diff --git a/.blueprint/manifest.yaml b/.blueprint/manifest.yaml index 68de532..efe4d20 100644 --- a/.blueprint/manifest.yaml +++ b/.blueprint/manifest.yaml @@ -372,8 +372,9 @@ modules: area: backend status: in_progress progress: 99 - note: "体系分发主线：T1/T2/B1/B2/B3/V1 ✅（T2 2026-07-04：git-workflow 复杂度感知分支策略；B3 2026-07-04：manifest agent_button_carrier + splitter 三级载体继承）；剩 B4 健康诊断遵守度（后置）" + note: "体系分发主线：T1/T2/B1/B2/B3/V1/U3 ✅（U3 2026-07-15：Rules-Applied ingest 层校验+回溯清洗）；剩 📋 U1（全托管升级通道+批量重分发）/ B4（遵守度+回传覆盖率，吸收被否决的 U2）" position: [3500, 0] + updated: 2026-07-15 - id: gitea-integration name: Gitea 集成 diff --git a/.blueprint/modules/onboarding-pack.md b/.blueprint/modules/onboarding-pack.md index 40e58f2..e84ce96 100644 --- a/.blueprint/modules/onboarding-pack.md +++ b/.blueprint/modules/onboarding-pack.md @@ -30,6 +30,8 @@ - 「完整分发」是陷阱，正确形态是分级菜单：新项目种全套，旧项目增量接入（复用 mcp_inject 的「命名空间隔离 + 合并不覆盖」语义，不动用户已有配置） - 接入动作：现有 `apply_blueprint_buff` 升格为 `apply_onboarding_pack`，按档编排注入 - 语义层进料口隐患：决策/验收/踩坑仍需主动写（与饿死复盘笔记同源），须配「从 PR 标题/commit body/review 评论半自动提取」，不再单纯指望人自觉 +- **分发文件三档升级语义（2026-07-15 择升环评估）**：① 版本化块文件（AGENTS/git-workflow）→ MANAGED 块原位升级；② 炼境全托管文件（envoy.md/blueprint-gate.cjs，用户不定制）→ 内容不一致即覆盖；③ 用户可定制文件（lefthook.yml）→ 绝不覆盖，但 verify_onboarding 须能检测落后。此前 ②③ 混用「存在即跳过」导致回流修正推不到存量项目 +- **回流触发不自动化（2026-07-15 多角度辩证）**：否决子项目侧 hook 自动触发 /envoy——自动化触发拆掉人肉过滤层，诱导低价值回传污染通道，伤「择」换「环」转速。回流频率问题用炼境侧可观测性解（「沉默项目」诊断信号，见 B4 卡），人留在过滤环里。校验类需求优先放统计消费端（炼境 ingest）而非分发 hook（R12 边界单点 + 零分发成本 + 可回溯） ## 复用的现有能力（~60% 地基已具备） @@ -132,6 +134,29 @@ - 版本化闭环要求（2026-07-03 用户澄清，体系分发的灵魂）: 分发不是"发了就不管"——每类资产带版本分发 → 子项目按该版本实践产生经验（成功/失败）→ 经验自吸反哺炼境（trailer/failure_cases/复盘/返工率）→ 按版本分组对比效果 → 修订出新版 → 再分发。CONVENTIONS 是模范生（version+队列对比+梦核+同步，1.8→1.9 已跑通一整圈）；B 系列每类资产按此标准建闭环。**前置依赖提升**：「接入包模板演进分发（MANAGED 块化）」concept 卡——模板无版本机制是当前最大缺口，没有它体系 v2 到不了存量项目 - notes: 2026-07-03 概念探讨收束后的分发需求（体系详见 memory concept_dual_projection 与 mcp-server 模块 concept 卡）。用户全 agent 开发形态下，"给开发者立规范"="给 agent 注入规范"，可完全机器化。四条现成管道分发四类资产：① B1(S) CONVENTIONS 1.10.0 加 R12 边界单点实现 + R13 验收须真实走通验证路径 + 载体选择原则；② B2(S) 接入包 AGENTS.md 模板加 agent 按钮章节（三要素/载体谱系/侦察驻军）；③ B3(M) init/sync 提示词升级——建蓝图时声明项目按钮载体、acceptance 生成为可执行验证路径格式；④ B4(S,后置) 健康诊断加体系遵守度检查。移植体验=零新增动作，接入流程不变。 +### 📋 U1. 全托管文件升级通道 + 批量重分发（择升环断点修复 ①） +- status: planned +- complexity: M +- depends: onboarding-pack（卡 T1） +- files: src-tauri/src/commands/onboarding.rs, src-tauri/src/mcp_tools.rs, src-tauri/resources/onboarding/.claude/commands/envoy.md, src-tauri/resources/onboarding/.claude/hooks/blueprint-gate.cjs +- acceptance: [Tauri command] 对存量项目重跑 apply_onboarding_pack → envoy.md / blueprint-gate.cjs 与内嵌模板不一致时覆盖（action=updated），一致时 skipped，文件头含「由炼境托管，重跑接入包会覆盖」声明；[cargo test] dispatch_fully_managed 三态（不存在→written / 一致→skipped / 不一致→updated）；[MCP tool] verify_onboarding 的 lefthook 检查项从内嵌 TMPL_LEFTHOOK 自省提取 commands 键集合，项目 lefthook.yml 缺键 → fail + 列出缺失键（禁止硬编码具体检查项名，模板演进校验器零维护）；[MCP tool] apply_onboarding_pack 不传 project_id → 对所有已登记项目批量执行并返回逐项目摘要（复用 ingest_git_history 先例），验收通过后删除 batch_apply.py / scan_onboarded.py +- notes: dispatch_one「存在即跳过」使回流修正只能推到 AGENTS/git-workflow 两份版本化文档——envoy/gate 的 bug 修复永远到不了存量 12 项目，批量重分发靠临时 python 脚本硬扛（batch_apply.py 为实锤）。三档升级语义见决策记录。已知盲点（辩证承认）：首次覆盖时子项目文件尚无托管声明，存在一次性盲覆盖窗口——envoy/gate 分发仅一周、被定制概率极低，且 updated 会在报告显式出现，风险接受 + +### ✅ U3. Rules-Applied 真实性校验（ingest 层单点实现，择升环断点修复 ②） +- status: done +- done_note: 2026-07-15 完成。纯函数 extract_rule_ids（R+数字 token 边界识别，防 PR13/R2D2 误提取）+ filter_valid_rules（valid=None 不校验 fail-open）；校验接入全部 4 条写入链路（buff 增量 / webhook push / gitea 轮询 / 全量导入），规则集每项目每批次加载一次；sanitize_rules_applied 显式回溯清洗通道（UPSERT「不覆盖」约束原样保留，两者并行不悖）；IngestResult 新增 sanitized 字段（Rust/TS/MCP 输出/FlywheelPanel 四层同步）；record_commit 收窄为 #[cfg(test)]，生产链路只剩 validated 版防绕过。cargo test 98 全过（新增 5 测：token 边界/过滤/入库剔除/message 链路/清洗幂等+仅清洗指定项目），pnpm typecheck 干净 +- complexity: S +- files: src-tauri/src/commands/commit_metrics.rs +- acceptance: [cargo test] trailer 解析校验纯函数——规则 ID 不存在于该项目 .blueprint/CONVENTIONS.md（如 R99）→ 从 rules_applied 剔除并可观测（日志或 invalid 计数），真实 ID 保留；项目无 CONVENTIONS.md → 不校验全保留；[MCP tool] ingest_git_history 重跑 → 「规则命中 × 返工」交叉统计不再含未知 ID（存量污染可回溯清洗） +- notes: 辩证翻转载体：原方案（子项目 commit-msg hook）违反 R12 边界单点——统计消费端在炼境，校验就该在炼境 ingest 层。收益：零分发成本覆盖全部项目、重跑即回溯清洗、cargo test 可测（炼境自身 agent 按钮载体）。commit-msg hook 即时反馈版降为可选增强（若做依赖 U1 通道，不在本卡）。上限诚实：只挡 typo/幻觉 ID，「真 ID 假触发」留给宁缺毋滥纪律 + +### 📋 B4. 健康诊断遵守度 + 回传覆盖率（吸收被否决的 U2，择升环断点修复 ③） +- status: planned +- complexity: S +- files: src-tauri/src/mcp_tools.rs, src-tauri/src/commands/onboarding.rs +- acceptance: [MCP tool] get_diagnostics（或 verify_onboarding 检查链）新增「沉默项目」信号：项目近 30 天 commit_metrics 有数据但使者回传（list_envoy_returns 同源数据）为 0 → 列出项目名 + commit 数，提示去该项目跑 /envoy；近 30 天有回传或无 commit 活动 → 不报 +- notes: U2（子项目侧 Stop hook 自动触发 /envoy）经多角度辩证否决（2026-07-15）：自动化触发会拆掉人肉过滤层，诱导 agent 用 under_validation 噪声交差灌满回传通道——伤「择」换「环」转速，本末倒置。正解是炼境侧可观测性：让遗忘可被看见（信号驱动），人留在过滤环里决定是否回传。同时销掉原 B4 后置欠账（体系遵守度检查） + ## 复盘笔记 【复盘】V1 接入生效验收（2026-07-04） diff --git a/src-tauri/src/commands/buff.rs b/src-tauri/src/commands/buff.rs index cd998d0..18d89e8 100644 --- a/src-tauri/src/commands/buff.rs +++ b/src-tauri/src/commands/buff.rs @@ -159,9 +159,16 @@ fn check_project(app: &AppHandle, project_path: &str) -> Result<(), String> { ); } // 批量写入用事务包裹（首次可达数十条，避免逐条 fsync） + // U3：规则 ID 集合每批次加载一次，剔除 trailer 里的未知 ID + let valid_rules = crate::commands::commit_metrics::load_project_rule_ids(root); let tx = conn.unchecked_transaction(); for entry in &commits { - let _ = crate::commands::commit_metrics::record_commit_entry(&conn, &project_id, entry); + let _ = crate::commands::commit_metrics::record_commit_entry( + &conn, + &project_id, + entry, + valid_rules.as_ref(), + ); } if let Ok(tx) = tx { let _ = tx.commit(); diff --git a/src-tauri/src/commands/commit_metrics.rs b/src-tauri/src/commands/commit_metrics.rs index 46f37c7..6aef022 100644 --- a/src-tauri/src/commands/commit_metrics.rs +++ b/src-tauri/src/commands/commit_metrics.rs @@ -5,6 +5,7 @@ use rusqlite::Connection; use serde::{Deserialize, Serialize}; +use std::collections::HashSet; /// 一条 commit 的解析结果。 #[derive(Debug, Clone, PartialEq, Eq)] @@ -107,6 +108,104 @@ pub fn parse_rules_applied(message: &str) -> Option { None } +// ── U3：Rules-Applied 真实性校验（ingest 层单点，防幻觉/typo ID 污染交叉统计）── + +/// 从 CONVENTIONS.md 内容提取全部规则 ID（大写 R+数字，前后不粘字母数字）。纯函数。 +pub fn extract_rule_ids(text: &str) -> HashSet { + let b = text.as_bytes(); + let mut ids = HashSet::new(); + let mut i = 0; + while i < b.len() { + if b[i] == b'R' && (i == 0 || !b[i - 1].is_ascii_alphanumeric()) { + let mut j = i + 1; + while j < b.len() && b[j].is_ascii_digit() { + j += 1; + } + if j > i + 1 && (j == b.len() || !b[j].is_ascii_alphanumeric()) { + ids.insert(text[i..j].to_string()); + } + i = j; + } else { + i += 1; + } + } + ids +} + +/// 过滤 rules_applied 串：剔除不在 valid 集合中的 ID；全部无效返回 None。 +/// valid 为 None（项目无 CONVENTIONS.md）时不校验、原样保留。纯函数。 +pub fn filter_valid_rules(rules: &str, valid: Option<&HashSet>) -> Option { + let Some(set) = valid else { + return Some(rules.to_string()); + }; + let kept: Vec<&str> = rules + .split(',') + .map(str::trim) + .filter(|t| !t.is_empty() && set.contains(*t)) + .collect(); + if kept.is_empty() { + None + } else { + Some(kept.join(",")) + } +} + +/// 读取项目 .blueprint/CONVENTIONS.md 提取规则 ID 集合；文件不存在返回 None（不校验）。 +/// 每项目每批次调用一次，勿放进逐 commit 循环。 +pub fn load_project_rule_ids(root: &std::path::Path) -> Option> { + let content = std::fs::read_to_string(root.join(".blueprint").join("CONVENTIONS.md")).ok()?; + Some(extract_rule_ids(&content)) +} + +/// 远程采集链路（webhook/轮询）无本地 root，按 project_id 查 blueprint_buffs 路径再加载。 +pub fn load_rule_ids_by_project_id(conn: &Connection, project_id: &str) -> Option> { + let path: String = conn + .query_row( + "SELECT project_path FROM blueprint_buffs WHERE project_id = ?1", + rusqlite::params![project_id], + |r| r.get(0), + ) + .ok()?; + load_project_rule_ids(std::path::Path::new(&path)) +} + +/// 回溯清洗存量行：剔除 rules_applied 中不在 valid 集合的 ID（全无效置 NULL）。 +/// 与 UPSERT 的「不覆盖已有值」约束并行不悖——本函数是显式清洗通道，仅在全量导入时调用。 +/// 返回被清洗的行数，重跑幂等（已干净的行不再计数）。 +pub fn sanitize_rules_applied( + conn: &Connection, + project_id: &str, + valid: &HashSet, +) -> Result { + let rows: Vec<(String, String)> = { + let mut stmt = conn + .prepare( + "SELECT sha, rules_applied FROM commit_metrics + WHERE project_id = ?1 AND rules_applied IS NOT NULL", + ) + .map_err(|e| e.to_string())?; + let mapped = stmt + .query_map(rusqlite::params![project_id], |r| { + Ok((r.get::<_, String>(0)?, r.get::<_, String>(1)?)) + }) + .map_err(|e| e.to_string())?; + mapped.filter_map(|r| r.ok()).collect() + }; + let mut cleaned = 0u32; + for (sha, rules) in rows { + let filtered = filter_valid_rules(&rules, Some(valid)); + if filtered.as_deref() != Some(rules.as_str()) { + conn.execute( + "UPDATE commit_metrics SET rules_applied = ?1 WHERE sha = ?2", + rusqlite::params![filtered, sha], + ) + .map_err(|e| e.to_string())?; + cleaned += 1; + } + } + Ok(cleaned) +} + /// UPSERT 写入一条 commit 记录：新 sha 插入；已存在且 rules_applied 为空时仅回填 /// rules_applied（绝不覆盖已有值——回填幂等的关键约束）。返回是否产生了写入。 fn upsert_metrics( @@ -140,9 +239,27 @@ fn upsert_metrics( Ok(n > 0) } -/// 解析并写入一条 commit 记录（从完整 message 提取 conventional + Rules-Applied trailer）。 -/// 生产链路的 git log 只给 subject，trailer 走 `record_commit_entry`；本函数保留给 -/// 持有完整 message 的调用方与测试。 +/// 解析并写入一条 commit 记录（从完整 message 提取 conventional + Rules-Applied trailer）， +/// 并按 valid_rules 剔除未知规则 ID（U3）。远程采集生产链路（webhook/轮询）用本函数， +/// valid_rules 每项目加载一次（`load_rule_ids_by_project_id`）。 +pub fn record_commit_validated( + conn: &Connection, + project_id: &str, + sha: &str, + message: &str, + committed_at: Option<&str>, + valid_rules: Option<&HashSet>, +) -> Result<(), String> { + let p = parse_conventional(message); + let rules_applied = + parse_rules_applied(message).and_then(|r| filter_valid_rules(&r, valid_rules)); + upsert_metrics(conn, project_id, sha, &p, rules_applied.as_deref(), committed_at)?; + Ok(()) +} + +/// 不校验规则 ID 的写入（等价 valid_rules=None）。仅供测试铺数据； +/// 生产链路一律走 `record_commit_validated` / `record_commit_entry`（U3 校验强制）。 +#[cfg(test)] pub fn record_commit( conn: &Connection, project_id: &str, @@ -150,21 +267,21 @@ pub fn record_commit( message: &str, committed_at: Option<&str>, ) -> Result<(), String> { - let p = parse_conventional(message); - let rules_applied = parse_rules_applied(message); - upsert_metrics(conn, project_id, sha, &p, rules_applied.as_deref(), committed_at)?; - Ok(()) + record_commit_validated(conn, project_id, sha, message, committed_at, None) } /// 写入一条来自 `read_git_log` 的记录（subject 与 trailer 分列，trailer 由 git 原生解析）。 /// 同时把触碰文件写入 commit_files（INSERT OR IGNORE 幂等，供文件级风险画像）。 +/// valid_rules 每项目加载一次传入（U3 校验）；None 表示项目无 CONVENTIONS.md，不校验。 pub fn record_commit_entry( conn: &Connection, project_id: &str, entry: &GitLogEntry, + valid_rules: Option<&HashSet>, ) -> Result { let p = parse_conventional(&entry.subject); - let rules = normalize_rule_tokens(&entry.rules_trailer); + let rules = normalize_rule_tokens(&entry.rules_trailer) + .and_then(|r| filter_valid_rules(&r, valid_rules)); let date = if entry.committed_at.is_empty() { None } else { @@ -300,6 +417,8 @@ pub struct IngestResult { pub imported: u32, /// 已存在且无需回填（跳过） pub skipped: u32, + /// 存量行中被清洗掉未知规则 ID 的行数（U3 回溯清洗；项目无 CONVENTIONS.md 时恒为 0） + pub sanitized: u32, } /// git log 单条记录。`rules_trailer` 是 git 原生解析出的 Rules-Applied trailer 值 @@ -406,16 +525,23 @@ pub fn ingest_full_git_history( ) .unwrap_or(project_id); + // U3：每项目加载一次规则 ID 集合，新写入过滤 + 存量回溯清洗 + let valid_rules = load_project_rule_ids(root); + let tx = conn.unchecked_transaction().map_err(|e| e.to_string())?; let mut imported = 0u32; for entry in &commits { // 新 sha 插入；已存在的行 UPSERT 回填 rules_applied（trailer 死链修复后的存量补数路径） - if record_commit_entry(&conn, &effective_id, entry)? { + if record_commit_entry(&conn, &effective_id, entry, valid_rules.as_ref())? { imported += 1; } } + let sanitized = match &valid_rules { + Some(set) => sanitize_rules_applied(&conn, &effective_id, set)?, + None => 0, + }; tx.commit().map_err(|e| e.to_string())?; - Ok(IngestResult { imported, skipped: total - imported }) + Ok(IngestResult { imported, skipped: total - imported, sanitized }) } // ── 跨项目 Git 健康度聚合 ───────────────────────────────────────────────────── @@ -784,22 +910,22 @@ mod tests { assert_eq!(rules_of(&conn, "old"), None); // 回填：同 sha 带 trailer 重写 → 补上 - let changed = record_commit_entry(&conn, "proj", &entry("old", "feat(a): x", "R01,R06")).unwrap(); + let changed = record_commit_entry(&conn, "proj", &entry("old", "feat(a): x", "R01,R06"), None).unwrap(); assert!(changed, "回填应计为写入"); assert_eq!(rules_of(&conn, "old").as_deref(), Some("R01,R06")); // 不覆盖：已有值的行，不同 trailer 重写 → 保持原值 - let changed = record_commit_entry(&conn, "proj", &entry("old", "feat(a): x", "R99")).unwrap(); + let changed = record_commit_entry(&conn, "proj", &entry("old", "feat(a): x", "R99"), None).unwrap(); assert!(!changed, "已有值不应被覆盖"); assert_eq!(rules_of(&conn, "old").as_deref(), Some("R01,R06")); // 无 trailer 重写已有 NULL 行 → 不产生无意义 UPDATE record_commit(&conn, "proj", "old2", "chore: y", None).unwrap(); - let changed = record_commit_entry(&conn, "proj", &entry("old2", "chore: y", "")).unwrap(); + let changed = record_commit_entry(&conn, "proj", &entry("old2", "chore: y", ""), None).unwrap(); assert!(!changed); // 新行带 trailer 直接写入 - assert!(record_commit_entry(&conn, "proj", &entry("new", "fix(b): z", "R05")).unwrap()); + assert!(record_commit_entry(&conn, "proj", &entry("new", "fix(b): z", "R05"), None).unwrap()); assert_eq!(rules_of(&conn, "new").as_deref(), Some("R05")); } @@ -844,8 +970,8 @@ mod tests { let conn = crate::db::conn_with_schema(); let mut e = entry("s1", "feat(a): x", ""); e.files = vec!["src/a.rs".into(), "src/b.rs".into()]; - record_commit_entry(&conn, "proj", &e).unwrap(); - record_commit_entry(&conn, "proj", &e).unwrap(); // 重跑 + record_commit_entry(&conn, "proj", &e, None).unwrap(); + record_commit_entry(&conn, "proj", &e, None).unwrap(); // 重跑 let count: i64 = conn .query_row("SELECT COUNT(*) FROM commit_files WHERE sha = 's1'", [], |r| r.get(0)) @@ -990,5 +1116,80 @@ Rules-Applied: R06", None).unwrap(); record_commit(&conn, "pz", "s9", "feat: plain", None).unwrap(); assert!(rule_hit_stats(&conn).unwrap().is_empty()); } + + // ── U3：Rules-Applied 真实性校验 ──────────────────────────────────────── + + fn valid_set(ids: &[&str]) -> HashSet { + ids.iter().map(|s| s.to_string()).collect() + } + + #[test] + fn extract_rule_ids_respects_token_boundaries() { + let text = "### R01 规则一\n- R12：xxx（见 |R06| 表格）\nPR13 前粘字母不算，R2D2 后粘字母不算，r03 小写不算"; + let ids = extract_rule_ids(text); + assert!(ids.contains("R01")); + assert!(ids.contains("R12")); + assert!(ids.contains("R06")); + assert!(!ids.contains("R13"), "PR13 内的 R13 不是独立 ID"); + assert!(!ids.contains("R2"), "R2D2 内的 R2 不是独立 ID"); + assert!(!ids.contains("R03") && !ids.contains("r03"), "小写 r 不提取"); + assert!(extract_rule_ids("没有任何规则").is_empty()); + } + + #[test] + fn filter_valid_rules_drops_unknown_ids() { + let set = valid_set(&["R01", "R06"]); + assert_eq!(filter_valid_rules("R01,R99", Some(&set)).as_deref(), Some("R01")); + assert_eq!(filter_valid_rules("R99", Some(&set)), None, "全无效返回 None"); + assert_eq!( + filter_valid_rules("R01,R99", None).as_deref(), + Some("R01,R99"), + "无 CONVENTIONS.md 时不校验" + ); + } + + #[test] + fn record_commit_entry_filters_unknown_ids_at_ingest() { + let conn = crate::db::conn_with_schema(); + let set = valid_set(&["R01"]); + record_commit_entry(&conn, "proj", &entry("v1", "feat(a): x", "R01,R99"), Some(&set)).unwrap(); + assert_eq!(rules_of(&conn, "v1").as_deref(), Some("R01"), "未知 ID 应在入库前剔除"); + record_commit_entry(&conn, "proj", &entry("v2", "feat(a): y", "R99"), Some(&set)).unwrap(); + assert_eq!(rules_of(&conn, "v2"), None, "全无效则该列为 NULL"); + } + + #[test] + fn record_commit_validated_filters_trailer_in_message() { + let conn = crate::db::conn_with_schema(); + let set = valid_set(&["R05"]); + record_commit_validated(&conn, "proj", "w1", "feat(x): a\n\nRules-Applied: R05,R99", None, Some(&set)).unwrap(); + assert_eq!(rules_of(&conn, "w1").as_deref(), Some("R05")); + } + + #[test] + fn sanitize_cleans_stored_pollution_idempotently() { + let conn = crate::db::conn_with_schema(); + // 存量污染：校验上线前写入的行 + record_commit_entry(&conn, "proj", &entry("p1", "feat: a", "R01,R99"), None).unwrap(); + record_commit_entry(&conn, "proj", &entry("p2", "fix: b", "R99"), None).unwrap(); + record_commit_entry(&conn, "proj", &entry("p3", "feat: c", "R01"), None).unwrap(); + record_commit_entry(&conn, "other", &entry("q1", "feat: d", "R99"), None).unwrap(); + + let set = valid_set(&["R01"]); + let cleaned = sanitize_rules_applied(&conn, "proj", &set).unwrap(); + assert_eq!(cleaned, 2, "p1 剔除 R99、p2 置 NULL；p3 干净不动"); + assert_eq!(rules_of(&conn, "p1").as_deref(), Some("R01")); + assert_eq!(rules_of(&conn, "p2"), None); + assert_eq!(rules_of(&conn, "p3").as_deref(), Some("R01")); + assert_eq!(rules_of(&conn, "q1").as_deref(), Some("R99"), "只清洗指定项目"); + + // 清洗后交叉统计不再含未知 ID（acceptance 断言） + let stats = rule_hit_stats(&conn).unwrap(); + assert!(stats.iter().filter(|s| s.rule_id == "R99").all(|s| s.project_count == 1), + "proj 的 R99 已清，仅剩 other 项目的"); + + // 幂等：重跑零清洗 + assert_eq!(sanitize_rules_applied(&conn, "proj", &set).unwrap(), 0); + } } diff --git a/src-tauri/src/commands/gitea.rs b/src-tauri/src/commands/gitea.rs index 7036610..2e8fb70 100644 --- a/src-tauri/src/commands/gitea.rs +++ b/src-tauri/src/commands/gitea.rs @@ -517,13 +517,16 @@ pub fn poll_all_repos() -> Result<(usize, usize), String> { Err(_) => continue, }; polled += 1; + // U3：按 project_id 加载规则 ID 集合（每仓库一次），剔除 trailer 里的未知 ID + let valid_rules = + crate::commands::commit_metrics::load_rule_ids_by_project_id(&conn, project_id); for (sha, message, date) in parse_gitea_commits(&json) { let before: i64 = conn.query_row( "SELECT COUNT(*) FROM commit_metrics WHERE sha = ?1", params![sha], |r| r.get(0), ).unwrap_or(0); if before == 0 - && crate::commands::commit_metrics::record_commit( - &conn, project_id, &sha, &message, date.as_deref(), + && crate::commands::commit_metrics::record_commit_validated( + &conn, project_id, &sha, &message, date.as_deref(), valid_rules.as_ref(), ).is_ok() { ingested += 1; diff --git a/src-tauri/src/commands/risk.rs b/src-tauri/src/commands/risk.rs index 33b88d5..b0a67c5 100644 --- a/src-tauri/src/commands/risk.rs +++ b/src-tauri/src/commands/risk.rs @@ -263,7 +263,7 @@ mod tests { rules_trailer: String::new(), files: files.iter().map(|s| s.to_string()).collect(), }; - record_commit_entry(conn, pid, &entry).unwrap(); + record_commit_entry(conn, pid, &entry, None).unwrap(); } #[test] diff --git a/src-tauri/src/mcp_tools.rs b/src-tauri/src/mcp_tools.rs index 6530389..c8a01bd 100644 --- a/src-tauri/src/mcp_tools.rs +++ b/src-tauri/src/mcp_tools.rs @@ -566,7 +566,10 @@ pub async fn tools_call(params: Option<&Value>) -> Value { ) { Ok(r) => { ok += 1; - lines.push(format!("- {}: 写入 {} 条，跳过 {} 条", id, r.imported, r.skipped)); + lines.push(format!( + "- {}: 写入 {} 条，跳过 {} 条，清洗 {} 条无效规则ID", + id, r.imported, r.skipped, r.sanitized + )); } Err(e) => { failed += 1; diff --git a/src-tauri/src/mcp_webhook.rs b/src-tauri/src/mcp_webhook.rs index de40985..116c70d 100644 --- a/src-tauri/src/mcp_webhook.rs +++ b/src-tauri/src/mcp_webhook.rs @@ -133,14 +133,18 @@ async fn handle_push(project_id: &str, body: &Bytes) -> Response { .into_response() } }; + // U3：按 project_id 找到本地路径加载规则 ID 集合，剔除 trailer 里的未知 ID + let valid_rules = + crate::commands::commit_metrics::load_rule_ids_by_project_id(&conn, project_id); let mut received = 0; for c in &commits { - if crate::commands::commit_metrics::record_commit( + if crate::commands::commit_metrics::record_commit_validated( &conn, project_id, &c.sha, &c.message, c.timestamp.as_deref(), + valid_rules.as_ref(), ) .is_ok() { diff --git a/src/components/insight/FlywheelPanel.tsx b/src/components/insight/FlywheelPanel.tsx index c50ddf7..f35cc65 100644 --- a/src/components/insight/FlywheelPanel.tsx +++ b/src/components/insight/FlywheelPanel.tsx @@ -112,17 +112,21 @@ export function FlywheelPanel() { setIngestSummary(null); let totalImported = 0; let totalSkipped = 0; + let totalSanitized = 0; for (const { id, path } of projectsWithPath) { try { const r = await ingestFullGitHistory(id, path); totalImported += r.imported; totalSkipped += r.skipped; + totalSanitized += r.sanitized; } catch { // 单个项目失败不中断其他项目 } } await refetchHealth(); - setIngestSummary(`新导入 ${totalImported} 条，跳过 ${totalSkipped} 条`); + setIngestSummary( + `新导入 ${totalImported} 条，跳过 ${totalSkipped} 条${totalSanitized > 0 ? `，清洗 ${totalSanitized} 条无效规则ID` : ""}`, + ); setIngesting(false); }; diff --git a/src/lib/commands.ts b/src/lib/commands.ts index f1707e2..c1d48c1 100644 --- a/src/lib/commands.ts +++ b/src/lib/commands.ts @@ -1580,6 +1580,8 @@ export const getCommitStats = (projectId: string) => export interface IngestResult { imported: number; skipped: number; + /** 存量行中被清洗掉未知规则 ID 的行数（U3；项目无 CONVENTIONS.md 时为 0） */ + sanitized: number; } export const ingestFullGitHistory = (projectId: string, projectPath: string) =>