)
0.7.5 (2026-05-04)【免费下载链接】graphifyTurn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.项目地址: https://gitcode.com/GitHub_Trending/graph/graphifyFeat:graphify extractnow runs incrementally — auto-detects priormanifest.jsonand re-extracts only changed/new files; semantic results cached by content hash so unchanged docs cost zero LLM tokens on repeat runs (#698)Feat: Entity deduplication pipeline runs on every build — entropy gate MinHash/LSH blocking Jaro-Winkler verification same-community boost collapses near-duplicate entities (typos, spacing, plurals) before clusteringFeat:--dedup-llmflag forgraphify extract— optional LLM tiebreaker for ambiguous entity pairs (~$0.01 for 10k-node graphs), off by defaultDeps:datasketchandrapidfuzzadded as base dependencies需要说明的适用前提该版本目标是计划时点的 0.7.5当前仓库版本已推进到 **0.9.52**[pyproject.toml](https://link.gitcode.com/i/87b9e7ee0f08732182792810dd6d731f) 第 7 行上述两项能力早已进入发布序列[CHANGELOG.md](https://link.gitcode.com/i/9b8397fe90cae71b85aec66ad3c9ffde) 中保留了大量围绕它们的后续修正记录可作为计划 → 生产演进的实证索引例如--no-dedup 开关允许运营者保留模糊命中的不同符号#2881、模糊合并计数在纯模糊运行下也被报告#1857/#1860、语义缓存写入按批次白名单限定#1757/#1835、--mode deep 独立缓存命名空间与 --force 绕过#1894、manifest 记录全部成功抽取文件而非仅零节点文件#1897、build_merge 无 root 时从 .graphify_root 标记回退推断以修复幽灵节点#1571等。 ## 八、Self-Review规格覆盖与类型一致性 计划结尾的自检Self-Review确认了规格点与任务的映射关系值得作为计划文档写作范式的参照 **Spec coverage全部勾选** - 语义缓存包裹 → Task 4 步骤 3–4增量自动检测manifest→ Task 4 步骤 3build_merge prune_sources → Task 4 步骤 5manifest 仅在成功时保存 → Task 4 步骤 6摘要输出 → Task 4 步骤 8 - dedup.py 新模块 → Task 2熵门控 / MinHash-LSH 阻断 / Jaro-Winkler 验证 / 同社区加成 / 并查集合并 → Task 2 步骤 3 - --dedup-llm 裁决器 → Task 5build() 与 build_merge() 接入 → Task 3 - datasketch rapidfuzz 依赖 → Task 1去重全链路测试 → Task 2 3增量测试 → Task 4 - CHANGELOG 版本号 → Task 6 **Placeholder scan** 无占位符。 **Type consistency签名一致性** deduplicate_entities(nodes, edges, *, communities, dedup_llm_backendNone) 在 Task 2/3/5 中一致build(extractions, *, directed, dedup) 在 Task 3 中一致build_merge(..., dedupTrue) 在 Task 3/4 中一致。当前实现的最终签名在此基础上追加了 root 与 hyperedges 两个关键字参数见 [graphify/dedup.py](https://link.gitcode.com/i/7bdc77c7c145362afb866339de30b1a0)。 ## 九、测试与回归验证 计划在每个 Task 内都固定了相同的验证节奏可整体复用于后续类似改动 bash # 目标测试 python -m pytest tests/test_dedup.py -v --tbshort python -m pytest tests/test_incremental.py -v --tbshort # 全量回归计划时点基线532 passed, 5 failed —— 预存 SQL 用例失败 python -m pytest tests/ -q --tbno【免费下载链接】graphifyTurn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.项目地址: https://gitcode.com/GitHub_Trending/graph/graphify创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考 0.7.5 (2026-05-04)
0.7.5 (2026-05-04)【免费下载链接】graphifyTurn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.项目地址: https://gitcode.com/GitHub_Trending/graph/graphifyFeat:graphify extractnow runs incrementally — auto-detects priormanifest.jsonand re-extracts only changed/new files; semantic results cached by content hash so unchanged docs cost zero LLM tokens on repeat runs (#698)Feat: Entity deduplication pipeline runs on every build — entropy gate MinHash/LSH blocking Jaro-Winkler verification same-community boost collapses near-duplicate entities (typos, spacing, plurals) before clusteringFeat:--dedup-llmflag forgraphify extract— optional LLM tiebreaker for ambiguous entity pairs (~$0.01 for 10k-node graphs), off by defaultDeps:datasketchandrapidfuzzadded as base dependencies需要说明的适用前提该版本目标是计划时点的 0.7.5当前仓库版本已推进到 **0.9.52**[pyproject.toml](https://link.gitcode.com/i/87b9e7ee0f08732182792810dd6d731f) 第 7 行上述两项能力早已进入发布序列[CHANGELOG.md](https://link.gitcode.com/i/9b8397fe90cae71b85aec66ad3c9ffde) 中保留了大量围绕它们的后续修正记录可作为计划 → 生产演进的实证索引例如--no-dedup 开关允许运营者保留模糊命中的不同符号#2881、模糊合并计数在纯模糊运行下也被报告#1857/#1860、语义缓存写入按批次白名单限定#1757/#1835、--mode deep 独立缓存命名空间与 --force 绕过#1894、manifest 记录全部成功抽取文件而非仅零节点文件#1897、build_merge 无 root 时从 .graphify_root 标记回退推断以修复幽灵节点#1571等。 ## 八、Self-Review规格覆盖与类型一致性 计划结尾的自检Self-Review确认了规格点与任务的映射关系值得作为计划文档写作范式的参照 **Spec coverage全部勾选** - 语义缓存包裹 → Task 4 步骤 3–4增量自动检测manifest→ Task 4 步骤 3build_merge prune_sources → Task 4 步骤 5manifest 仅在成功时保存 → Task 4 步骤 6摘要输出 → Task 4 步骤 8 - dedup.py 新模块 → Task 2熵门控 / MinHash-LSH 阻断 / Jaro-Winkler 验证 / 同社区加成 / 并查集合并 → Task 2 步骤 3 - --dedup-llm 裁决器 → Task 5build() 与 build_merge() 接入 → Task 3 - datasketch rapidfuzz 依赖 → Task 1去重全链路测试 → Task 2 3增量测试 → Task 4 - CHANGELOG 版本号 → Task 6 **Placeholder scan** 无占位符。 **Type consistency签名一致性** deduplicate_entities(nodes, edges, *, communities, dedup_llm_backendNone) 在 Task 2/3/5 中一致build(extractions, *, directed, dedup) 在 Task 3 中一致build_merge(..., dedupTrue) 在 Task 3/4 中一致。当前实现的最终签名在此基础上追加了 root 与 hyperedges 两个关键字参数见 [graphify/dedup.py](https://link.gitcode.com/i/7bdc77c7c145362afb866339de30b1a0)。 ## 九、测试与回归验证 计划在每个 Task 内都固定了相同的验证节奏可整体复用于后续类似改动 bash # 目标测试 python -m pytest tests/test_dedup.py -v --tbshort python -m pytest tests/test_incremental.py -v --tbshort # 全量回归计划时点基线532 passed, 5 failed —— 预存 SQL 用例失败 python -m pytest tests/ -q --tbno【免费下载链接】graphifyTurn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.项目地址: https://gitcode.com/GitHub_Trending/graph/graphify创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考