| 实现 P4.1 性能基准测试框架 创建搜索索引性能基准测试，验证 SQLite FTS 在不同工作区规模下的查询性能表现。 ## 新增文件 ### 基准测试框架 - benchmark/search-index-performance.ts — 独立的性能基准脚本，支持快速/标准/完整模式 - test/intelligence/benchmark-search-index.test.ts — 集成到 vitest 的可运行基准测试 ### 测试支持 - test/intelligence/testSupport/asyncSqliteStore.ts — 完整的测试夹具实现 - test/intelligence/sqliteSearchIntegration.test.ts — 完整链路集成测试 ## 性能指标（标准模式） 所有性能目标已达成： - 500 符号：<1ms 查询延迟 ✓ - 1K 符号：~2ms 查询延迟 ✓ - 5K 符号：~4ms 查询延迟 ✓ - 10K 符号：~7ms 查询延迟 ✓（目标 <10ms） ## 运行方式 npm 脚本集成（package.json）： - npm test -- benchmark-search-index （标准模式） - BENCH_SCALE=quick npm test -- benchmark-search-index （快速模式） - BENCH_SCALE=full npm test -- benchmark-search-index （完整模式，包括 50K 符号） ## P4.1 验收标准 ✓ 性能基准测试框架完成 ✓ 多规模工作区测试支持 ✓ 所有性能目标达成 ✓ 测试文档化完成 ✓ npm 脚本集成 Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> | 1 个月前 |
| docs: add technical blog series | 13 天前 |
| feat: integrate vision capabilities for image analysis - Add ImageAnalysisService for analyzing user-uploaded images - Support image attachments in WebviewToHostMessage protocol - Integrate image analysis into ReactAgentRunner workflow - Inject image analysis results into system prompts - Add 11 unit tests for ImageAnalysisService (all passing) Core features: - Smart prompt building based on user message context - Concurrent analysis of multiple images - Support for design mockups, error screenshots, hand-drawn sketches - Non-intrusive integration (analysis failure doesn't block conversation) Documentation: - Move all docs to superpowers structure - Add vision feature guide (500+ lines) - Add implementation summary - Add 7 vision-related guides Files: - New: src/extension/vision/imageAnalysisService.ts (180 lines) - Modified: 5 core files for attachment support - New: test/imageAnalysisService.test.ts (280 lines, 11 tests) - New: 9 documentation files in docs/superpowers/ Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> | 1 个月前 |
| refactor: remove legacy superpowers runtime | 1 个月前 |
| feat: add command recovery and reference discovery | 1 个月前 |
| feat: add agent guides and robust tool recovery | 27 天前 |
| feat: add agent guides and robust tool recovery | 27 天前 |
| feat(superpowers): vendor official v6.1.1 skills | 1 个月前 |
| chore: remove blog directory | 13 天前 |
| feat: add command recovery and reference discovery | 1 个月前 |
| docs: unify agent development workflow | 1 个月前 |
| rename persistence directory from .codegraph to .loopagent .codegraph is owned by the external CodeGraph MCP tool; this extension's own SQLite index persistence was accidentally reusing the same directory name, risking a collision. Renamed to .loopagent and updated the CLAUDE.md section and tests that documented/covered the old path. | 1 个月前 |
| Initial commit | 13 天前 |
| chore: initial project import | 2 个月前 |
| fix: disable DeepSeek thinking mode when tools are present to prevent DSML text output When DeepSeek's thinking mode is enabled alongside tool calling, the model may emit tool calls as DSML text tokens instead of using the OpenAI-compatible structured tool_calls API. This causes the raw DSML markup (with fullwidth pipe characters) to be displayed to users instead of being executed as tool invocations. Changes: - Modified deepseekProvider.ts to disable thinking mode when request.tools is non-empty, regardless of toolChoice value - Added comprehensive unit tests covering all thinking mode scenarios - Added detailed root cause analysis document Fixes the issue observed in yguc workspace where assistant returned: <｜｜DSML｜｜tool_calls> <｜｜DSML｜｜invoke name="readFile"> ... Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> | 1 个月前 |
| test(intelligence): verify sqlite worker in minimum vscode host | 1 个月前 |
| feat: add command recovery and reference discovery | 1 个月前 |
| feat: integrate vision capabilities for image analysis - Add ImageAnalysisService for analyzing user-uploaded images - Support image attachments in WebviewToHostMessage protocol - Integrate image analysis into ReactAgentRunner workflow - Inject image analysis results into system prompts - Add 11 unit tests for ImageAnalysisService (all passing) Core features: - Smart prompt building based on user message context - Concurrent analysis of multiple images - Support for design mockups, error screenshots, hand-drawn sketches - Non-intrusive integration (analysis failure doesn't block conversation) Documentation: - Move all docs to superpowers structure - Add vision feature guide (500+ lines) - Add implementation summary - Add 7 vision-related guides Files: - New: src/extension/vision/imageAnalysisService.ts (180 lines) - Modified: 5 core files for attachment support - New: test/imageAnalysisService.test.ts (280 lines, 11 tests) - New: 9 documentation files in docs/superpowers/ Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> | 1 个月前 |
| feat: add command recovery and reference discovery | 1 个月前 |
| feat: add model runner and enhance conversation persistence - Introduce ModelRunner for unified model client management - Enhance persistent conversation store with structured metadata - Expand test coverage for agent runner and conversation flows - Update workspace intelligence and model client implementations - Add conversation history stop plan and design docs Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> | 1 个月前 |
| chore: initial project import | 2 个月前 |
| @ feat: dynamic workflow recovery, cycle management & context engineering - add cycleManager: cycle exit strategies and skipped-node handling - add workflowRecovery: reasoned recovery and failure visibility - add codeReviewTool: integrated code review workflow tool - add layeredSystemPrompt: context-layered system prompt builder - add messageCompression: message history compression - add spoolManager: explore-code spool management - add workspaceMutationLock: workspace mutation serialization - add javaAdapter / javaAstExtractor: Java language intelligence - add edgeRanking: graph edge ranking for code intelligence - remove toolDispatcher (migrated to dynamicWorkflowTools) - update dynamicGraphEngine/Types: state-driven cycle and expression eval - update workflowCompiler: compiled graph failure visibility - update workflowOrchestrator: recovery governance and data flow - update conversationManager, chatTypes, messages: workflow state types - add scripts: e2e probes, CDP test drivers, Java logistics tests - add docs/plans: 2026-07-31 workflow review, recovery, expression fixes - add test coverage: cycle, recovery, expression, failure visibility Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> @ | 1 个月前 |