codeblast 0.3.4 → 0.3.5

package/README.md

@@ -9,7 +9,7 @@
 </p>
 
 <p align="center">
-  <a href="#the-three-queries"><img src="https://img.shields.io/badge/TypeScript-function--level-3178c6?style=flat-square" alt="TypeScript function-level"/></a>
+  <a href="#the-four-views"><img src="https://img.shields.io/badge/TypeScript-function--level-3178c6?style=flat-square" alt="TypeScript function-level"/></a>
   <a href="#the-precision-promise-bounded-and-evidence-backed"><img src="https://img.shields.io/badge/recall-28%2F28_%3D_100%25-3fb950?style=flat-square" alt="mutation-tested recall 100%"/></a>
   <a href="SKILL.md"><img src="https://img.shields.io/badge/Agent-Skill-7c3aed?style=flat-square" alt="agent skill"/></a>
   <img src="https://img.shields.io/badge/license-MIT-8b949e?style=flat-square" alt="MIT"/>
@@ -31,7 +31,7 @@ jobs:
     steps:
       - uses: actions/checkout@v4
         with: { fetch-depth: 0 }
-      - uses: alloevil/codeblast@v0.3.3
+      - uses: alloevil/codeblast@v0.3.5
 ```
 
 On a structural PR, codeblast posts a bounded review decision, affected tests, `file:line` evidence,
@@ -40,14 +40,15 @@ locally in the runner; source is not uploaded to a codeblast service.
 
 ## What it is
 
-**codeblast parses your repository into a deterministic code graph and answers the three most expensive questions around any code change:**
+**codeblast parses your repository into a deterministic code graph and exposes four views of a change:**
 > 🔗 **[Live interactive demo](https://alloevil.github.io/codeblast/)** — real architecture maps of tRPC / Tabby / sgp, with three-level drill-down
 
 | | Question | Command |
 |---|---|---|
-| 🎯 | **What breaks if I change this?** | `impact` — direct / transitive / affected-tests, in three tiers |
-| 🔍 | **What did this PR structurally change?** | `change` — symbols and dependency edges added, removed, renamed |
-| 🗺️ | **What does this project look like?** | `archmap` — collapsible module map + circular-dependency detection |
+| 🎯 | **What breaks if I change this?** | `impact` — direct, transitive and affected-test reachability |
+| 🛡️ | **Should this change receive targeted review?** | `check-change` — risk, graph health, affected tests and warnings |
+| 🔍 | **What structurally changed?** | `change` — symbols and dependency edges added, removed or renamed |
+| 🗺️ | **What does this project look like?** | `archmap` — module/file/symbol drill-down and cycle detection |
 
 Built for humans (CLI / interactive HTML / PR comments) and for AI agents ([SKILL.md](SKILL.md)) — one graph, two front-ends.
 <table>
@@ -72,7 +73,7 @@ Built for humans (CLI / interactive HTML / PR comments) and for AI agents ([SKIL
 ```bash
 npx codeblast demo            # build a graph of the current repo, run one impact query, emit the map
 npm i -g codeblast            # or install globally; needs Node ≥ 22.13 (built-in sqlite) or Bun
-                              # npm serves 0.3.3
+                              # npm serves 0.3.5
 
 # Install as an agent skill (Claude Code, Codex, Cursor, and 14 more harnesses)
 npx skills add alloevil/codeblast
@@ -95,12 +96,12 @@ jobs:
     steps:
       - uses: actions/checkout@v4
         with: { fetch-depth: 0 }
-      - uses: alloevil/codeblast@v0.3.3
+      - uses: alloevil/codeblast@v0.3.5
 ```
 
-The action builds the analyzer from the ref you pinned, posts one sticky comment per PR, and stays
-silent when the diff has no structural change. For the full input/output contract, use the
-[workflow template](.github/workflows-template/codeblast.yml).
+The action builds the analyzer from the ref you pinned, posts one sticky comment when a PR contains a
+meaningful structural or tested behavior signal, and stays silent for docs-only, test-only, or
+auxiliary-only noise. For the full input/output contract, use the [workflow template](.github/workflows-template/codeblast.yml).
 
 
 ## Why not yet another LLM diagram tool
@@ -113,10 +114,10 @@ codeblast:     code → deterministic tsc/AST parse → graph → project  graph
 **Every node and every static-analysis edge carries `file:line` evidence** you can open and verify (`co_change` edges put a co-commit count in that field instead of a source line; file-level nodes carry line 1).
 The LLM does exactly one job in the pipeline: giving modules human-readable names — node membership and edges always come from static analysis.
 
-## The three queries
+## The four views
 
 <p align="center">
-  <img src="assets/readme/three-queries.svg" width="100%" alt="The three codeblast queries — impact: check the blast radius first; change: structural diff between two refs; archmap: module, file and symbol drill-down"/>
+  <img src="assets/readme/three-queries.svg" width="100%" alt="codeblast views — impact, change and architecture; check-change adds the machine-readable pre-merge safety decision"/>
 </p>
 
 ```bash
@@ -130,11 +131,15 @@ codeblast impact graph.db "createOrder" --json
 #    → two channels: call-graph reachable (precision ~0.70, read first)
 #      + import reachable (conservative supplement, don't skip)
 
-# ② Change Map — structural diff between two refs
+# ② Safety — one machine-readable pre-merge decision
+codeblast check-change <repo> <base> <head> --json
+#    → decision / risk / graph_health / affected_test_files / warnings
+
+# ③ Change Map — structural diff between two refs
 codeblast change <repo> main~5 main --json
 #    → unexpected edges_added = a signal the change is out of scope
 
-# ③ Architecture Map — interactive HTML: module → file → symbol drill-down,
+# ④ Architecture Map — interactive HTML: module → file → symbol drill-down,
 #    symbols link to source lines
 codeblast archmap graph.db --out arch.html --repo-url <github-url> \
   --site-url https://example.github.io/repo --og-image <share-image-url>
@@ -159,12 +164,11 @@ The self-pilot record is [`eval/pilot-2026-09-24.json`](eval/pilot-2026-09-24.js
 package indexed this repository with 0 extraction failures and returned separate review-first and test
 guidance. It is evidence that the workflow runs, not a claim of universal accuracy.
 
-Copy [`.github/workflows-template/codeblast.yml`](.github/workflows-template/codeblast.yml) into your repo (it runs `npx codeblast pr-comment`, no other setup):
-every PR gets an automatic comment with structural changes + blast radius + new symbols with no test coverage; **PRs with no structural change get zero comments**.
-Replayed against 50 real commits: 42 correctly stayed silent. Comment usefulness is the honest weak spot —
-four review rounds — rounds 1–3 independent blind review, round 4 by the current model — scored 25% / 75% / 57% / 20% useful, against 7/8 = 87.5% when the
-authoring agent rated its own comments; both numbers and the fixes that followed each round are logged in
-[intent.md](intent.md).
+Copy [`.github/workflows-template/codeblast.yml`](.github/workflows-template/codeblast.yml) into your repo.
+PR comments contain structural changes, tested behavior signals, affected tests and evidence links;
+docs-only, test-only and auxiliary-only changes stay silent. The deterministic offline replay matrix
+guards these routing rules. Earlier historical replays and their human-review limitations remain logged
+in [intent.md](intent.md).
 ## Evidence you can rerun
 
 The headline promise is bounded: TypeScript, within the statically analyzable scope, and measured by
@@ -180,6 +184,28 @@ and every published figure has a machine-readable receipt in [`docs/claims.json`
 The pilot is not a benchmark and does not establish a universal accuracy rate. It is a reproducible
 smoke run of the published package against this repository.
 
+### Current automated evidence
+
+The current main branch is guarded by one command:
+
+```bash
+bun run evolution-check
+```
+
+It runs build, tests, typecheck, deterministic replay, full/incremental graph equivalence and the
+published-claims gate. Current committed evidence:
+
+| Evidence | Current result |
+|---|---:|
+| Offline safety replay | 12/12, 0 incorrect silence, 0 invalid evidence, 0 nondeterministic output |
+| Full vs incremental graph | nodes, edges and blind spots equal; deleted/renamed rows removed |
+| Star-project graph completeness | n8n, Excalidraw and OpenCode: 0 extraction failures on pinned commits |
+| Pinned mutation recall | tRPC 28/28; graphql-tools 10/10 |
+| Modern compatibility radar | tRPC bounded sample 3/3; informational, not a release gate |
+
+Literal `import.meta.glob("./fixtures/**/*.tsx")` patterns are represented as conservative imports.
+Non-literal loaders, subprocess boundaries and unresolved runtime dispatch remain explicit blind spots.
+
 ### A reviewer's decision, not a diagram
 
 For agents and CI, `check-change --json` returns a routing decision plus evidence:

package/README.zh-CN.md

@@ -1,4 +1,4 @@
-**codeblast** 是一个确定性代码图谱 CLI，面向 TypeScript / Python 仓库，让开发者和 AI agent 在合并之前就知道改动会炸到哪里。
+**codeblast** 是面向 TypeScript monorepo 和 AI coding agent 的合并前变更安全检查器：告诉你该看哪里、该跑哪些测试，以及图谱是否完整。
 
 <p align="center">
   <img src="assets/readme/hero.svg" width="100%" alt="codeblast — deterministic code graph: know what breaks before you merge"/>
@@ -9,7 +9,7 @@
 </p>
 
 <p align="center">
-  <a href="#三个查询"><img src="https://img.shields.io/badge/TypeScript-函数级-3178c6?style=flat-square" alt="TypeScript function-level"/></a>
+  <a href="#四个视图"><img src="https://img.shields.io/badge/TypeScript-函数级-3178c6?style=flat-square" alt="TypeScript function-level"/></a>
   <a href="#精度承诺有边界有证据"><img src="https://img.shields.io/badge/recall-28%2F28_%3D_100%25-3fb950?style=flat-square" alt="mutation-tested recall 100%"/></a>
   <a href="SKILL.md"><img src="https://img.shields.io/badge/Agent-Skill-7c3aed?style=flat-square" alt="agent skill"/></a>
   <img src="https://img.shields.io/badge/license-MIT-8b949e?style=flat-square" alt="MIT"/>
@@ -17,14 +17,15 @@
 
 ## 是什么
 
-**codeblast 把仓库解析成一份确定性代码图谱，回答改代码前后最贵的三个问题：**
-> 🔗 **[在线交互演示](https://alloevil.github.io/codeblast/)** — tRPC / Tabby / sgp 的实时架构图,点开即可三层下钻
+**codeblast 把仓库解析成确定性代码图谱，为一次改动提供四个视图：**
+> 🔗 **[在线交互演示](https://alloevil.github.io/codeblast/)** — tRPC / Tabby / sgp 的真实架构图，可下钻到源码行
 
 | | 问题 | 命令 |
 |---|---|---|
-| 🎯 | **改这个会炸哪里？** | `impact` — 直接/传递/受影响测试三级清单 |
-| 🔍 | **这个 PR 在结构上改了什么？** | `change` — 符号与依赖边的增删/重命名 |
-| 🗺️ | **这个项目长什么样？** | `archmap` — 模块折叠图 + 循环依赖检测 |
+| 🎯 | **改这个会影响哪里？** | `impact` — 直接、传递与受影响测试 |
+| 🛡️ | **这个改动是否需要重点评审？** | `check-change` — 风险、图健康度、测试与警告 |
+| 🔍 | **结构上改了什么？** | `change` — 符号和依赖边的新增、删除与重命名 |
+| 🗺️ | **项目长什么样？** | `archmap` — 模块/文件/符号下钻与循环依赖 |
 
 给人看（CLI / 交互 HTML / PR 评论），也给 AI agent 用（[SKILL.md](SKILL.md)）——同一份图谱，两个出口。
 
@@ -33,7 +34,7 @@
 ```bash
 npx codeblast demo            # 给当前仓库建图、跑一次 impact 查询、导出架构图
 npm i -g codeblast            # 或全局安装；需要 Node ≥ 22.13（内置 sqlite）或 Bun
-                              # npm 上的版本是 0.3.3，与本仓库一致
+                              # npm 当前版本为 0.3.5
 
 # 作为 agent skill 安装（Claude Code、Codex、Cursor 等）
 npx skills add alloevil/codeblast
@@ -49,7 +50,7 @@ codeblast:   代码 → tsc/AST 确定性解析 → 图谱 → 投影    图 =
 **每个节点和每条静态分析边都带 `file:line` 证据**，可直接打开核对（`co_change` 边在该字段存的是共提交次数而非源码行；文件级节点行号为 1）。
 LLM 在管线里只做一件事：给模块起人话名字——节点归属和边永远来自静态分析。
 
-## 三个查询
+## 四个视图
 
 ```bash
 # 建图：TS monorepo / Python 自动识别，hash 增量更新（tRPC 957 文件：6248 节点 / 17072 边，样例运行见 SKILL.md）
@@ -60,23 +61,45 @@ codeblast impact graph.db "createOrder" --json
 #    → direct 清单 = 必须检查的 callsite；tests 清单 = 必须跑的测试
 #    → 双通道：调用链可达（精确率 ~0.70,优先看）+ import 可达（保守补充,勿跳过）
 
-# ② Change Map —— 两个 ref 之间的结构 diff
+# ② Safety —— 合并前机器判断
+codeblast check-change <repo> <base> <head> --json
+#    → decision / risk / graph_health / affected_test_files / warnings
+
+# ③ Change Map —— 两个 ref 之间的结构 diff
 codeblast change <repo> main~5 main --json
 #    → 意料之外的 edges_added = 改动越界信号
 
-# ③ Architecture Map —— 交互 HTML：模块→文件→符号三层下钻，符号跳源码行
+# ④ Architecture Map —— 交互 HTML：模块→文件→符号三层下钻，符号跳源码行
 codeblast archmap graph.db --out arch.html --repo-url <github-url>
 
 # 可选：git 历史耦合挖掘（协议两端、配置与消费者——静态分析看不见的边）
 codeblast cochange <repo> graph.db
 ```
 
-### PR bot（CI 内跑，宁静默不刷屏）
+### PR bot（CI 内运行，宁静默不刷屏）
+
+复制 [`.github/workflows-template/codeblast.yml`](.github/workflows-template/codeblast.yml) 到目标仓库。
+有意义的结构变化或“生产函数体 + 同 PR 测试”变化会得到 review decision、测试建议和 `file:line` 证据；
+docs-only、test-only、auxiliary-only 噪音默认静默。离线 replay 会自动防止错误静默和无效证据回归。
+
+## 当前自动证据
 
-复制 [`.github/workflows-template/codeblast.yml`](.github/workflows-template/codeblast.yml) 到目标仓库：
-每个 PR 自动评论结构变化 + 影响半径 + 无测试覆盖的新增符号；**无结构变化的 PR 零评论**。
-50 个真实提交回放：42 个正确静默。评论有效率是诚实的弱项——四轮评审（1–3 轮为独立盲评，第 4 轮由现模型担任评审人）为 25% / 75% / 57% / 20%，
-而作者 agent 自评同一批 8 条评论为 7/8 = 87.5%；两个数字与每轮之后的修复都记在 [intent.md](intent.md)。
+```bash
+bun run evolution-check
+```
+
+统一运行 build、测试、类型检查、离线 replay、full/incremental 图等价和 claims gate。当前结果：
+
+| 证据 | 结果 |
+|---|---:|
+| 离线安全 replay | 12/12；错误静默、无效证据、非确定输出均为 0 |
+| 全量与增量图 | nodes / edges / blind spots 相等；删除与重命名后的旧记录已清理 |
+| 明星项目图完整性 | n8n、Excalidraw、OpenCode 固定 commit 均为 0 extraction failure |
+| 固定 mutation gate | tRPC 28/28；graphql-tools 10/10 |
+| modern 兼容性雷达 | tRPC 小样本 3/3；仅 informational，不是 release gate |
+
+字面量 `import.meta.glob("./fixtures/**/*.tsx")` 会展开成保守 import 边；非字面量 loader、
+子进程边界和无法解析的运行时派发仍作为显式 blind spot。
 
 ## 精度承诺（有边界，有证据）
 
@@ -107,13 +130,13 @@ codeblast cochange <repo> graph.db
 
 ## 给 AI Agent 用
 
-```
-改前:  impact "symbol" --json   → callsite 清单进上下文，防漏改
-改后:  change HEAD~1 HEAD --json → 自查结构越界与意外删除
+```text
+改前：impact "symbol" --json → review_first / run_tests / conservative / warnings
+改后：check-change repo base head --json → decision / risk / graph_health / affected tests
 ```
 
-完整契约与解读纪律（含"禁止假装盲区清单完整"）见 [SKILL.md](SKILL.md)。
-Agent 规范另见 [AGENTS.md](AGENTS.md)。
+`safe-to-review` 只是注意力路由，不等于“可以安全合并”。完整契约、schema version、exit code
+和解读纪律见 [SKILL.md](SKILL.md)；Agent 规范另见 [AGENTS.md](AGENTS.md)。
 
 ## 常见问题
 

package/dist/bin.js

@@ -467,21 +467,24 @@ class Extractor {
         enclosing.push(id);
       if (ts.isCallExpression(node) || ts.isNewExpression(node)) {
         const caller = [...enclosing].reverse().find(Boolean) ?? relPath;
-        if (ts.isCallExpression(node) && node.expression.kind === ts.SyntaxKind.ImportKeyword) {
+        if (ts.isCallExpression(node) && ts.isPropertyAccessExpression(node.expression) && node.expression.name.text === "glob" && node.expression.expression.getText(sf) === "import.meta") {
+          const arg = node.arguments[0];
+          const patterns = arg && ts.isStringLiteral(arg) ? [arg.text] : arg && ts.isArrayLiteralExpression(arg) && arg.elements.every(ts.isStringLiteral) ? arg.elements.map((element) => element.text) : null;
+          if (patterns) {
+            const excluded = new Set(patterns.filter((pattern) => pattern.startsWith("!")).flatMap((pattern) => this.expandGlob(pattern.slice(1), sf.fileName)));
+            const matches = new Set(patterns.filter((pattern) => !pattern.startsWith("!")).flatMap((pattern) => this.expandGlob(pattern, sf.fileName)));
+            for (const match of matches)
+              if (!excluded.has(match))
+                edges.push({ src: relPath, dst: this.rel(match), kind: "imports", file: relPath, line: lineOf(node), confidence: "conservative", src_file: relPath });
+          } else {
+            blindSpots.push({ file: relPath, line: lineOf(node), reason: `dynamic import.meta.glob: ${(arg?.getText() ?? "").slice(0, 80)}`, src_file: relPath });
+          }
+        } else if (ts.isCallExpression(node) && node.expression.kind === ts.SyntaxKind.ImportKeyword) {
           const arg = node.arguments[0];
           if (arg && ts.isStringLiteral(arg)) {
             const resolved = this.resolveModule(arg.text, sf.fileName);
-            if (resolved) {
-              edges.push({
-                src: relPath,
-                dst: this.rel(resolved),
-                kind: "imports",
-                file: relPath,
-                line: lineOf(node),
-                confidence: "exact",
-                src_file: relPath
-              });
-            }
+            if (resolved)
+              edges.push({ src: relPath, dst: this.rel(resolved), kind: "imports", file: relPath, line: lineOf(node), confidence: "exact", src_file: relPath });
           } else {
             blindSpots.push({ file: relPath, line: lineOf(node), reason: `dynamic import: ${(arg?.getText() ?? "").slice(0, 80)}`, src_file: relPath });
           }
@@ -659,6 +662,28 @@ class Extractor {
   isExported(node) {
     return (ts.getCombinedModifierFlags(node) & ts.ModifierFlags.Export) !== 0;
   }
+  expandGlob(pattern, fromFile) {
+    if (!pattern.startsWith("."))
+      return [];
+    const base = path.resolve(path.dirname(fromFile), pattern.split("*")[0]);
+    if (!fs.existsSync(base))
+      return [];
+    const rawSuffix = pattern.slice(pattern.indexOf("*"));
+    const suffix = rawSuffix.split("**").map((part) => part.split("*").map((piece) => piece.replace(/[.+?^${}()|[\]\\]/g, "\\$&")).join("[^/]*")).join(".*");
+    const re = new RegExp(`^${suffix}$`);
+    const out = [];
+    const walk = (dir) => {
+      for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {
+        const file = path.join(dir, entry.name);
+        if (entry.isDirectory())
+          walk(file);
+        else if (re.test(path.relative(base, file).replaceAll(path.sep, "/")))
+          out.push(file);
+      }
+    };
+    walk(base);
+    return out;
+  }
 }
 var TEST_FILE_RE;
 var init_extract = __esm(() => {
@@ -879,25 +904,33 @@ var init_cli = __esm(() => {
   }
   pyProbe = spawnSync(["python3", path2.join(import.meta.dirname, "py_extract.py"), repoRoot]);
   if (pyProbe.exitCode === 0) {
-    const payload = JSON.parse(pyProbe.stdout);
-    for (const f of payload.files) {
-      if (seenFiles.has(f.path))
-        continue;
-      seenFiles.add(f.path);
-      const existing = getHash.get(f.path);
-      if (existing?.hash === f.hash && f.hash !== "") {
-        skipped++;
-        continue;
+    try {
+      const payload = JSON.parse(pyProbe.stdout);
+      for (const f of payload.files) {
+        if (seenFiles.has(f.path))
+          continue;
+        seenFiles.add(f.path);
+        const existing = getHash.get(f.path);
+        if (existing?.hash === f.hash && f.hash !== "") {
+          skipped++;
+          continue;
+        }
+        writeBatch(f.path, f.hash, f.nodes, f.edges, f.blind_spots);
+        indexed++;
+        nodeCount += f.nodes.length;
+        edgeCount += f.edges.length;
+        blindCount += f.blind_spots.length;
       }
-      writeBatch(f.path, f.hash, f.nodes, f.edges, f.blind_spots);
-      indexed++;
-      nodeCount += f.nodes.length;
-      edgeCount += f.edges.length;
-      blindCount += f.blind_spots.length;
+      const pyFiles = payload.files.length;
+      if (pyFiles > 0)
+        console.error(`python files ingested: ${pyFiles}`);
+    } catch (err) {
+      failures++;
+      console.error(`PYTHON INGEST FAILED: ${err instanceof Error ? err.message : err}`);
     }
-    const pyFiles = payload.files.length;
-    if (pyFiles > 0)
-      console.error(`python files ingested: ${pyFiles}`);
+  } else {
+    failures++;
+    console.error(`PYTHON INGEST FAILED: ${pyProbe.stderr.slice(0, 1000)}`);
   }
   if (failures === 0) {
     const stale = db.prepare("SELECT path FROM files").all().filter((row) => !seenFiles.has(row.path));
@@ -2348,7 +2381,7 @@ var SCHEMA_VERSION = "1", EXIT_OK = 0, EXIT_REVIEW = 1, EXIT_ERROR = 2, readVers
   nodes: Number(db.prepare("SELECT COUNT(*) c FROM nodes").get().c),
   edges: Number(db.prepare("SELECT COUNT(*) c FROM edges").get().c),
   blind_spots: Number(db.prepare("SELECT COUNT(*) c FROM blind_spots").get().c)
-}), healthBase, healthHead, diff3, healthWarnings, prodNodesAdded2, total3, affectedTests = 0, truncated = false, blindSpots = 0, bodyChanged2, structuralIds2, diffText, currentFile = "", diffLineCount2, changedFiles2, auxOnly, decision2, output;
+}), healthBase, healthHead, diff3, healthWarnings, prodNodesAdded2, total3, affectedTestFiles, truncated = false, blindSpots = 0, affectedTests, bodyChanged2, structuralIds2, diffText, currentFile = "", diffLineCount2, changedFiles2, auxOnly, decision2, output;
 var init_check_change = __esm(() => {
   init_db();
   init_proc();
@@ -2384,14 +2417,18 @@ var init_check_change = __esm(() => {
   ];
   prodNodesAdded2 = diff3.nodesAdded.filter((n) => !TEST_RE3.test(n.file));
   total3 = structuralTotal(diff3);
+  affectedTestFiles = new Set;
   for (const node of [...diff3.nodesAdded, ...diff3.renamed.map((r) => ({ id: `${r.file}#${r.to}`, kind: r.kind, name: r.to, file: r.file, line: 0 }))].slice(0, 15)) {
     try {
       const result = impact(dbB3, node.id, 2000);
-      affectedTests += new Set(result.items.filter((item) => item.level === "tests").map((item) => item.file)).size;
+      for (const item of result.items)
+        if (item.level === "tests")
+          affectedTestFiles.add(item.file);
       truncated ||= result.truncated;
       blindSpots += result.blind_spot_count;
     } catch {}
   }
+  affectedTests = affectedTestFiles.size;
   bodyChanged2 = [];
   structuralIds2 = new Set([
     ...diff3.nodesAdded.map((node) => node.id),
@@ -2442,6 +2479,7 @@ var init_check_change = __esm(() => {
     recommended_actions: decision2.recommendedActions,
     structural_changes: total3,
     affected_test_files: affectedTests,
+    affected_test_file_paths: [...affectedTestFiles].sort((a, b) => a.localeCompare(b)),
     blind_spot_count: blindSpots,
     graph_health: { base: healthBase, head: healthHead, warnings: healthWarnings },
     truncated,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeblast",
-  "version": "0.3.4",
+  "version": "0.3.5",
   "description": "Know what breaks before you merge — mutation-tested code graph with architecture, change & impact maps. Evidence on every edge. For humans and AI agents.",
   "keywords": [
     "impact-analysis",
@@ -39,6 +39,7 @@
     "prepack": "bun run build",
     "typecheck": "tsc -p .",
     "agent-smoke": "bun run build && node eval/agent-workflow-smoke.mjs",
+    "evolution-check": "PYTHONPATH=. node eval/evolution-check.mjs",
     "guidance-stability": "node eval/guidance-stability.mjs",
     "incremental-equivalence": "bun run build && node eval/incremental-equivalence.mjs",
     "offline-replay": "bun run build && node eval/offline-replay.mjs",