011 文档化规则。
删除 PricingPolicy 并简化 file 适配器 crates/tokscale-core/src/adapters/file.rs	NonCachedFileAdapter 移除 pricing_policy 字段与构造参数，parse 改为调用新的 finalize_token_priced_messages；WARP_ADAPTER 静态初始化被删除。
cache 与 codex 适配器迁移定价入口 crates/tokscale-core/src/adapters/cache.rs, crates/tokscale-core/src/adapters/codex.rs	cache 的三处定价位点（fingerprint 缺失、缓存命中、缓存插入前）与 codex 的消息定价都改为调用 finalize_token_priced_messages，删除私有的 apply_pricing_to_messages 函数。
其他适配器从逐消息定价改为批量处理 crates/tokscale-core/src/adapters/crush.rs, crates/tokscale-core/src/adapters/gjc.rs, crates/tokscale-core/src/adapters/goose.rs, crates/tokscale-core/src/adapters/hermes.rs, crates/tokscale-core/src/adapters/kilo.rs, crates/tokscale-core/src/adapters/kiro.rs, crates/tokscale-core/src/adapters/trae.rs	删除 PricingPolicy 导入，parse 内对消息逐条应用定价的逻辑改为调用 finalize_token_priced_messages；测试期望与 cost 数值同步调整。
适配器注册表与模块导出清理 crates/tokscale-core/src/adapters/mod.rs, crates/tokscale-core/src/sessions/mod.rs, crates/tokscale-core/src/adapters/opencode.rs	从 LOCAL_SOURCE_ADAPTERS 移除 WARP_ADAPTER 与 CRUSH_ADAPTER；从 sessions/mod.rs 移除 pub mod crush/warp 导出；新增单测验证 Warp/Crush 不注册为本地适配器；opencode 适配器移除 Crush 分支。
Session 解析器移除成本字段：AMP、Codebuff、Cursor、GJC crates/tokscale-core/src/sessions/amp.rs, crates/tokscale-core/src/sessions/codebuff.rs, crates/tokscale-core/src/sessions/cursor.rs, crates/tokscale-core/src/sessions/gjc.rs	删除结构体中的 cost/credits 字段及对应读取逻辑，统一 token 归一化（各项 max(0)）与正性过滤（positive_token_total==0 时跳过），消息初始成本改为 0.0，更新单测期望。
Session 解析器成本清理与 token 过滤：Hermes、Kilo、Mux、OpenClaw crates/tokscale-core/src/sessions/hermes.rs, crates/tokscale-core/src/sessions/kilo.rs, crates/tokscale-core/src/sessions/mux.rs, crates/tokscale-core/src/sessions/openclaw.rs	SQL 查询删除 estimated/actual cost 列与相关条件，JSON 反序列化删除 cost 字段，UnifiedMessage 初始成本改为 0.0，统一 token 的 max(0) 与正性过滤，更新单测。
Session 解析器指纹清理与 token 过滤：OpenCode、RooCode、Trae crates/tokscale-core/src/sessions/opencode.rs, crates/tokscale-core/src/sessions/roocode.rs, crates/tokscale-core/src/sessions/trae.rs	JSON 与 SQLite 反序列化删除 cost 字段，SQLite 指纹逻辑移除成本维度，token 统一 clamp 与 positive_token_total 过滤，消息初始成本改为 0.0，更新单测。
Crush 解析器置空实现 crates/tokscale-core/src/sessions/crush.rs	parse_crush_sqlite 改为直接返回空向量，删除所有 SQLite 查询/递归会话树逻辑；测试精简为仅两种空返回场景（缺失数据库、cost-only 记录）。
聚合层成本零值归一化 crates/tokscale-core/src/aggregator.rs, crates/tokscale-core/src/aggregate/engine.rs, crates/tokscale-core/src/aggregate/tui.rs	新增 clean_total_cost 函数确保 cost==0.0 时写入精确 0.0；在 MonthlyReport、HourlyReport、ModelReport 与 calculate_summary 中应用于 total_cost/max_cost；sane_cost 条件从 >=0.0 改为 >0.0。
lib.rs 定价测试切换与重基线 crates/tokscale-core/src/lib.rs	大量定价测试导入与函数名从 apply_pricing_if_available 改为 apply_token_pricing；无定价场景断言改为 0.0；新增 finalize 过滤、token 正性汇总与饱和行为测试；所有定价相关测试保持 cost 期望不变。
Session 与集成测试重基线 crates/tokscale-core/src/sessions/cline.rs, crates/tokscale-core/src/sessions/kilocode.rs, crates/tokscale-core/tests/codebuff.rs, crates/tokscale-core/tests/gjc.rs, crates/tokscale-core/tests/hermes.rs	cline、kilocode 单元测试 cost 期望从 0.05/0.05 改为 0.0；codebuff 集成测试从 1.25 改为 0.0；gjc/hermes 从"嵌入/回退"语义改为"统一 token 定价"，期望值基于 token·price 计算。
CLI 离线与 Warp 告警更新 crates/tokscale-cli/tests/cli_tests.rs, crates/tokscale-cli/src/main.rs, crates/tokscale-cli/src/warp.rs	CLI 离线无定价缓存场景的 totalCost 断言统一改为严格等于 0.0（四个命令）；warp_setup_warnings_for_report 简化为仅在显式请求时返回固定告警；删除 has_usage_cache_in_home 辅助函数。

估计代码审查工作量

🎯 4 (Complex) | ⏱️ ~60 minutes

可能相关的 PR

• makoMakoGo/tokscale#19：同样修改了 amp.rs 的 AMP 会话解析/成本语义，将解析得到的 credits 字段通过定价处理，与本 PR 将 credits 完全移除属同一模块的强相关变更。

诗

🐇 小兔子算账忙，
旧成本字段统统藏。
Token 桶里量一量，
apply_token_pricing 新路畅！
零点零归零点零，
缓存版本 26 上场。✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	PR标题清晰准确地概括了本次变更的核心目标：从本地应用成本转变为基于Token定价推导成本。
Docstring Coverage	✅ Passed	Docstring coverage is 80.77% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/token-derived-local-cost

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[makoMakoGo]

/juya review

[codspeed-hq]

Merging this PR will not alter performance

✅ 18 untouched benchmarks

---

_{Comparing fix/token-derived-local-cost (9ea9f30) with personal/local-clients (51ca010)¹}

Footnotes

1. No successful run was found on personal/local-clients (eef2342) during the generation of this report, so 51ca010 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[cubic-dev-ai]

1 issue found across 34 files

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[makoMakoGo]

/juya review

[juya-review-bot]

OpenCodeReview failed before posting review comments. Check bot logs for this pull request and retry after fixing the service configuration.

[makoMakoGo]

/juya review

[juya-review-bot]

OpenCodeReview failed before posting review comments. Check bot logs for this pull request and retry after fixing the service configuration.

[makoMakoGo]

/juya review

[juya-review-bot]

OpenCodeReview inline comment.

[juya-review-bot]

OpenCodeReview inline comment.

[juya-review-bot]

OpenCodeReview inline comment.