Repository navigation
Feat/3.1.0 beta1 - #2699
Merged
Merged
Feat/3.1.0 beta1#2699
Conversation
Validate optional workbench config safely and preserve original TGZ bytes with structured errors. Support sequential imports of up to 50 workbench TGZ files, 100 MiB per file, same-version plugin/workbench overwrites, and revision-checked workbench deletion. Fix cached-task overwrite races and confirmation lifecycle handling. Include three-language copy, regression tests, API contracts and 109 deployment evidence. Verified 118 backend tests on frozen Python 3.11, 100 frontend tests, quality gates and production build.
… handling - Parse and validate optional screenshot declarations (dsh.workbench.screenshots) from package.json without disk extraction - Enforce screenshot format validation (PNG, JPEG, WebP), dimension and payload limits - Store and retrieve workbench screenshots by image SHA-256 - Add screenshot HTTP endpoint with caching headers and browser auth token extraction - Add comprehensive test suites for screenshots, browser auth, and deleted equivalent builds - Document enterprise workbench screenshot protocol and verification
Merge codex/custom-web-search after development completed. Adapt the shared client settings to versioned search settings and unified image generation, retaining legacy tool references.
删除成功后计数归零,页面不再把赞或踩画成仍然存在。 Co-authored-by: Cursor <cursoragent@cursor.com>
## Summary Fix DSH account availability updates so authorization is evaluated against the exact target tool's owning category. Previously, the endpoint enumerated the visible catalog and could reject a valid target when the catalog projection omitted it, or authorize based on the wrong visible set. The update now loads the target tool and category directly, enforces tenant ownership/deletion/availability checks, calls the business authorization service for the category, and preserves the non-disclosing 404 behavior for unavailable resources. ## Validation - `ruff check` passed for changed backend files - `ruff format --check` passed - `scripts/arch-guard.sh` passed - `py_compile` passed - Focused pytest startup was blocked by the local x86_64 pandoc binary on this ARM host; no test assertion failure was observed Target branch: `feat/3.1.0-beta1`
## Summary - Replace full tool catalog enumeration during an account availability switch with exact tool/category loading and one category-scoped `check_business_action(use)` decision. - Authorize the owning category ID (F048 tool resource), not the child tool ID. Retain current-tenant and deletion checks, account preference persistence, and audit behavior. - Preserve the existing non-disclosing HTTP 404 for denied/unavailable targets; propagate authorization infrastructure failures. ## Why Saving one switch currently calls get_tool_list(action="use"), loading and checking every category before finding the requested child. Production traces reported 11–14s saves and 22 permission HTTP requests / 1,099 check items. This change removes unrelated resource checks; no post-fix production timing is claimed. ## Scope Hotfix, no API/schema contract change. Does not modify catalog listing, frontend refresh, MCP invocation, or permission runtime. Based directly on feat/3.1.0-beta1. ## Validation - Focused regression suite: **74 passed, 9 subtests passed** (5 files: availability target, capability catalog, work capabilities, user availability, executor availability). - Existing selection against original release service: **55 passed, 9 subtests passed**. - 19 new parameterized cases cover API/builtin/MCP enable+disable, exact category ID, no full enumeration, invalid/deleted/cross-tenant targets, revoked permission, unavailable permission targets, and infrastructure failures without writes/audits. - Ruff format/check, architecture guard, git diff --check: PASS. - Independent seven-dimension sdd-code-review: PASS; error-contract finding fixed and re-reviewed. ### Local test environment Used the existing Python 3.12 venv and an untracked empty settings file. Stubbed only unrelated pypandoc.get_pandoc_path in the test launcher because the local x86 Pandoc discovery triggers exit() before pytest collection; JUnit output confirmed actual execution. No live database/OpenFGA/DM8 integration test, production deployment, or tool toggle was performed. CI should run supported-runtime integration checks.
# Conflicts: # src/backend/bisheng/dsh/domain/services/model_catalog.py # src/backend/bisheng/workstation/api/endpoints/config.py
## What 简要描述做了什么改动。 ## Why 为什么需要这个改动? ## How 实现方式、设计决策(如有)。 ## Test - [ ] 本地测试通过 - [ ] 114 测试服务器验证通过 ## Related - Issue/ticket:
grep lists `path` first, and ls ends the prefix at a directory boundary (781f5b0), so grep(pattern, path="/uploads/x.md") listed nothing and answered "No matches found" for every pattern. On 106 (2026-11-16) one open-API session got 29 empty greps, read an 8489-line file 200 lines at a time, compressed its context four times and lost the task. When the listing is empty, grep now looks up the object stored exactly at `path` and searches it. Directory paths behave as before. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ession
After a compression the task instruction survived only as the summary
model's paraphrase, and each chained compression summarised the previous
summary. On 106 (2026-11-16) the third of four compressions restated an
open-API task ("return three JSON arrays, write no files") as "review this
prospectus for investment value", and the run wrote a report instead.
BishengSummarizationMiddleware now puts the user's own messages that a
compression cuts off into the summary message verbatim, and forwards them
from the previous summary message on every later compression (also on the
compact tool path). The summary text follows them unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Any HTTP 500 or business status_code 500 used to raise the full-screen service-maintenance overlay, so one failing endpoint (e.g. a linsight file_download 500) replaced the whole page until reload. The backend's resp_500 also uses code 500 for ordinary endpoint errors. A 500 now triggers one shared probe of the public /api/v1/env endpoint; only when the probe fails too is the overlay raised. Otherwise the error is rejected to its caller like any other business error and toasted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
## Summary Bring the committed DSH and shared tool work from `v3.0.0-fix-dsh` into the 3.1 beta release line. This branch contains the DSH market/workbench fixes, usage batching and quota dispatch changes, shared client tool support, custom web search, image generation, and their backend/frontend tests and release documentation. The PR includes only commits already present on `v3.0.0-fix-dsh`; uncommitted files in the local worktree are excluded. ## Validation The branch includes focused backend and frontend regression tests and local verification artifacts for the included DSH/tool changes. The release target is `feat/3.1.0-beta1`.
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
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.
What
简要描述做了什么改动。
Why
为什么需要这个改动?
How
实现方式、设计决策(如有)。
Test
Related