This repository was archived by the owner on Jul 30, 2026. It is now read-only.
feat(core): add Devin CLI SQLite integration for local token accounting - #196
Open
vlln wants to merge 3 commits into
Open
feat(core): add Devin CLI SQLite integration for local token accounting#196vlln wants to merge 3 commits into
vlln wants to merge 3 commits into
Conversation
Read assistant token metrics from the Devin CLI session store at ~/.local/share/devin/cli/sessions.db. The integration joins message_nodes with sessions, extracts the four metadata.metrics token buckets (input/output/cache_read/cache_creation) from assistant chat_message blobs, and maps them to TokenBreakdown. Twin child nodes sharing a message_id are deduplicated by (session_id, message_id) so each assistant turn is counted once per session. The forest structure is not reconstructed; Tokscale accounts for token usage only. Session model labels are preserved verbatim and canonicalized by the shared pricing/identity pipeline. Provider attribution is inferred from the model id; unresolvable providers stay unknown. Vendor credit_cost metadata is ignored per ADR 0001 — cost is derived from token buckets and the Tokscale pricing table. Hidden sessions are excluded; zero-token rows are filtered; malformed JSON and missing model/timestamp are reported as record rejections without aborting the scan. Discovery resolves the default sessions.db plus scanner.extraScanPaths roots, opens the database read-only with committed WAL state, and uses the uncached parse path since Devin maintains a single live database. ADR 0031 documents the scope, credit-cost boundary, and rejection behavior. Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
🚧 Files skipped from review as they are similar to previous changes (1)
Walkthrough新增 Devin 客户端目录项与解码器标识，解析本地 SQLite 会话中的 token 使用量，支持数据库发现、WAL 指纹、消息去重和拒绝记录，并接入 CLI/TUI、缓存流水线及相关文档。 ChangesDevin 客户端集成
Estimated code review effort: 4 (Complex) | ~45 minutes Sequence Diagram(s)sequenceDiagram
participant CLI
participant DevinIntegration
participant sessions.db
participant DevinDecoder
participant MessageCache
CLI->>DevinIntegration: 请求发现并解析 Devin 输入
DevinIntegration->>sessions.db: 查找默认及额外数据库
DevinIntegration->>DevinDecoder: 解析每个 InputUnit
DevinDecoder->>sessions.db: 读取会话与消息节点
DevinDecoder-->>DevinIntegration: 返回消息与拒绝记录
DevinIntegration->>MessageCache: 去重并折叠解析结果
Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@crates/tokscale-core/src/integrations/devin/decode.rs`:
- Around line 222-258: 负数 token 当前会被 extract_assistant_usage 当作缺失值并静默归零，与
non_negative_i64 的文档及 malformed-record 预期不符。更新 non_negative_i64 与
extract_assistant_usage 的处理，使字段缺失仍按零处理，但检测到负数时拒绝整条记录并上报
malformed-record；保留现有正常值解析行为。
- Around line 46-56: Update the query and extraction flow around
extract_assistant_usage to select a node-level identifier such as n.node_id or
n.row_id and pass it through. Change node_fallback and dedup_key construction so
assistant messages without message_id use a unique node-<id> fallback instead of
an empty message_id-derived value, while preserving message_id-based
deduplication. Extend node_without_message_id_uses_node_fallback_for_dedup with
two such nodes in one session and verify both are retained.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 33c4e2c3-b2bc-47d3-8ccd-4b8c8ff80738
📒 Files selected for processing (10)
README.mdREADME.zh-cn.mdcrates/tokscale-cli/src/tui/data/mod.rscrates/tokscale-core/client-catalog.jsoncrates/tokscale-core/src/integrations/devin/decode.rscrates/tokscale-core/src/integrations/devin/mod.rscrates/tokscale-core/src/integrations/mod.rscrates/tokscale-core/src/message_cache.rsdocs/adr/0031-devin-cli-sqlite-integration.mddocs/clients.md
… buckets Two data-integrity fixes to the Devin decoder, both flagged by CodeRabbit review on PR makoMakoGo#196: 1. Assistant messages without a message_id were collapsed into a single empty-string dedup key per session, so all but the first were silently dropped by the fold-layer dedup. The SQL query now selects n.node_id and extract_assistant_usage builds a unique `node-<id>` fallback, matching the module doc and ADR 0031. End-to-end against real local data recovers ~6.67M previously-lost tokens. 2. Negative token buckets were silently zeroed via unwrap_or(0) instead of being rejected as malformed-record, contradicting the non_negative_i64 doc. Replaced with token_bucket, which returns Ok(None) for absent/null fields (treated as zero) but Err(NegativeToken) for negative values, so the caller rejects the row as malformed-record. Regression tests cover two message_id-less nodes in one session (both retained with distinct dedup keys) and a negative input_tokens row (rejected while a healthy sibling is kept). Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@crates/tokscale-core/src/integrations/devin/decode.rs`:
- Around line 202-204: 在记录解码逻辑及其 token 桶转换处，更新 as_f64 到 i64
的处理：转换前先拒绝负数、非整数值以及超出 i64 范围的浮点数，再执行安全转换并返回 NegativeToken 等现有错误。保留
malformed-record 的现有拒绝流程，并在 negative_token_bucket_is_rejected_as_malformed
附近补充非负小数和可溢出浮点数的回归测试。
- Line 110: 在解析行的逻辑中更新 node_id 读取，移除 row.get(4) 的 unwrap_or(-1) 回退值；读取 NULL
或类型错误时记录 MalformedRecord 并跳过当前行，避免生成重复的伪节点 ID。保留有效 node_id 记录的现有处理流程。
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 084bfaf7-bf2a-4735-ada2-c010b7876580
📒 Files selected for processing (1)
crates/tokscale-core/src/integrations/devin/decode.rs
…version Two more data-integrity fixes from CodeRabbit review on PR makoMakoGo#196: 1. row.get(4).unwrap_or(-1) collapsed any NULL or type-mismatched node_id into a shared `node--1` fallback key, so multiple malformed rows would be deduplicated to one and silently lose usage. Now a read failure records MalformedRecord and skips the row. 2. as_f64().map(|v| v as i64) silently truncated problematic floats: -0.5 became 0 (bypassing the negative check), 1.5 became 1 (losing data), and 1e19 overflowed to a wrapped value. token_bucket now validates finiteness, non-negativity, integrality, and i64 range before converting, returning NegativeToken otherwise. Regression tests cover a non-integer node_id (rejected as malformed) and fractional/overflowing float token buckets (-0.5, 1.5, 1e19 all rejected; 10.0 accepted). Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Author
|
@coderabbitai review |
✅ Action performedReview finished.
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
devinclient identity toclient-catalog.jsonand wireDecoderId::Devininto the message cache andintegration_fordispatchintegrations/devinmodule reads assistant token metrics from the Devin CLI SQLite store at~/.local/share/devin/cli/sessions.db, joiningmessage_nodeswithsessionsand extracting the fourmetadata.metricstoken buckets intoTokenBreakdownmessage_idare deduplicated by(session_id, message_id)so each assistant turn is counted once per session; the forest structure is not reconstructedunknowntotal_credit_cost/total_acu_costmetadata is ignored per ADR 0001 — cost is derived from token buckets and the Tokscale pricing tablesessions.dbplusscanner.extraScanPathsroots, opens the database read-only with committed WAL state, and uses the uncached parse pathdocs/clients.mdand both READMEs list the new clientTest plan
cargo test -p tokscale-core --lib devin— 12 new unit tests pass (schema error, assistant metrics extraction, twin dedup, zero-token filter, hidden exclusion, credit-cost ignored, missing-model rejection, unknown provider, malformed JSON, node-fallback dedup, discovery)cargo test -p tokscale-cli --bin tokscale— 750 tests pass (updatedtest_client_allordering assertion)cargo build -p tokscale-cli— clean buildcargo fmt --check -p tokscale-core— cleantokscale models --client devin --no-spinnerreports 315M tokens / $223.56, with unpriceable models (glm-5-2,swe-1-6-slow) explicitly at $0.00Generated with Devin
Summary by CodeRabbit