fix: Windows-native compatibility for the derive-reorganization branch by wangdahoo · Pull Request #63 · bybit-exchange/kaas

wangdahoo · 2026-09-09T13:27:36Z

Summary

Cherry-picks the Windows-native compatibility fixes (originally developed on
fix/windows-native-build, merged into the local testing branch) on top of
feat/derive-knowledge-reorganization, so this branch also builds and runs
cleanly on a zh-CN Windows machine. All five commits apply without conflicts
and the full suite is green natively.

build natively on Windows — split the Unix-only process-group setup
(syscall.Setpgid, group SIGKILL) into daemon_unix.go /
daemon_windows.go (taskkill tree-kill), and guard the SIGPIPE
registration in the Python entrypoint.
keep CJK payloads intact — a piped Python daemon on Windows decodes
stdin with the ANSI code page (cp936) + surrogateescape, corrupting the
Go bridge's UTF-8 JSON into lone surrogates, which then fail the strict
UTF-8 encode of the LLM request (extract: ... UnicodeEncodeError ... '\udcac' ... surrogates not allowed). server_daemon.main() now pins
stdin/stdout/stderr to UTF-8.
UTF-8 for every KB text file regardless of OS locale — explicit
encoding="utf-8" on all bare text IO in py/src and py/tests
(master/topic indexes, timeline, caches, compile state, people stubs,
compile log, manifests, test fixtures). Without this, files land on disk
as GBK on Chinese Windows and every strict UTF-8 reader downstream
(including the Go bridge and the web UI) breaks.
stop Windows Path.resolve() races from faking kb_dir escapes —
under parallel extraction, resolve() intermittently returns
\?\-prefixed verbatim paths while another thread mkdirs a parent
directory, which fails is_relative_to(base) and makes compile drop
random documents (~3-12% of suite runs failed on rotating tests).
_strip_verbatim() now removes the prefix before containment checks
(KBStore._resolve and resolve_kb_dir).

Test plan

Full py suite green natively on zh-CN Windows: 2087 passed, 2 skipped
(one skip is NTFS-forbidden newline dir names)
go build ./... and go test ./internal/bridge/ pass natively on Windows
Threaded stress loop over KBStore._resolve with concurrent mkdir:
escapes 0 (intermittent before)

Co-Authored-By: Claude Opus 4.7 noreply@anthropic.com

The daemon lifecycle code in internal/bridge used Unix-only syscall.Setpgid and group SIGKILL, breaking the native Windows build of internal/bridge, internal/worker and cmd/kaas; the Python entrypoint also registered SIGPIPE unconditionally, which Windows Python does not provide. - Split the process-group setup into daemon_unix.go and add daemon_windows.go, which tree-kills via taskkill /T /F and falls back to Process.Signal when the process already exited. - Guard the SIGPIPE registration with hasattr(signal, "SIGPIPE"). With these, `go run ./cmd/kaas -f etc/kaas-dev.toml` plus `pnpm dev` run the full stack on Windows; `make dev` keeps working on Unix. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

The daemon's wire protocol is UTF-8 JSON lines, but a piped Python process on Windows decodes stdin with the ANSI code page (cp936 on Chinese Windows) and errors='surrogateescape'. The UTF-8 bytes the Go bridge writes became mojibake laced with lone surrogates, which then failed the strict UTF-8 encode of the LLM request: extract: bridge: AI engine error INTERNAL_ERROR: UnicodeEncodeError: 'utf-8' codec can't encode character '\udcac': surrogates not allowed - server_daemon.main() now reconfigures stdin/stdout/stderr to UTF-8 before serving (streams without reconfigure(), like StringIO in tests, are left alone). - KBStore built rel_path via str(relative_to(...)), which yields backslash separators on Windows, so the CLI route keyed extractions as raw\a.md while the worker route keyed them raw/a.md. Both sites now use Path.as_posix(), restoring the byte-identical-extraction contract between the two ingestion routes. - The parity test read extraction files with the locale codec; it now reads them as UTF-8 explicitly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Python text IO without an explicit encoding follows the OS locale: UTF-8 on the Linux CI, but cp936 (GBK) on Chinese Windows. On that platform the engine wrote GBK bytes for master-index.md, topic indexes, timeline, classify cache, compile state, article writes, people stubs, the compile log and derived manifests, which then broke every strict UTF-8 reader downstream — including the Go bridge and the web UI — and failed ~110 py tests natively. - Add encoding="utf-8" to every bare read_text/write_text/open text call in py/src and py/tests. - KBStore.extraction_rel_path built paths with str(Path(...)), which yields backslash separators on Windows; use a slash join so the layer's rel-path keys match the Go/daemon form everywhere. - copy_documents now rejects POSIX-style absolute rel_paths too — Windows pathlib treats "/etc/passwd" as relative, so the previous is_absolute() guard let it slip through to the escape check and surface as a ValueError instead of the specified DeriveError. - test_ask_uses_kb_dir compares resolved forms (resolve() rewrites a POSIX-style root to a drive path on Windows); the trailing-newline list_derived test is skipped on NTFS, which forbids such names. py suite on zh-CN Windows: 107 failing -> 3 (all three are a pre-existing intermittent write-phase race that also reproduces without these changes). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

These two test files only exist on this branch (added with the derive-reorganization merge), so the locale sweep on fix/windows-native-build could not cover them. Their fixtures wrote master-index and article files with the cp936 default, which broke once the product side started reading and writing strict UTF-8. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Under a parallel-extract workload on Windows, Path.resolve() intermittently returns a \?\-prefixed verbatim path when another thread creates a parent directory (e.g. extraction/) at the same moment. The verbatim prefix then fails every is_relative_to(base) comparison, so KBStore._resolve raised "path escapes kb_dir" for files inside the KB and compile lost random documents (reproducible: ~3 in 12 suite runs failed on different tests each time). _resolve and resolve_kb_dir now strip the \?\ and \?\UNC\ prefixes before comparing. A 8-thread stress loop over _resolve with a concurrent mkdir went from intermittent escapes to zero, and the flaky compile test files went from 3-5 failures per 12 runs to 0. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…ation (#64) * fix(upload): skip unsupported files inside ZIP instead of rejecting entire archive Previously, if a ZIP contained any file with an extension not in allowedInnerExtensions (e.g. .jpg, .png), the entire ZIP was rejected with "extension not allowed". This made it impossible to upload ZIPs containing a mix of documents and non-document files. Now unsupported entries are silently skipped and only supported files are extracted and processed. * fix(derive): hash-based fallback slug for pure CJK topics slugFromTopic (Go) and normalise_slug (Python) now produce a deterministic "t-{sha256[:10]}" slug when the regex normalization strips every character (e.g. pure Chinese topics like '定价'). Previously these topics were rejected with 'invalid slug'. * fix(worker): don't burn retry attempts on circuit breaker rejection Two bugs caused a cascading failure where 126 out of 136 failed tasks were killed by 'breaker open' without ever reaching the LLM: 1. Extract phase treated circuit.ErrOpen as a regular failure, calling Nack and burning a retry attempt. The pipeline phase already handled this correctly by abandoning without Nack. Apply the same pattern to extract: abandon and let RecoverExpired requeue the task. 2. Dispatcher claimed maxConc tasks per tick in half-open state, but the breaker only allows one in-flight probe — the rest were instantly rejected. Limit claiming to 1 task when the breaker is half-open. Root cause: transient LLM outage (timeouts + connection errors) tripped the breaker; the 30s cooldown/half-open cycle then permanently killed queued tasks every cycle by burning all 3 retry attempts on instant ErrOpen rejections. * docs(plan): add technical plan - derive knowledge reorganization * feat(derive): add data structures and entity index builder (Feature: p1-feat-001) * feat(classify): add topic-aware classify prompt and function (Feature: p1-feat-004) * feat(derive): add entity name resolution with LLM canonicalization (Feature: p1-feat-002) * feat(derive): add aggregation planning and reorganize orchestrator (Feature: p1-feat-003) * feat(derive): wire reorganize into compile_kb and derive_kb (Feature: p1-feat-005) * docs(plan): add technical plan - 2026-09-07-derive-split-oversized-merge * feat(merge): add budget estimation helpers (Feature: p2-feat-001) * feat(compile): add batch packing and iterative merge (Feature: p2-feat-002) * feat(compile): wire batch splitting into _process_article (Feature: p2-feat-003) * test(compile): add batch splitting tests (Feature: p2-feat-004) * docs(plan): add technical plan - 2026-09-07-derive-sub-article-split * feat(compile): add sub-article threshold and helpers (Feature: p3-feat-001) * feat(compile): add sub-article support to _merge_batch_split (Feature: p3-feat-002) * feat(compile): wire sub-article splitting into _process_article (Feature: p3-feat-003) * test(compile): add sub-article splitting tests (Feature: p3-feat-004) * fix(compile): cap per-batch budget to prevent LLM timeout (Feature: p4-feat-001) * fix(compile): lower _MAX_BATCH_BUDGET to 20K to prevent merge-path timeouts * feat(derive): add reorganize flag and derive-flow diagrams - .gitignore: exclude .ph directory - server_daemon: hardcode reorganize=True in _handle_derive for validation - docs/assets: add derive-flow SVG diagrams (en + zh) * docs(plan): add technical plan - 20260908-derive-reorganize-config-v2 * feat(daemon): read reorganize from payload instead of hardcoding (Feature: p5-feat-003) * feat(config): add DeriveConf with reorganize setting (Feature: p5-feat-001) * feat(derive): thread reorganize config through bridge and runner (Feature: p5-feat-002) * test(daemon): add reorganize parameter tests for _handle_derive (Feature: p5-feat-005) * test(derive): add Go tests for reorganize config, bridge, and runner (Feature: p5-feat-004) * docs(plan): normalize date format to yyyy-mm-dd * docs(plan): add technical plan - 2026-09-09-batch-parallel-write * feat(compile): add batch-parallel infrastructure (Feature: p6-feat-001) Add _ChainResult dataclass, _get_batch_parallel_sem() semaphore factory, _sem_guard context manager, and _pre_split_chains chunking function. These are the building blocks for parallelizing batch-level LLM calls within _merge_batch_split. The semaphore follows the exact pattern of _get_section_sem() in merge.py: reads KB_BATCH_PARALLEL_MAX_CONCURRENT per call, warns once on invalid values, caches keyed on bound. Tests cover semaphore lifecycle (default, env override, invalid values), _sem_guard acquire/release/noop/exception-safety, _pre_split_chains chunking correctness, order preservation, and edge cases. * feat(compile): implement _run_chain for batch-parallel chains (Feature: p6-feat-002) * feat(compile): refactor _merge_batch_split to three-phase parallel structure (Feature: p6-feat-003) * test(compile): add comprehensive batch-parallel tests (Feature: p6-feat-004) * fix(derive): auto-deduplicate slug on conflict instead of erroring When normalise_slug generates a slug from a pure-CJK topic (or any topic that produces a deterministic hash-based slug like t-e999a859f8), re-running the same topic would hit SlugExistsError. The user had to manually pass --force or --slug to retry. Now, when the slug is auto-generated (not user-provided) and force=False, derive_kb() automatically appends -2, -3, etc. until a free slug is found. User-provided slugs still error on conflict (existing behavior preserved). Python: _deduplicate_slug() in _layout.py, wired into derive_kb(). Go: deduplicateSlug() in derive.go for the HTTP API handler. Tests: 7 new tests covering no-conflict, single/double conflict, truncation, validation, user-slug preservation, and Go-side dedup. * docs(plan): add technical plan - 2026-09-09-parallel-reorg * feat(reorganize): add aggregate-plan-feedback prompt template (Feature: p7-feat-002) * feat(compile): add batch_parallel_threshold for serial fallback (Feature: p7-feat-001) * feat(reorganize): add two-round feedback to plan_aggregation (Feature: p7-feat-003) * docs(plan): add technical plan - 2026-09-09-builds-page-redesign * feat(i18n): add Builds page i18n strings (Feature: p8-feat-007) * feat(web): extend derived API client and add build types (Feature: p8-feat-002) * feat(api): add ListDerivedJobsPaged and DeleteDerivedJob endpoints (Feature: p8-feat-001) * feat(web): add shared build components (useAutoPolling, StatusStageBadge, StatsBar) (Feature: p8-feat-003) * feat(web): add DeriveJobsTab and DeriveJobDetailDialog components (Feature: p8-feat-005) * feat(web): extract TasksTab and TaskDetailDialog components (Feature: p8-feat-004) * feat(web): add Builds page shell, routing, nav, and simplify DeriveDialog (Feature: p8-feat-006) * fix(web): adjust StatsBar tab sizing and DeriveJobsTab column widths * fix(web): set appropriate column widths for Builds tables (Feature: col-widths) * fix(web): widen Attempts column (Feature: attempts-width) * fix(web): make StatsBar tabs inline instead of full-width * fix(web): add divider and padding around StatsBar * fix(web): horizontal layout for tabs and search bar with divider * docs(plan): add technical plan - 2026-09-09-write-phase-early-parallel * fix(web): remove divider from filter bar * feat(compile): add batch_parallel_batch_limit config and lower default concurrency (Feature: p9-feat-001) * feat(compile): add batch-count early break to Phase A and _run_chain (Feature: p9-feat-002) * fix: Windows-native compatibility for the derive-reorganization branch (#63) * fix: build and run the dev stack natively on Windows The daemon lifecycle code in internal/bridge used Unix-only syscall.Setpgid and group SIGKILL, breaking the native Windows build of internal/bridge, internal/worker and cmd/kaas; the Python entrypoint also registered SIGPIPE unconditionally, which Windows Python does not provide. - Split the process-group setup into daemon_unix.go and add daemon_windows.go, which tree-kills via taskkill /T /F and falls back to Process.Signal when the process already exited. - Guard the SIGPIPE registration with hasattr(signal, "SIGPIPE"). With these, `go run ./cmd/kaas -f etc/kaas-dev.toml` plus `pnpm dev` run the full stack on Windows; `make dev` keeps working on Unix. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix: keep CJK payloads intact when running natively on Windows The daemon's wire protocol is UTF-8 JSON lines, but a piped Python process on Windows decodes stdin with the ANSI code page (cp936 on Chinese Windows) and errors='surrogateescape'. The UTF-8 bytes the Go bridge writes became mojibake laced with lone surrogates, which then failed the strict UTF-8 encode of the LLM request: extract: bridge: AI engine error INTERNAL_ERROR: UnicodeEncodeError: 'utf-8' codec can't encode character '\udcac': surrogates not allowed - server_daemon.main() now reconfigures stdin/stdout/stderr to UTF-8 before serving (streams without reconfigure(), like StringIO in tests, are left alone). - KBStore built rel_path via str(relative_to(...)), which yields backslash separators on Windows, so the CLI route keyed extractions as raw\a.md while the worker route keyed them raw/a.md. Both sites now use Path.as_posix(), restoring the byte-identical-extraction contract between the two ingestion routes. - The parity test read extraction files with the locale codec; it now reads them as UTF-8 explicitly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix: make every KB text file UTF-8 regardless of OS locale Python text IO without an explicit encoding follows the OS locale: UTF-8 on the Linux CI, but cp936 (GBK) on Chinese Windows. On that platform the engine wrote GBK bytes for master-index.md, topic indexes, timeline, classify cache, compile state, article writes, people stubs, the compile log and derived manifests, which then broke every strict UTF-8 reader downstream — including the Go bridge and the web UI — and failed ~110 py tests natively. - Add encoding="utf-8" to every bare read_text/write_text/open text call in py/src and py/tests. - KBStore.extraction_rel_path built paths with str(Path(...)), which yields backslash separators on Windows; use a slash join so the layer's rel-path keys match the Go/daemon form everywhere. - copy_documents now rejects POSIX-style absolute rel_paths too — Windows pathlib treats "/etc/passwd" as relative, so the previous is_absolute() guard let it slip through to the escape check and surface as a ValueError instead of the specified DeriveError. - test_ask_uses_kb_dir compares resolved forms (resolve() rewrites a POSIX-style root to a drive path on Windows); the trailing-newline list_derived test is skipped on NTFS, which forbids such names. py suite on zh-CN Windows: 107 failing -> 3 (all three are a pre-existing intermittent write-phase race that also reproduces without these changes). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * test: read/write derive-branch test fixtures as UTF-8 These two test files only exist on this branch (added with the derive-reorganization merge), so the locale sweep on fix/windows-native-build could not cover them. Their fixtures wrote master-index and article files with the cp936 default, which broke once the product side started reading and writing strict UTF-8. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix: stop Windows resolve() races from faking kb_dir escapes Under a parallel-extract workload on Windows, Path.resolve() intermittently returns a \?\-prefixed verbatim path when another thread creates a parent directory (e.g. extraction/) at the same moment. The verbatim prefix then fails every is_relative_to(base) comparison, so KBStore._resolve raised "path escapes kb_dir" for files inside the KB and compile lost random documents (reproducible: ~3 in 12 suite runs failed on different tests each time). _resolve and resolve_kb_dir now strip the \?\ and \?\UNC\ prefixes before comparing. A 8-thread stress loop over _resolve with a concurrent mkdir went from intermittent escapes to zero, and the flaky compile test files went from 3-5 failures per 12 runs to 0. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * docs(plan): add technical plan - 2026-09-09-builds-master-detail.md * feat(api): add AbortSignal support to getTask (Feature: p10-feat-002) * feat(i18n): add Sheet description entries for builds detail (Feature: p10-feat-003) * feat(ui): extend Sheet component with close button and SheetDescription (Feature: p10-feat-001) * feat(builds): create TaskDetailSheet and DeriveJobDetailSheet (Feature: p10-feat-004) * feat(builds): refactor tabs to use Sheet detail with clickable rows (Feature: p10-feat-005) * refactor(builds): remove obsolete Dialog detail components (Feature: p10-feat-006) * test(builds): update tests for Sheet components and clickable rows (Feature: p10-feat-007) * docs(plan): add technical plan - 2025-09-10-builds-tab-refactor.md * feat(ui): update StatusStageBadge for partial status and optional stage (Feature: p11-feat-008) * feat(i18n): add build job keys and rename tab labels (Feature: p11-feat-009) * feat(api): create build jobs API client (Feature: p11-feat-007) * feat(store): add BuildJob model and BuildJobStore interface (Feature: p11-feat-001) * feat(ui): create BuildJobDetailSheet component (Feature: p11-feat-011) * feat(ui): create BuildJobsTab component (Feature: p11-feat-010) * feat(store): implement SQLite build_jobs table and store methods (Feature: p11-feat-002) * test(store): add build_jobs store tests (Feature: p11-feat-003) * feat(api): add build job API handlers and routes (Feature: p11-feat-004) * feat(ui): update Builds page to use BuildJobsTab (Feature: p11-feat-012) - Replace TasksTab import/usage with BuildJobsTab in Builds.tsx - Replace listTasks stats calls with listBuildJobs for Normal tab counts - Delete TasksTab.tsx and its test file - Update Builds.test.tsx to mock listBuildJobs instead of listTasks - Fix test assertions to match current i18n tab labels (Normal, Derive Topic) * feat(api): create BuildJob on submit (Feature: p11-feat-005) * test(builds): update all frontend tests for tab refactor (Feature: p11-feat-013) - StatusStageBadge.test.tsx: add partial status amber badge tests, optional stage tests - BuildJobsTab.test.tsx: new test file covering table rendering, sort, search, pagination, delete confirmation, detail sheet, partial in filter/delete - BuildJobDetailSheet.test.tsx: new test file covering job detail display, task list rendering, nested TaskDetailSheet, error indicator - BuildJobDetailSheet.tsx: guard against undefined tasks in optimistic display - StatsBar.test.tsx, Builds.test.tsx: already updated by prior features - TasksTab.test.tsx: already deleted by prior features * test(api): add build job API tests and update submit tests (Feature: p11-feat-006) * fix: enable TOC sidebar scrolling when content overflows Add overflow-y-auto to the TOC aside so long tables of contents can scroll independently. Remove the ineffective sticky positioning on the inner wrapper since the aside itself is the scroll container. * docs(plan): add technical plan - 2026-09-10-chat-kb-switcher * feat(ui): update Session type, API functions, and KB store (Feature: p12-feat-005) * feat(store): add KBSlug to Session model and update SQLite schema (Feature: p12-feat-001) * test(store): update session tests for kb_slug filter (Feature: p12-feat-002) * feat(ui): create ChatKBSelector component (Feature: p12-feat-006) * test(api): update session tests for kb_slug support (Feature: p12-feat-004) * feat(ui): update SessionList and Chat.tsx for KB-scoped sessions (Feature: p12-feat-007) * test(ui): update frontend tests for KB switcher (Feature: p12-feat-008) * docs(plan): add technical plan - 2026-09-10-derive-quality-fix.md * feat(prompts): add language directives to 6 prompts (Feature: p13-feat-004) * feat(derive): implement multi-round voting in select_by_topic (Feature: p13-feat-001) * test(merge): add regression test for _create_system language directive (Feature: p13-feat-005) * test(derive): add multi-round voting test suite (Feature: p13-feat-003) * feat(derive): propagate filter_rounds through derive_kb, CLI, and daemon (Feature: p13-feat-002) * fix: make classify tokenizer CJK-aware with character bigrams _title_words() stripped all CJK characters, causing _relevance_score() to return 0 for every Chinese title and dedup_create_new() to skip all Chinese titles. Together these made the classifier create one wiki article per source file instead of merging related content. Add _tokenize() that produces character bigrams for CJK runs (standard IR technique, comparable granularity to Latin words) and splits Latin runs on non-alphanumeric boundaries. _title_words() and the inline topic tokenization in _relevance_score() both delegate to it. Side fix: hyphenated topic tags like 'sun-wukong' now split into {'sun', 'wukong'} instead of collapsing to {'sunwukong'}, so they match article titles correctly. * fix(derive): wire filter_rounds from TOML config through Go to Python daemon * test(derive): add filter stability test script from diagnosis * docs(plan): add technical plan - 2026-09-10-cr-fixes * style(api): fix import grouping in session.go (Feature: p14-feat-002) * fix(api): replace log.Printf with structured slog (Feature: p14-feat-001) * refactor(ui): remove unused kbSlug prop from SessionList (Feature: p14-feat-005) * refactor(ui): extract shared formatDate utility (Feature: p14-feat-003) * refactor(store): extract duplicated status CASE/WHEN SQL to const (Feature: p14-feat-006) * refactor(api): embed buildJobDTO in buildJobDetailDTO (Feature: p14-feat-004) * refactor(store): extract generic paged query builder (Feature: p14-feat-007) * refactor(ui): extract shared PagedJobTable component (Feature: p14-feat-008) * docs(plan): add technical plan - 2026-09-10-derive-drawer-article-tasks.md * feat(derive): make filter voting threshold configurable via filter_threshold Add filter_threshold (float64, range (0,1]) to the full config chain: TOML [derive] section, KAAS_DERIVE_FILTER_THRESHOLD env override, Go bridge DeriveRequest, Python CLI --filter-threshold, and the select_by_topic engine function. When set, the acceptance threshold becomes ceil(filter_rounds * filter_threshold). When unset (0), the existing supermajority formula ceil(2N/3) is preserved, so the change is fully backward-compatible. Examples with filter_rounds=5: filter_threshold=0.5 → threshold=3 (simple majority) filter_threshold=0.67 → threshold=4 (supermajority, current default) filter_threshold=1.0 → threshold=5 (unanimity) * feat(i18n): add derive article table keys (Feature: p15-feat-001) * feat(ui): create DeriveWikiPreviewSheet component (Feature: p15-feat-002) * test(ui): add DeriveWikiPreviewSheet tests (Feature: p15-feat-004) * feat(ui): add article table to DeriveJobDetailSheet (Feature: p15-feat-003) * test(ui): add DeriveJobDetailSheet article table tests (Feature: p15-feat-005) * docs(plan): add UI polish plan - 2026-09-10-builds-ui-polish.md * style(ui): unify column widths to 160px (Feature: p16-feat-002) * feat(ui): add immediate thinking indicator in chat (Feature: p16-feat-005) * style(ui): fix tab label, badge spacing, sheet widths (Feature: p16-feat-004) * feat(ui): replace row click with details button (Feature: p16-feat-003) * feat(ui): restyle status badges with WCAG-compliant colors (Feature: p16-feat-001) --------- Co-authored-by: 王大虎 <157195705@qq.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

wangdahoo and others added 5 commits September 9, 2026 21:26

tybot02 approved these changes Sep 9, 2026

View reviewed changes

tybot02 merged commit 7594157 into bybit-exchange:feat/derive-knowledge-reorganization Sep 9, 2026
10 checks passed

tybot02 mentioned this pull request Sep 10, 2026

feat: derive knowledge reorganization with entity-driven topic aggregation #64

Merged

4 tasks

Navigation Menu

fix: Windows-native compatibility for the derive-reorganization branch - #63

fix: Windows-native compatibility for the derive-reorganization branch#63
tybot02 merged 5 commits into
bybit-exchange:feat/derive-knowledge-reorganizationfrom
wangdahoo:fix/windows-native-build-on-derive

wangdahoo commented Sep 9, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

wangdahoo commented Sep 9, 2026

Summary

Test plan

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants