Add JetBrains Junie client support - #727
Merged
Merged
Conversation
Contributor
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
junhoyeo
approved these changes
Jun 21, 2026
junhoyeo
left a comment
junhoyeo
left a comment
Owner
There was a problem hiding this comment.
Reviewed via multi-agent review + local build/test/clippy + frontend parity test; full CI now green across all platforms (the first-time-contributor gate had blocked the native builds — I approved them and they pass). Adds JetBrains Junie end-to-end with exhaustive registry parity (Rust define_clients! Junie=32, COUNT->33; constants.ts/types.ts/SourceLogo.tsx; clientRegistry.test.ts passes), robust events.jsonl parsing, and embedded-cost preservation (reprice only when absent). Synced the ja/ko/zh READMEs with the English Junie entries. Thanks for the contribution — LGTM!
This was referenced Jun 22, 2026
junhoyeo
added a commit
that referenced
this pull request
Jun 22, 2026
…bal ordinal The file-global `usage_ordinal` gave a replayed identical `LlmResponseMetadataEvent` a new suffix (:0 then :1), so both the original and the replay passed the in-file `seen` set and the cross-file `should_keep_deduped_message` filter — double-counting the same tokens/cost. Revert the uniqueness component to the row's position within the event's own `modelUsage` array. A replayed event reproduces the same dedup key (collapsed to one), while multiple distinct rows inside one event keep distinct indices. Genuinely-distinct calls already differ via `timestampMs` in the key, so the #727 undercount fix is preserved. Constraint: cross-file Junie dedup relies on dedup_key being stable across replays Rejected: process-global monotonic counter | breaks replay collapse (regresses this PR's own fix) Confidence: high Scope-risk: narrow Not-tested: two byte-identical events with identical timestamps are treated as a replay (indistinguishable from a true replay)
junhoyeo
added a commit
that referenced
this pull request
Jun 22, 2026
…749) * fix(junie): unique dedup ordinal, scoped turn-start, kind-based skip Three confirmed bugs in the Junie session parser (PR #727): (a) The dedup key embedded `usage_index` from `usages.iter().enumerate()`, which resets to 0 on every `modelUsage` row. Two distinct LLM calls with identical token counts produced the same key and the second was silently dropped. Replace it with a file-global monotonic ordinal so every usage row across the whole session earns a unique key. (b) `pending_turn_start` was set by a UserPromptEvent and only cleared when a usage message was actually pushed. A prompt that produced no counted usage leaked the flag onto a later, unrelated turn's usage event. Clear it once per usage event (the first response event consumes the prompt's turn-start) and mark only the first emitted message of that event. (c) SKIP_EVENT_KINDS was matched via `line.contains(kind)`, a raw substring search that could drop legitimate usage/prompt lines merely mentioning a skipped kind in their text. Match on the parsed event kind instead, keeping the substring filter only as a cheap pre-parse guard. Adds regression tests, including two distinct usage rows with identical token counts that must both be counted (fails without the ordinal fix). Confidence: high Scope-risk: narrow Not-tested: real multi-megabyte Junie state snapshots (perf path) * fix(junie): derive dedup suffix from within-event index, not file-global ordinal The file-global `usage_ordinal` gave a replayed identical `LlmResponseMetadataEvent` a new suffix (:0 then :1), so both the original and the replay passed the in-file `seen` set and the cross-file `should_keep_deduped_message` filter — double-counting the same tokens/cost. Revert the uniqueness component to the row's position within the event's own `modelUsage` array. A replayed event reproduces the same dedup key (collapsed to one), while multiple distinct rows inside one event keep distinct indices. Genuinely-distinct calls already differ via `timestampMs` in the key, so the #727 undercount fix is preserved. Constraint: cross-file Junie dedup relies on dedup_key being stable across replays Rejected: process-global monotonic counter | breaks replay collapse (regresses this PR's own fix) Confidence: high Scope-risk: narrow Not-tested: two byte-identical events with identical timestamps are treated as a replay (indistinguishable from a true replay)
pinion05
added a commit
to pinion05/tokscale
that referenced
this pull request
Jun 23, 2026
Adds the JetBrains Junie coding agent as a first-class client. - Session source: ~/.junie/sessions/*/events.jsonl (ClientId::Junie = 32, COUNT -> 33) - Parser reads modelUsage cost/tokens/turn-start with provider inference; preserves Junie's authoritative per-call cost and reprices only when absent; tolerant of malformed lines and large state snapshots - Full registry parity: frontend constants (display name, logo, color), SUPPORTED_CLIENT_TYPES, TUI client UI + hotkey; clientRegistry.test.ts green - Docs: README EN + ja/ko/zh client tables, multi-platform list, --client values, data-location table
makoMakoGo
added a commit
to makoMakoGo/tokscale
that referenced
this pull request
Jun 23, 2026
ported from upstream junhoyeo#727 ported from upstream junhoyeo#749
t1000040
pushed a commit
to tmobi-internal/tokscale
that referenced
this pull request
Jun 30, 2026
Adds the JetBrains Junie coding agent as a first-class client. - Session source: ~/.junie/sessions/*/events.jsonl (ClientId::Junie = 32, COUNT -> 33) - Parser reads modelUsage cost/tokens/turn-start with provider inference; preserves Junie's authoritative per-call cost and reprices only when absent; tolerant of malformed lines and large state snapshots - Full registry parity: frontend constants (display name, logo, color), SUPPORTED_CLIENT_TYPES, TUI client UI + hotkey; clientRegistry.test.ts green - Docs: README EN + ja/ko/zh client tables, multi-platform list, --client values, data-location table
t1000040
pushed a commit
to tmobi-internal/tokscale
that referenced
this pull request
Jun 30, 2026
…unhoyeo#749) * fix(junie): unique dedup ordinal, scoped turn-start, kind-based skip Three confirmed bugs in the Junie session parser (PR junhoyeo#727): (a) The dedup key embedded `usage_index` from `usages.iter().enumerate()`, which resets to 0 on every `modelUsage` row. Two distinct LLM calls with identical token counts produced the same key and the second was silently dropped. Replace it with a file-global monotonic ordinal so every usage row across the whole session earns a unique key. (b) `pending_turn_start` was set by a UserPromptEvent and only cleared when a usage message was actually pushed. A prompt that produced no counted usage leaked the flag onto a later, unrelated turn's usage event. Clear it once per usage event (the first response event consumes the prompt's turn-start) and mark only the first emitted message of that event. (c) SKIP_EVENT_KINDS was matched via `line.contains(kind)`, a raw substring search that could drop legitimate usage/prompt lines merely mentioning a skipped kind in their text. Match on the parsed event kind instead, keeping the substring filter only as a cheap pre-parse guard. Adds regression tests, including two distinct usage rows with identical token counts that must both be counted (fails without the ordinal fix). Confidence: high Scope-risk: narrow Not-tested: real multi-megabyte Junie state snapshots (perf path) * fix(junie): derive dedup suffix from within-event index, not file-global ordinal The file-global `usage_ordinal` gave a replayed identical `LlmResponseMetadataEvent` a new suffix (:0 then :1), so both the original and the replay passed the in-file `seen` set and the cross-file `should_keep_deduped_message` filter — double-counting the same tokens/cost. Revert the uniqueness component to the row's position within the event's own `modelUsage` array. A replayed event reproduces the same dedup key (collapsed to one), while multiple distinct rows inside one event keep distinct indices. Genuinely-distinct calls already differ via `timestampMs` in the key, so the junhoyeo#727 undercount fix is preserved. Constraint: cross-file Junie dedup relies on dedup_key being stable across replays Rejected: process-global monotonic counter | breaks replay collapse (regresses this PR's own fix) Confidence: high Scope-risk: narrow Not-tested: two byte-identical events with identical timestamps are treated as a replay (indistinguishable from a true replay)
This was referenced Jul 6, 2026
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
This add support for JetBrains Junie.
Summary by cubic
Adds first-class support for JetBrains Junie. Tokscale now scans
~/.junie/sessions/*/events.jsonl, imports usage with embedded costs (falls back to pricing when missing), and surfaces it in CLI, TUI, and the web UI.junie; scansevents.jsonl, parses usage events, deduplicates, infers provider, marks turn starts, preservesmodelUsage.cost; tests added.--client junie, proper display name and logo; tests added.#7B61FFand hotkeyp.https://github.com/JetBrains.png), color, and type added.Written for commit e60a16f. Summary will update on new commits.