sigmap 8.2.0 → 8.4.0

package/CHANGELOG.md

@@ -10,6 +10,20 @@ Format: [Semantic Versioning](https://semver.org/)
 
 ---
 
+## [8.4.0] — 2026-07-05
+
+Minor release — **PR Evidence Report (v9.0 G3): a branded, deterministic review artifact.** SigMap already had the pieces — `review-pr` findings and `get_diff_context` — but no single Markdown comment an agent or CI could post on a PR. This adds it: one report that answers *"what changed, what it touches, and what to test"*, with no LLM.
+
+### Added
+- **PR Evidence Report (#417, PR #418):** new `src/review/pr-evidence.js` — `buildPrEvidence(changedFiles, cwd)` folds together, per changed file, its extracted **signatures**, **blast radius** (direct/transitive importers, impacted tests + routes), cross-language **related tests**, a **risk label**, and the **`review-pr` findings** (scope drift, god-node edits, missing tests, security-sensitive files). `formatPrEvidenceMarkdown` renders the branded **"🔍 PR Evidence Report"** — with **no wall-clock timestamp**, so it's byte-stable given a fixed tree (diff-friendly as a comment). Exposed via `sigmap review-pr --markdown` (alias `--evidence`); honors `--staged`/`--base`; the exit code reflects the review pass/fail so CI can both post the comment and gate on it. Reuses shipped zero-dep modules only; git stays behind the shell-free `git()` util.
+
+## [8.3.0] — 2026-07-05
+
+Minor release — **Python site-packages grounding: the moat now spans both major ecosystems.** v8.1/v8.2 built local-library grounding for JS/TS (`node_modules` `.d.ts`); this extends it to **Python**, so `verify-ai-output` and the `verify_suggestion` MCP tool ground AI-suggested Python code against the libraries actually installed in the project's venv — with pinned versions (D8). Zero-dependency, no Python runtime, deterministic.
+
+### Added
+- **Python site-packages grounding (#413, PR #414):** `buildLibraryIndex` (`src/verify/lib-index.js`) gains a Python pass alongside the JS/TS one. It reads direct deps from `requirements.txt` / `pyproject.toml` (PEP 621 `[project].dependencies` + Poetry), discovers the venv `site-packages` (`.venv|venv|env` → `lib/python*/site-packages`, or `Lib/site-packages` on Windows) **without spawning Python**, resolves each dep's installed module + version (`*.dist-info`, D8) with PEP 503 import-name normalization, and extracts exported names from the package's `__init__.py`/`.pyi` (`__all__`, top-level `def`/`class`, public assignments, and `from … import` re-exports). Both ecosystems merge into one symbol index — genuine installed-Python-library calls stop being false-flagged as `fake-symbol`. Byte-stable given a fixed installed tree; cached via `src/cache/sig-cache.js`; graceful on missing venv / unresolved deps.
+
 ## [8.2.0] — 2026-07-04
 
 Minor release — **`verify_suggestion` MCP tool: the grounding moat, made consumable by agents.** v8.1.0 built local-library grounding inside the `verify-ai-output` CLI; this exposes it as the **18th MCP tool**, so a coding agent can verify its own generated code against the repo **and the libraries actually installed** in `node_modules` — *before it writes* — and get back the flagged issues plus the pinned versions it verified against (D8).

package/README.md

@@ -98,7 +98,7 @@ Ask → Rank → Context → Validate → Judge → Learn
 
 <!--SM:benchmarkBlock-->
 ```
-Benchmark : sigmap-v8.2-main (21 repositories, including R language)
+Benchmark : sigmap-v8.4-main (21 repositories, including R language)
 Date      : 2026-07-04
 
 Hit@5          : 86.7%   (baseline 13.6%  — 6.4× lift)

package/gen-context.js

@@ -13108,7 +13108,7 @@ __factories["./src/mcp/server"] = function(module, exports) {
 
   const SERVER_INFO = {
     name: 'sigmap',
-    version: '8.2.0',
+    version: '8.4.0',
     description: 'SigMap MCP server — code signatures on demand',
   };
 
@@ -14610,6 +14610,149 @@ __factories["./src/retrieval/tokenizer"] = function(module, exports) {
   
 };
 
+// ── ./src/review/pr-evidence ──
+__factories["./src/review/pr-evidence"] = function(module, exports) {
+  
+  /**
+   * PR Evidence Report (v9.0 G3).
+   *
+   * A single, branded, deterministic Markdown artifact for code review: for each
+   * changed file it folds together the signature context, blast radius (direct /
+   * transitive importers, impacted tests + routes), cross-language related tests,
+   * a risk label, and the `review-pr` findings (scope drift, god-node edits,
+   * missing tests, security-sensitive files). Posted as a PR comment, it answers
+   * "what changed, what it touches, and what to test" — without an LLM.
+   *
+   * Built entirely from shipped zero-dep modules (reviewPr, graph/impact,
+   * evidence/pack, extractors/dispatch). Carries NO wall-clock timestamp, so the
+   * report is byte-stable given a fixed tree — diff-friendly as a comment.
+   */
+
+  const fs = require('fs');
+  const path = require('path');
+  const { reviewPr } = __require('./src/review/review-pr');
+
+  /**
+   * Build the structured PR evidence for a changed-file list.
+   * @param {Array<{path:string,status?:string}>|string[]} changedFiles
+   * @param {string} cwd
+   * @param {object} [opts]
+   * @param {number} [opts.depth=2]   blast-radius BFS depth
+   * @param {string} [opts.scope]     label for the diff scope (e.g. "vs main")
+   * @returns {{ scope:string, files:object[], review:object }}
+   */
+  function buildPrEvidence(changedFiles, cwd, opts = {}) {
+    const files = (changedFiles || []).map((f) =>
+      typeof f === 'string' ? { path: f, status: 'M' } : { path: f.path, status: f.status || 'M' });
+
+    const review = reviewPr(files, cwd, opts);
+
+    let riskLabelFor = () => 'source';
+    let findRelatedTests = () => [];
+    try { ({ riskLabelFor, findRelatedTests } = __require('./src/evidence/pack')); } catch (_) { /* defaults */ }
+    const { extractFile, langFor } = __require('./src/extractors/dispatch');
+
+    let allFiles = [];
+    try { const { buildSigIndex } = __require('./src/retrieval/ranker'); allFiles = [...buildSigIndex(cwd).keys()]; } catch (_) { /* no index */ }
+
+    const depth = Number.isFinite(opts.depth) ? opts.depth : 2;
+    const srcPaths = files.filter((f) => f.status !== 'D' && langFor(f.path)).map((f) => f.path);
+    let impactByFile = new Map();
+    try {
+      const { analyzeImpact } = __require('./src/graph/impact');
+      impactByFile = new Map(analyzeImpact(srcPaths, cwd, { depth }).map((r) => [r.file, r.impact]));
+    } catch (_) { /* graph optional */ }
+
+    const fileReports = files.map((f) => {
+      const deleted = f.status === 'D';
+      let signatures = [];
+      if (!deleted && langFor(f.path)) {
+        try { signatures = extractFile(f.path, fs.readFileSync(path.resolve(cwd, f.path), 'utf8')); } catch (_) { /* unreadable */ }
+      }
+      const impact = impactByFile.get(f.path) || null;
+      return {
+        path: f.path,
+        status: f.status,
+        riskLabel: riskLabelFor(f.path),
+        signatures,
+        blast: impact ? {
+          total: impact.totalImpact,
+          direct: impact.direct || [],
+          transitive: (impact.transitive || []).length,
+          tests: impact.tests || [],
+          routes: impact.routes || [],
+        } : null,
+        relatedTests: deleted ? [] : findRelatedTests(f.path, allFiles),
+      };
+    });
+
+    return { scope: opts.scope || 'diff', files: fileReports, review };
+  }
+
+  const STATUS_LABEL = { M: 'modified', A: 'added', D: 'deleted', R: 'renamed', C: 'copied' };
+
+  /** Render the branded, deterministic "PR Evidence Report" Markdown. */
+  function formatPrEvidenceMarkdown(evidence, opts = {}) {
+    const L = [];
+    const s = evidence.review.summary;
+    const maxSigs = Number.isFinite(opts.maxSignatures) ? opts.maxSignatures : 30;
+
+    L.push('## 🔍 PR Evidence Report');
+    L.push('');
+    L.push(
+      `**${s.filesChanged} file(s) changed** — ${s.sourceChanged} source, ${s.testsChanged} test · ` +
+      (s.ok ? '✅ no review findings' : `⚠️ ${s.findings} finding(s)`) +
+      ` · scope: ${evidence.scope}`
+    );
+    L.push('');
+
+    if (!s.ok) {
+      L.push('### Review findings');
+      for (const f of evidence.review.findings) {
+        if (f.type === 'missing-tests') L.push(`- ⚠️ **missing tests** — \`${f.file}\` changed with no matching test`);
+        else if (f.type === 'security-file') L.push(`- ⚠️ **security-sensitive file** — \`${f.file}\``);
+        else if (f.type === 'god-node') L.push(`- ⚠️ **god node** — \`${f.file}\` → ${f.count} dependents (high blast radius)`);
+        else if (f.type === 'scope-drift') L.push(`- ⚠️ **scope drift** — ${f.count} top-level dirs touched (${f.dirs.join(', ')})`);
+      }
+      L.push('');
+    }
+
+    L.push('### Changed files');
+    for (const f of evidence.files) {
+      const st = STATUS_LABEL[f.status] || f.status;
+      L.push(`#### \`${f.path}\`  _(${st} · risk: ${f.riskLabel})_`);
+      if (f.status === 'D') { L.push('_deleted_', ''); continue; }
+
+      if (f.blast) {
+        L.push(
+          `**Blast radius:** ${f.blast.total} file(s) impacted — ${f.blast.direct.length} direct, ${f.blast.transitive} transitive` +
+          (f.blast.tests.length ? `, ${f.blast.tests.length} test(s)` : '') +
+          (f.blast.routes.length ? `, ${f.blast.routes.length} route(s)` : '')
+        );
+        if (f.blast.tests.length) L.push(`Tests to run: ${f.blast.tests.slice(0, 8).map((t) => '`' + t + '`').join(', ')}`);
+      } else {
+        L.push('**Blast radius:** _(not in dependency graph — new or leaf file)_');
+      }
+      if (f.relatedTests.length) L.push(`Related tests: ${f.relatedTests.slice(0, 8).map((t) => '`' + t + '`').join(', ')}`);
+
+      if (f.signatures.length) {
+        L.push('```');
+        for (const sig of f.signatures.slice(0, maxSigs)) L.push(sig);
+        if (f.signatures.length > maxSigs) L.push(`… +${f.signatures.length - maxSigs} more`);
+        L.push('```');
+      }
+      L.push('');
+    }
+
+    L.push('---');
+    L.push('_Deterministic PR Evidence Report — generated by [SigMap](https://sigmap.io). No LLM; byte-stable given a fixed tree._');
+    return L.join('\n');
+  }
+
+  module.exports = { buildPrEvidence, formatPrEvidenceMarkdown };
+  
+};
+
 // ── ./src/review/review-pr ──
 __factories["./src/review/review-pr"] = function(module, exports) {
   
@@ -16805,12 +16948,17 @@ __factories["./src/verify/lib-index"] = function(module, exports) {
    * in `node_modules` and verify AI suggestions against repo + private +
    * installed-lib symbols. This module builds the installed-lib half.
    *
-   * For each **direct** dependency declared in `package.json`, it locates the
-   * package under `node_modules/<dep>`, reads its version (D8 version pinning),
-   * and extracts the exported symbol names from its TypeScript declaration entry
-   * (`types`/`typings`, else `index.d.ts`). Pure, zero-dependency, deterministic:
-   * byte-stable given a fixed installed tree. Bounded (per-file read cap + dep
-   * cap) and cached via `src/cache/sig-cache.js` so repeat builds are near-free.
+   * Two ecosystems, one index:
+   *   - **JS/TS** — each **direct** dependency in `package.json` resolved under
+   *     `node_modules/<dep>`; exports read from its TypeScript declaration entry
+   *     (`types`/`typings`, else `index.d.ts`).
+   *   - **Python** — each direct dependency in `requirements.txt`/`pyproject.toml`
+   *     resolved in the project's venv `site-packages`; exports read from the
+   *     package's `__init__.py`/`.pyi`. No Python runtime is spawned (North-Star #1).
+   *
+   * Pure, zero-dependency, deterministic: byte-stable given a fixed installed
+   * tree. Bounded (per-file read cap + dep cap) and cached via
+   * `src/cache/sig-cache.js` so repeat builds are near-free.
    */
 
   const fs = require('fs');
@@ -16820,6 +16968,7 @@ __factories["./src/verify/lib-index"] = function(module, exports) {
   const MAX_DTS_BYTES = 512 * 1024; // per-file read cap
   const MAX_DEPS = 1000;            // dep count cap
   const DEP_KEYS = ['dependencies', 'devDependencies', 'peerDependencies', 'optionalDependencies'];
+  const VENV_DIRS = ['.venv', 'venv', 'env', '.env'];
 
   /**
    * Extract exported symbol names from a `.d.ts` declaration file. Deterministic,
@@ -16895,6 +17044,130 @@ __factories["./src/verify/lib-index"] = function(module, exports) {
     return { version, dtsPath: null }; // installed but untyped
   }
 
+  // ── Python ──────────────────────────────────────────────────────────────────
+
+  /**
+   * Extract exported symbol names from a Python module's `__init__.py`/`.pyi`.
+   * Deterministic, regex-based, top-level only: `__all__`, `def`/`class`, public
+   * module-level assignments, and `from … import …` re-exports (a package's
+   * public API is largely re-exports). Private names (leading `_`) are skipped
+   * unless listed in `__all__`.
+   * @param {string} src
+   * @returns {string[]} sorted unique exported names
+   */
+  function extractPyExports(src) {
+    const names = new Set();
+    if (!src) return [];
+
+    // __all__ = [ 'a', 'b', ... ]  (authoritative when present; keeps privates)
+    const allMatch = src.match(/^__all__\s*[:+]?=\s*[\[(]([\s\S]*?)[\])]/m);
+    if (allMatch) {
+      for (const m of allMatch[1].matchAll(/['"]([A-Za-z_]\w*)['"]/g)) names.add(m[1]);
+    }
+
+    // top-level def / class (column 0)
+    for (const m of src.matchAll(/^(?:async\s+)?def\s+([A-Za-z_]\w*)/gm)) if (!m[1].startsWith('_')) names.add(m[1]);
+    for (const m of src.matchAll(/^class\s+([A-Za-z_]\w*)/gm)) if (!m[1].startsWith('_')) names.add(m[1]);
+
+    // top-level public assignments: NAME = …  /  NAME: type = …  (not ==, +=, etc.)
+    for (const m of src.matchAll(/^([A-Za-z_]\w*)\s*(?::[^=\n]+)?=(?!=)/gm)) {
+      if (!m[1].startsWith('_')) names.add(m[1]);
+    }
+
+    // re-exports: from .mod import Name, Other as Alias
+    for (const m of src.matchAll(/^from\s+[^\n]+?\s+import\s+([^\n#]+)/gm)) {
+      for (const part of m[1].split(',')) {
+        const name = part.trim().replace(/[()]/g, '').split(/\s+as\s+/).pop().trim();
+        if (/^[A-Za-z_]\w*$/.test(name) && !name.startsWith('_')) names.add(name);
+      }
+    }
+
+    return [...names].sort();
+  }
+
+  /** Read direct Python dependency names from requirements.txt + pyproject.toml. */
+  function pythonDirectDeps(cwd) {
+    const names = new Set();
+    try {
+      const req = fs.readFileSync(path.join(cwd, 'requirements.txt'), 'utf8');
+      for (const line of req.split('\n')) {
+        const t = line.trim();
+        if (!t || t.startsWith('#') || t.startsWith('-')) continue;
+        const m = t.match(/^([A-Za-z0-9][A-Za-z0-9._-]*)/);
+        if (m) names.add(m[1]);
+      }
+    } catch (_) { /* none */ }
+    try {
+      const py = fs.readFileSync(path.join(cwd, 'pyproject.toml'), 'utf8');
+      // PEP 621: [project] dependencies = ["foo>=1", "bar"]
+      const projDeps = py.match(/^\s*dependencies\s*=\s*\[([\s\S]*?)\]/m);
+      if (projDeps) for (const m of projDeps[1].matchAll(/['"]([A-Za-z0-9][A-Za-z0-9._-]*)/g)) names.add(m[1]);
+      // Poetry: [tool.poetry.dependencies]\n foo = "^1"
+      const poetry = py.match(/\[tool\.poetry\.dependencies\]([\s\S]*?)(?:\n\[|$)/);
+      if (poetry) for (const m of poetry[1].matchAll(/^([A-Za-z0-9][A-Za-z0-9._-]*)\s*=/gm)) {
+        if (m[1] !== 'python') names.add(m[1]);
+      }
+    } catch (_) { /* none */ }
+    return [...names].sort();
+  }
+
+  /** Locate the project's venv `site-packages` directories (no Python runtime). */
+  function findSitePackages(cwd) {
+    const out = [];
+    for (const v of VENV_DIRS) {
+      const base = path.join(cwd, v);
+      const libDir = path.join(base, 'lib'); // POSIX: <venv>/lib/pythonX.Y/site-packages
+      let pyDirs = [];
+      try { pyDirs = fs.readdirSync(libDir).filter((d) => /^python\d/.test(d)).sort(); } catch (_) { /* none */ }
+      for (const py of pyDirs) {
+        const sp = path.join(libDir, py, 'site-packages');
+        try { if (fs.statSync(sp).isDirectory()) out.push(sp); } catch (_) { /* next */ }
+      }
+      const winSp = path.join(base, 'Lib', 'site-packages'); // Windows
+      try { if (fs.statSync(winSp).isDirectory()) out.push(winSp); } catch (_) { /* next */ }
+    }
+    return out;
+  }
+
+  /** PEP 503 name normalization (case-insensitive, `-`/`_`/`.` collapsed). */
+  function normalizePy(name) {
+    return String(name).toLowerCase().replace(/[-_.]+/g, '-');
+  }
+
+  /** Find an installed distribution's version from its `*.dist-info`/`*.egg-info`. */
+  function findPyVersion(sitePkgsDir, dep) {
+    const norm = normalizePy(dep);
+    let entries;
+    try { entries = fs.readdirSync(sitePkgsDir); } catch (_) { return null; }
+    for (const e of entries.sort()) {
+      const m = e.match(/^(.+?)-(\d[^-]*)\.(?:dist-info|egg-info)$/);
+      if (m && normalizePy(m[1]) === norm) return m[2];
+    }
+    return null;
+  }
+
+  /**
+   * Resolve a Python dependency to its installed module entry file + version.
+   * @returns {{ version: string|null, sourcePath: string|null }|null} null if not installed
+   */
+  function resolvePyEntry(sitePkgsDirs, dep) {
+    const candidates = [...new Set([dep, dep.replace(/-/g, '_'), dep.toLowerCase(), dep.toLowerCase().replace(/-/g, '_')])];
+    for (const sp of sitePkgsDirs) {
+      const version = findPyVersion(sp, dep);
+      for (const cand of candidates) {
+        for (const entry of ['__init__.pyi', '__init__.py']) { // package
+          const p = path.join(sp, cand, entry);
+          try { if (fs.statSync(p).isFile()) return { version, sourcePath: p }; } catch (_) { /* next */ }
+        }
+        for (const ext of ['.pyi', '.py']) { // single-module
+          const p = path.join(sp, cand + ext);
+          try { if (fs.statSync(p).isFile()) return { version, sourcePath: p }; } catch (_) { /* next */ }
+        }
+      }
+    }
+    return null;
+  }
+
   /**
    * Build the installed-library signature index for `cwd`.
    *
@@ -16907,17 +17180,24 @@ __factories["./src/verify/lib-index"] = function(module, exports) {
   function buildLibraryIndex(cwd, opts = {}) {
     const version = opts.version || '0';
     const useCache = opts.cache !== false;
-    const deps = directDeps(cwd).slice(0, MAX_DEPS);
 
-    const entries = [];
-    for (const dep of deps) {
+    // Collect entries from both ecosystems; each carries its extractor kind.
+    const entries = []; // { name, version, sourcePath, kind: 'dts'|'py' }
+    for (const dep of directDeps(cwd).slice(0, MAX_DEPS)) {
       const r = resolveEntry(cwd, dep);
-      if (r) entries.push({ dep, version: r.version, dtsPath: r.dtsPath });
+      if (r) entries.push({ name: dep, version: r.version, sourcePath: r.dtsPath, kind: 'dts' });
+    }
+    const sitePkgs = findSitePackages(cwd);
+    if (sitePkgs.length) {
+      for (const dep of pythonDirectDeps(cwd).slice(0, MAX_DEPS)) {
+        const r = resolvePyEntry(sitePkgs, dep);
+        if (r) entries.push({ name: dep, version: r.version, sourcePath: r.sourcePath, kind: 'py' });
+      }
     }
 
     const cache = useCache ? loadCache(cwd, version) : new Map();
-    const dtsFiles = entries.filter((e) => e.dtsPath).map((e) => e.dtsPath);
-    const { unchanged } = getChangedFiles(dtsFiles, cache);
+    const files = entries.filter((e) => e.sourcePath).map((e) => e.sourcePath);
+    const { unchanged } = getChangedFiles(files, cache);
     const unchangedSet = new Set(unchanged);
 
     const symbols = new Set();
@@ -16926,20 +17206,20 @@ __factories["./src/verify/lib-index"] = function(module, exports) {
 
     for (const e of entries) {
       let names;
-      if (!e.dtsPath) {
+      if (!e.sourcePath) {
         names = [];
-      } else if (unchangedSet.has(e.dtsPath) && cache.get(e.dtsPath)) {
-        names = cache.get(e.dtsPath).sigs || [];
+      } else if (unchangedSet.has(e.sourcePath) && cache.get(e.sourcePath)) {
+        names = cache.get(e.sourcePath).sigs || [];
       } else {
         let src = '';
         try {
-          if (fs.statSync(e.dtsPath).size <= MAX_DTS_BYTES) src = fs.readFileSync(e.dtsPath, 'utf8');
+          if (fs.statSync(e.sourcePath).size <= MAX_DTS_BYTES) src = fs.readFileSync(e.sourcePath, 'utf8');
         } catch (_) { /* unreadable → empty */ }
-        names = extractDtsExports(src);
-        fresh.push({ file: e.dtsPath, sigs: names });
+        names = e.kind === 'py' ? extractPyExports(src) : extractDtsExports(src);
+        fresh.push({ file: e.sourcePath, sigs: names });
       }
       for (const n of names) symbols.add(n);
-      libraries.push({ name: e.dep, version: e.version, symbols: names.length, typed: !!e.dtsPath });
+      libraries.push({ name: e.name, version: e.version, symbols: names.length, typed: !!e.sourcePath });
     }
 
     if (useCache && fresh.length) {
@@ -16958,7 +17238,10 @@ __factories["./src/verify/lib-index"] = function(module, exports) {
       .map((l) => `${l.name}@${l.version}`);
   }
 
-  module.exports = { buildLibraryIndex, extractDtsExports, directDeps, resolveEntry, formatVersionPins };
+  module.exports = {
+    buildLibraryIndex, extractDtsExports, directDeps, resolveEntry, formatVersionPins,
+    extractPyExports, pythonDirectDeps, findSitePackages, resolvePyEntry,
+  };
   
 };
 
@@ -17291,7 +17574,7 @@ function __tryGit(args, opts = {}) {
   catch (_) { return ''; }
 }
 
-const VERSION = '8.2.0';
+const VERSION = '8.4.0';
 const MARKER = '\n\n## Auto-generated signatures\n<!-- Updated by gen-context.js -->\n';
 
 function requireSourceOrBundled(key) {
@@ -19092,7 +19375,8 @@ Usage:
   ${cmd} conventions                       Extract repo file-naming/export/test conventions (--conflicts, --inject, --report, --fix)
   ${cmd} scaffold "<name>"                 Propose a convention-matched file/dir scaffold (--ext, --threshold, --force, --json)
   ${cmd} verify-plan <plan.md|->           Check a plan vs the live index — files/symbols exist, blast radius, scope (--json)
-  ${cmd} review-pr                         Audit a diff — scope drift, god-node edits, missing tests, security files (--staged, --json)
+  ${cmd} review-pr                         Audit a diff — scope drift, god-node edits, missing tests, security files (--staged, --base, --json, --markdown)
+  ${cmd} review-pr --markdown              PR Evidence Report — branded Markdown (signatures + blast radius + tests) to post as a PR comment
   ${cmd} create "<task>"                   Grounded-creation pipeline: scaffold → verify-plan → verify-ai-output → review-pr (--staged)
   ${cmd} squeeze <file|->                  Minimize a pasted stacktrace/CI-log/JSON blob (--json for stats)
   ${cmd} ask "<query>" --squeeze           Auto-accept input minimization (no prompt; for scripts/CI)
@@ -21033,6 +21317,15 @@ function main() {
       return { path: file, status };
     });
 
+    // --markdown / --evidence: emit the branded, deterministic PR Evidence Report.
+    if (args.includes('--markdown') || args.includes('--evidence')) {
+      const { buildPrEvidence, formatPrEvidenceMarkdown } = requireSourceOrBundled('./src/review/pr-evidence');
+      const scope = staged ? 'staged' : (baseArg ? `vs ${baseArg}` : 'branch');
+      const ev = buildPrEvidence(changedFiles, cwd, { scope });
+      process.stdout.write(formatPrEvidenceMarkdown(ev) + '\n');
+      process.exit(ev.review.summary.ok ? 0 : 1);
+    }
+
     const { reviewPr } = requireSourceOrBundled('./src/review/review-pr');
     const result = reviewPr(changedFiles, cwd, {});
 

package/llms-full.txt

@@ -11,13 +11,13 @@ ranking keeps the relevant context in scope (cutting tokens ~97% as a side
 effect), with no LLM calls, embeddings, or vector database. Works with Claude,
 Cursor, GitHub Copilot, Aider, Windsurf, local LLMs, and MCP.
 
-# Version: 8.2.0 | Benchmark: sigmap-v8.2-main (2026-07-04)
+# Version: 8.4.0 | Benchmark: sigmap-v8.4-main (2026-07-04)
 # Source: auto-generated from package.json, version.json, benchmarks/latest.json, src/mcp/tools.js, src/config/defaults.js
 # Regenerate: npm run generate:llms   |   Validate: npm run validate:llms
 
 ---
 
-## Core metrics (benchmark: sigmap-v8.2-main, 2026-07-04)
+## Core metrics (benchmark: sigmap-v8.4-main, 2026-07-04)
 
 | Metric | Without SigMap | With SigMap |
 |--------|----------------|-------------|
@@ -105,7 +105,8 @@ sigmap verify-ai-output <answer.md> --report  Write a standalone HTML report (re
 sigmap conventions                       Extract repo file-naming/export/test conventions (--conflicts, --inject, --report, --fix)
 sigmap scaffold "<name>"                 Propose a convention-matched file/dir scaffold (--ext, --threshold, --force, --json)
 sigmap verify-plan <plan.md|->           Check a plan vs the live index — files/symbols exist, blast radius, scope (--json)
-sigmap review-pr                         Audit a diff — scope drift, god-node edits, missing tests, security files (--staged, --json)
+sigmap review-pr                         Audit a diff — scope drift, god-node edits, missing tests, security files (--staged, --base, --json, --markdown)
+sigmap review-pr --markdown              PR Evidence Report — branded Markdown (signatures + blast radius + tests) to post as a PR comment
 sigmap create "<task>"                   Grounded-creation pipeline: scaffold → verify-plan → verify-ai-output → review-pr (--staged)
 sigmap squeeze <file|->                  Minimize a pasted stacktrace/CI-log/JSON blob (--json for stats)
 sigmap ask "<query>" --squeeze           Auto-accept input minimization (no prompt; for scripts/CI)

package/llms.txt

@@ -11,7 +11,7 @@ ranking keeps the relevant context in scope (cutting tokens ~97% as a side
 effect), with no LLM calls, embeddings, or vector database. Works with Claude,
 Cursor, GitHub Copilot, Aider, Windsurf, local LLMs, and MCP.
 
-# Version: 8.2.0 | Benchmark: sigmap-v8.2-main (2026-07-04)
+# Version: 8.4.0 | Benchmark: sigmap-v8.4-main (2026-07-04)
 # Source: auto-generated from package.json, version.json, benchmarks/latest.json, src/mcp/tools.js, src/config/defaults.js
 # Regenerate: npm run generate:llms   |   Validate: npm run validate:llms
 
@@ -23,7 +23,7 @@ Cursor, GitHub Copilot, Aider, Windsurf, local LLMs, and MCP.
 - No blast-radius awareness before editing a hub file — `--impact` shows every file a change touches.
 - Pasted stack traces, CI logs, and JSON bloat the prompt — `squeeze` minimizes them and enriches the top frame from the symbol index.
 
-## Core metrics (benchmark: sigmap-v8.2-main, 2026-07-04)
+## Core metrics (benchmark: sigmap-v8.4-main, 2026-07-04)
 
 - hit@5 retrieval: 86.7% vs 13.6% random baseline (6.4× lift)
 - Token reduction: 97.0% average across benchmark repos

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sigmap",
-  "version": "8.2.0",
+  "version": "8.4.0",
   "description": "97% token reduction for AI coding. Extracts function & class signatures with TF-IDF ranking to feed only the right files to Claude, Cursor, Copilot, Aider, Windsurf, local LLMs & MCP. Zero dependencies, runs offline via npx.",
   "main": "packages/core/index.js",
   "exports": {

package/packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sigmap-cli",
-  "version": "8.2.0",
+  "version": "8.4.0",
   "description": "SigMap CLI wrapper — thin adapter for programmatic CLI invocation",
   "main": "index.js",
   "keywords": [

package/packages/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sigmap-core",
-  "version": "8.2.0",
+  "version": "8.4.0",
   "description": "SigMap core library — zero-dependency code signature extraction, retrieval, and security scanning",
   "main": "index.js",
   "keywords": [

package/src/mcp/server.js

@@ -18,7 +18,7 @@ const { readContext, searchSignatures, getMap, createCheckpoint, getRouting, exp
 
 const SERVER_INFO = {
   name: 'sigmap',
-  version: '8.2.0',
+  version: '8.4.0',
   description: 'SigMap MCP server — code signatures on demand',
 };
 

package/src/review/pr-evidence.js

@@ -0,0 +1,139 @@
+'use strict';
+
+/**
+ * PR Evidence Report (v9.0 G3).
+ *
+ * A single, branded, deterministic Markdown artifact for code review: for each
+ * changed file it folds together the signature context, blast radius (direct /
+ * transitive importers, impacted tests + routes), cross-language related tests,
+ * a risk label, and the `review-pr` findings (scope drift, god-node edits,
+ * missing tests, security-sensitive files). Posted as a PR comment, it answers
+ * "what changed, what it touches, and what to test" — without an LLM.
+ *
+ * Built entirely from shipped zero-dep modules (reviewPr, graph/impact,
+ * evidence/pack, extractors/dispatch). Carries NO wall-clock timestamp, so the
+ * report is byte-stable given a fixed tree — diff-friendly as a comment.
+ */
+
+const fs = require('fs');
+const path = require('path');
+const { reviewPr } = require('./review-pr');
+
+/**
+ * Build the structured PR evidence for a changed-file list.
+ * @param {Array<{path:string,status?:string}>|string[]} changedFiles
+ * @param {string} cwd
+ * @param {object} [opts]
+ * @param {number} [opts.depth=2]   blast-radius BFS depth
+ * @param {string} [opts.scope]     label for the diff scope (e.g. "vs main")
+ * @returns {{ scope:string, files:object[], review:object }}
+ */
+function buildPrEvidence(changedFiles, cwd, opts = {}) {
+  const files = (changedFiles || []).map((f) =>
+    typeof f === 'string' ? { path: f, status: 'M' } : { path: f.path, status: f.status || 'M' });
+
+  const review = reviewPr(files, cwd, opts);
+
+  let riskLabelFor = () => 'source';
+  let findRelatedTests = () => [];
+  try { ({ riskLabelFor, findRelatedTests } = require('../evidence/pack')); } catch (_) { /* defaults */ }
+  const { extractFile, langFor } = require('../extractors/dispatch');
+
+  let allFiles = [];
+  try { const { buildSigIndex } = require('../retrieval/ranker'); allFiles = [...buildSigIndex(cwd).keys()]; } catch (_) { /* no index */ }
+
+  const depth = Number.isFinite(opts.depth) ? opts.depth : 2;
+  const srcPaths = files.filter((f) => f.status !== 'D' && langFor(f.path)).map((f) => f.path);
+  let impactByFile = new Map();
+  try {
+    const { analyzeImpact } = require('../graph/impact');
+    impactByFile = new Map(analyzeImpact(srcPaths, cwd, { depth }).map((r) => [r.file, r.impact]));
+  } catch (_) { /* graph optional */ }
+
+  const fileReports = files.map((f) => {
+    const deleted = f.status === 'D';
+    let signatures = [];
+    if (!deleted && langFor(f.path)) {
+      try { signatures = extractFile(f.path, fs.readFileSync(path.resolve(cwd, f.path), 'utf8')); } catch (_) { /* unreadable */ }
+    }
+    const impact = impactByFile.get(f.path) || null;
+    return {
+      path: f.path,
+      status: f.status,
+      riskLabel: riskLabelFor(f.path),
+      signatures,
+      blast: impact ? {
+        total: impact.totalImpact,
+        direct: impact.direct || [],
+        transitive: (impact.transitive || []).length,
+        tests: impact.tests || [],
+        routes: impact.routes || [],
+      } : null,
+      relatedTests: deleted ? [] : findRelatedTests(f.path, allFiles),
+    };
+  });
+
+  return { scope: opts.scope || 'diff', files: fileReports, review };
+}
+
+const STATUS_LABEL = { M: 'modified', A: 'added', D: 'deleted', R: 'renamed', C: 'copied' };
+
+/** Render the branded, deterministic "PR Evidence Report" Markdown. */
+function formatPrEvidenceMarkdown(evidence, opts = {}) {
+  const L = [];
+  const s = evidence.review.summary;
+  const maxSigs = Number.isFinite(opts.maxSignatures) ? opts.maxSignatures : 30;
+
+  L.push('## 🔍 PR Evidence Report');
+  L.push('');
+  L.push(
+    `**${s.filesChanged} file(s) changed** — ${s.sourceChanged} source, ${s.testsChanged} test · ` +
+    (s.ok ? '✅ no review findings' : `⚠️ ${s.findings} finding(s)`) +
+    ` · scope: ${evidence.scope}`
+  );
+  L.push('');
+
+  if (!s.ok) {
+    L.push('### Review findings');
+    for (const f of evidence.review.findings) {
+      if (f.type === 'missing-tests') L.push(`- ⚠️ **missing tests** — \`${f.file}\` changed with no matching test`);
+      else if (f.type === 'security-file') L.push(`- ⚠️ **security-sensitive file** — \`${f.file}\``);
+      else if (f.type === 'god-node') L.push(`- ⚠️ **god node** — \`${f.file}\` → ${f.count} dependents (high blast radius)`);
+      else if (f.type === 'scope-drift') L.push(`- ⚠️ **scope drift** — ${f.count} top-level dirs touched (${f.dirs.join(', ')})`);
+    }
+    L.push('');
+  }
+
+  L.push('### Changed files');
+  for (const f of evidence.files) {
+    const st = STATUS_LABEL[f.status] || f.status;
+    L.push(`#### \`${f.path}\`  _(${st} · risk: ${f.riskLabel})_`);
+    if (f.status === 'D') { L.push('_deleted_', ''); continue; }
+
+    if (f.blast) {
+      L.push(
+        `**Blast radius:** ${f.blast.total} file(s) impacted — ${f.blast.direct.length} direct, ${f.blast.transitive} transitive` +
+        (f.blast.tests.length ? `, ${f.blast.tests.length} test(s)` : '') +
+        (f.blast.routes.length ? `, ${f.blast.routes.length} route(s)` : '')
+      );
+      if (f.blast.tests.length) L.push(`Tests to run: ${f.blast.tests.slice(0, 8).map((t) => '`' + t + '`').join(', ')}`);
+    } else {
+      L.push('**Blast radius:** _(not in dependency graph — new or leaf file)_');
+    }
+    if (f.relatedTests.length) L.push(`Related tests: ${f.relatedTests.slice(0, 8).map((t) => '`' + t + '`').join(', ')}`);
+
+    if (f.signatures.length) {
+      L.push('```');
+      for (const sig of f.signatures.slice(0, maxSigs)) L.push(sig);
+      if (f.signatures.length > maxSigs) L.push(`… +${f.signatures.length - maxSigs} more`);
+      L.push('```');
+    }
+    L.push('');
+  }
+
+  L.push('---');
+  L.push('_Deterministic PR Evidence Report — generated by [SigMap](https://sigmap.io). No LLM; byte-stable given a fixed tree._');
+  return L.join('\n');
+}
+
+module.exports = { buildPrEvidence, formatPrEvidenceMarkdown };

package/src/verify/lib-index.js

@@ -8,12 +8,17 @@
  * in `node_modules` and verify AI suggestions against repo + private +
  * installed-lib symbols. This module builds the installed-lib half.
  *
- * For each **direct** dependency declared in `package.json`, it locates the
- * package under `node_modules/<dep>`, reads its version (D8 version pinning),
- * and extracts the exported symbol names from its TypeScript declaration entry
- * (`types`/`typings`, else `index.d.ts`). Pure, zero-dependency, deterministic:
- * byte-stable given a fixed installed tree. Bounded (per-file read cap + dep
- * cap) and cached via `src/cache/sig-cache.js` so repeat builds are near-free.
+ * Two ecosystems, one index:
+ *   - **JS/TS** — each **direct** dependency in `package.json` resolved under
+ *     `node_modules/<dep>`; exports read from its TypeScript declaration entry
+ *     (`types`/`typings`, else `index.d.ts`).
+ *   - **Python** — each direct dependency in `requirements.txt`/`pyproject.toml`
+ *     resolved in the project's venv `site-packages`; exports read from the
+ *     package's `__init__.py`/`.pyi`. No Python runtime is spawned (North-Star #1).
+ *
+ * Pure, zero-dependency, deterministic: byte-stable given a fixed installed
+ * tree. Bounded (per-file read cap + dep cap) and cached via
+ * `src/cache/sig-cache.js` so repeat builds are near-free.
  */
 
 const fs = require('fs');
@@ -23,6 +28,7 @@ const { loadCache, saveCache, getChangedFiles, updateCacheEntries } = require('.
 const MAX_DTS_BYTES = 512 * 1024; // per-file read cap
 const MAX_DEPS = 1000;            // dep count cap
 const DEP_KEYS = ['dependencies', 'devDependencies', 'peerDependencies', 'optionalDependencies'];
+const VENV_DIRS = ['.venv', 'venv', 'env', '.env'];
 
 /**
  * Extract exported symbol names from a `.d.ts` declaration file. Deterministic,
@@ -98,6 +104,130 @@ function resolveEntry(cwd, dep) {
   return { version, dtsPath: null }; // installed but untyped
 }
 
+// ── Python ──────────────────────────────────────────────────────────────────
+
+/**
+ * Extract exported symbol names from a Python module's `__init__.py`/`.pyi`.
+ * Deterministic, regex-based, top-level only: `__all__`, `def`/`class`, public
+ * module-level assignments, and `from … import …` re-exports (a package's
+ * public API is largely re-exports). Private names (leading `_`) are skipped
+ * unless listed in `__all__`.
+ * @param {string} src
+ * @returns {string[]} sorted unique exported names
+ */
+function extractPyExports(src) {
+  const names = new Set();
+  if (!src) return [];
+
+  // __all__ = [ 'a', 'b', ... ]  (authoritative when present; keeps privates)
+  const allMatch = src.match(/^__all__\s*[:+]?=\s*[\[(]([\s\S]*?)[\])]/m);
+  if (allMatch) {
+    for (const m of allMatch[1].matchAll(/['"]([A-Za-z_]\w*)['"]/g)) names.add(m[1]);
+  }
+
+  // top-level def / class (column 0)
+  for (const m of src.matchAll(/^(?:async\s+)?def\s+([A-Za-z_]\w*)/gm)) if (!m[1].startsWith('_')) names.add(m[1]);
+  for (const m of src.matchAll(/^class\s+([A-Za-z_]\w*)/gm)) if (!m[1].startsWith('_')) names.add(m[1]);
+
+  // top-level public assignments: NAME = …  /  NAME: type = …  (not ==, +=, etc.)
+  for (const m of src.matchAll(/^([A-Za-z_]\w*)\s*(?::[^=\n]+)?=(?!=)/gm)) {
+    if (!m[1].startsWith('_')) names.add(m[1]);
+  }
+
+  // re-exports: from .mod import Name, Other as Alias
+  for (const m of src.matchAll(/^from\s+[^\n]+?\s+import\s+([^\n#]+)/gm)) {
+    for (const part of m[1].split(',')) {
+      const name = part.trim().replace(/[()]/g, '').split(/\s+as\s+/).pop().trim();
+      if (/^[A-Za-z_]\w*$/.test(name) && !name.startsWith('_')) names.add(name);
+    }
+  }
+
+  return [...names].sort();
+}
+
+/** Read direct Python dependency names from requirements.txt + pyproject.toml. */
+function pythonDirectDeps(cwd) {
+  const names = new Set();
+  try {
+    const req = fs.readFileSync(path.join(cwd, 'requirements.txt'), 'utf8');
+    for (const line of req.split('\n')) {
+      const t = line.trim();
+      if (!t || t.startsWith('#') || t.startsWith('-')) continue;
+      const m = t.match(/^([A-Za-z0-9][A-Za-z0-9._-]*)/);
+      if (m) names.add(m[1]);
+    }
+  } catch (_) { /* none */ }
+  try {
+    const py = fs.readFileSync(path.join(cwd, 'pyproject.toml'), 'utf8');
+    // PEP 621: [project] dependencies = ["foo>=1", "bar"]
+    const projDeps = py.match(/^\s*dependencies\s*=\s*\[([\s\S]*?)\]/m);
+    if (projDeps) for (const m of projDeps[1].matchAll(/['"]([A-Za-z0-9][A-Za-z0-9._-]*)/g)) names.add(m[1]);
+    // Poetry: [tool.poetry.dependencies]\n foo = "^1"
+    const poetry = py.match(/\[tool\.poetry\.dependencies\]([\s\S]*?)(?:\n\[|$)/);
+    if (poetry) for (const m of poetry[1].matchAll(/^([A-Za-z0-9][A-Za-z0-9._-]*)\s*=/gm)) {
+      if (m[1] !== 'python') names.add(m[1]);
+    }
+  } catch (_) { /* none */ }
+  return [...names].sort();
+}
+
+/** Locate the project's venv `site-packages` directories (no Python runtime). */
+function findSitePackages(cwd) {
+  const out = [];
+  for (const v of VENV_DIRS) {
+    const base = path.join(cwd, v);
+    const libDir = path.join(base, 'lib'); // POSIX: <venv>/lib/pythonX.Y/site-packages
+    let pyDirs = [];
+    try { pyDirs = fs.readdirSync(libDir).filter((d) => /^python\d/.test(d)).sort(); } catch (_) { /* none */ }
+    for (const py of pyDirs) {
+      const sp = path.join(libDir, py, 'site-packages');
+      try { if (fs.statSync(sp).isDirectory()) out.push(sp); } catch (_) { /* next */ }
+    }
+    const winSp = path.join(base, 'Lib', 'site-packages'); // Windows
+    try { if (fs.statSync(winSp).isDirectory()) out.push(winSp); } catch (_) { /* next */ }
+  }
+  return out;
+}
+
+/** PEP 503 name normalization (case-insensitive, `-`/`_`/`.` collapsed). */
+function normalizePy(name) {
+  return String(name).toLowerCase().replace(/[-_.]+/g, '-');
+}
+
+/** Find an installed distribution's version from its `*.dist-info`/`*.egg-info`. */
+function findPyVersion(sitePkgsDir, dep) {
+  const norm = normalizePy(dep);
+  let entries;
+  try { entries = fs.readdirSync(sitePkgsDir); } catch (_) { return null; }
+  for (const e of entries.sort()) {
+    const m = e.match(/^(.+?)-(\d[^-]*)\.(?:dist-info|egg-info)$/);
+    if (m && normalizePy(m[1]) === norm) return m[2];
+  }
+  return null;
+}
+
+/**
+ * Resolve a Python dependency to its installed module entry file + version.
+ * @returns {{ version: string|null, sourcePath: string|null }|null} null if not installed
+ */
+function resolvePyEntry(sitePkgsDirs, dep) {
+  const candidates = [...new Set([dep, dep.replace(/-/g, '_'), dep.toLowerCase(), dep.toLowerCase().replace(/-/g, '_')])];
+  for (const sp of sitePkgsDirs) {
+    const version = findPyVersion(sp, dep);
+    for (const cand of candidates) {
+      for (const entry of ['__init__.pyi', '__init__.py']) { // package
+        const p = path.join(sp, cand, entry);
+        try { if (fs.statSync(p).isFile()) return { version, sourcePath: p }; } catch (_) { /* next */ }
+      }
+      for (const ext of ['.pyi', '.py']) { // single-module
+        const p = path.join(sp, cand + ext);
+        try { if (fs.statSync(p).isFile()) return { version, sourcePath: p }; } catch (_) { /* next */ }
+      }
+    }
+  }
+  return null;
+}
+
 /**
  * Build the installed-library signature index for `cwd`.
  *
@@ -110,17 +240,24 @@ function resolveEntry(cwd, dep) {
 function buildLibraryIndex(cwd, opts = {}) {
   const version = opts.version || '0';
   const useCache = opts.cache !== false;
-  const deps = directDeps(cwd).slice(0, MAX_DEPS);
 
-  const entries = [];
-  for (const dep of deps) {
+  // Collect entries from both ecosystems; each carries its extractor kind.
+  const entries = []; // { name, version, sourcePath, kind: 'dts'|'py' }
+  for (const dep of directDeps(cwd).slice(0, MAX_DEPS)) {
     const r = resolveEntry(cwd, dep);
-    if (r) entries.push({ dep, version: r.version, dtsPath: r.dtsPath });
+    if (r) entries.push({ name: dep, version: r.version, sourcePath: r.dtsPath, kind: 'dts' });
+  }
+  const sitePkgs = findSitePackages(cwd);
+  if (sitePkgs.length) {
+    for (const dep of pythonDirectDeps(cwd).slice(0, MAX_DEPS)) {
+      const r = resolvePyEntry(sitePkgs, dep);
+      if (r) entries.push({ name: dep, version: r.version, sourcePath: r.sourcePath, kind: 'py' });
+    }
   }
 
   const cache = useCache ? loadCache(cwd, version) : new Map();
-  const dtsFiles = entries.filter((e) => e.dtsPath).map((e) => e.dtsPath);
-  const { unchanged } = getChangedFiles(dtsFiles, cache);
+  const files = entries.filter((e) => e.sourcePath).map((e) => e.sourcePath);
+  const { unchanged } = getChangedFiles(files, cache);
   const unchangedSet = new Set(unchanged);
 
   const symbols = new Set();
@@ -129,20 +266,20 @@ function buildLibraryIndex(cwd, opts = {}) {
 
   for (const e of entries) {
     let names;
-    if (!e.dtsPath) {
+    if (!e.sourcePath) {
       names = [];
-    } else if (unchangedSet.has(e.dtsPath) && cache.get(e.dtsPath)) {
-      names = cache.get(e.dtsPath).sigs || [];
+    } else if (unchangedSet.has(e.sourcePath) && cache.get(e.sourcePath)) {
+      names = cache.get(e.sourcePath).sigs || [];
     } else {
       let src = '';
       try {
-        if (fs.statSync(e.dtsPath).size <= MAX_DTS_BYTES) src = fs.readFileSync(e.dtsPath, 'utf8');
+        if (fs.statSync(e.sourcePath).size <= MAX_DTS_BYTES) src = fs.readFileSync(e.sourcePath, 'utf8');
       } catch (_) { /* unreadable → empty */ }
-      names = extractDtsExports(src);
-      fresh.push({ file: e.dtsPath, sigs: names });
+      names = e.kind === 'py' ? extractPyExports(src) : extractDtsExports(src);
+      fresh.push({ file: e.sourcePath, sigs: names });
     }
     for (const n of names) symbols.add(n);
-    libraries.push({ name: e.dep, version: e.version, symbols: names.length, typed: !!e.dtsPath });
+    libraries.push({ name: e.name, version: e.version, symbols: names.length, typed: !!e.sourcePath });
   }
 
   if (useCache && fresh.length) {
@@ -161,4 +298,7 @@ function formatVersionPins(libraries) {
     .map((l) => `${l.name}@${l.version}`);
 }
 
-module.exports = { buildLibraryIndex, extractDtsExports, directDeps, resolveEntry, formatVersionPins };
+module.exports = {
+  buildLibraryIndex, extractDtsExports, directDeps, resolveEntry, formatVersionPins,
+  extractPyExports, pythonDirectDeps, findSitePackages, resolvePyEntry,
+};