diff --git a/.github/workflows/pr-risk-scan-comment.yml b/.github/workflows/pr-risk-scan-comment.yml
new file mode 100644
index 000000000..5870594e2
--- /dev/null
+++ b/.github/workflows/pr-risk-scan-comment.yml
@@ -0,0 +1,96 @@
+name: PR Risk Scan — Comment
+
+on:
+  workflow_run:
+    workflows: ["PR Risk Scan — Gate"]
+    types: [completed]
+
+permissions:
+  issues: write
+  pull-requests: write
+  actions: read
+
+jobs:
+  comment:
+    runs-on: ubuntu-latest
+    if: github.event.workflow_run.event == 'pull_request'
+    steps:
+      - name: Download scan artifact
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: pr-risk-scan-results
+          run-id: ${{ github.event.workflow_run.id }}
+          github-token: ${{ github.token }}
+
+      - name: Upsert PR comment
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        with:
+          script: |
+            const fs = require('fs');
+            const marker = '<!-- pr-risk-scan-results -->';
+            const reportPath = 'report.md';
+            const prNumberPath = 'pr-number.txt';
+
+            if (!fs.existsSync(reportPath)) {
+              core.setFailed('Risk scan report.md artifact was not found.');
+              return;
+            }
+
+            let body = fs.readFileSync(reportPath, 'utf8');
+
+            // Treat artifact content as untrusted (the gate workflow runs on PR code).
+            // Prevent spam/notification abuse and avoid API failures on oversized bodies.
+            body = body.replace(/@/g, '@\u200b');
+            const maxLength = 65000;
+            if (body.length > maxLength) {
+              body = `${body.slice(0, maxLength)}\n\n_...(truncated)..._`;
+            }
+            if (!body.includes(marker)) {
+              body = `${marker}\n${body}`;
+            }
+            let prNumber = null;
+            if (fs.existsSync(prNumberPath)) {
+              const parsed = parseInt(fs.readFileSync(prNumberPath, 'utf8').trim(), 10);
+              if (!Number.isNaN(parsed)) {
+                prNumber = parsed;
+              }
+            }
+
+            if (!prNumber) {
+              const fallback = context.payload.workflow_run.pull_requests?.[0]?.number;
+              if (fallback) {
+                prNumber = fallback;
+              }
+            }
+
+            if (!prNumber) {
+              core.setFailed('Could not determine PR number for comment upsert.');
+              return;
+            }
+
+            const { data: comments } = await github.rest.issues.listComments({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: prNumber,
+              per_page: 100,
+            });
+
+            const existing = comments.find((comment) => comment.body.includes(marker));
+
+            if (existing) {
+              await github.rest.issues.updateComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                comment_id: existing.id,
+                body,
+              });
+              console.log(`Updated existing risk scan comment ${existing.id}`);
+            } else {
+              await github.rest.issues.createComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                issue_number: prNumber,
+                body,
+              });
+              console.log('Created new risk scan comment');
+            }
diff --git a/.github/workflows/pr-risk-scan.yml b/.github/workflows/pr-risk-scan.yml
new file mode 100644
index 000000000..2dc22b412
--- /dev/null
+++ b/.github/workflows/pr-risk-scan.yml
@@ -0,0 +1,51 @@
+name: PR Risk Scan — Gate
+
+on:
+  pull_request:
+    branches: [staged]
+    types: [opened, synchronize, reopened]
+    paths:
+      - "skills/**"
+      - "agents/**"
+      - "workflows/**"
+      - "plugins/**"
+      - "hooks/**"
+      - "instructions/**"
+
+permissions:
+  contents: read
+  actions: read
+
+jobs:
+  scan:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
+        with:
+          fetch-depth: 0
+
+      - name: Collect changed files
+        run: |
+          git diff --name-only --diff-filter=ACMR "origin/${{ github.base_ref }}...HEAD" > changed-files.txt
+          echo "Changed files:"
+          cat changed-files.txt || true
+
+      - name: Run PR risk scanner
+        run: |
+          mkdir -p pr-risk-results
+          node ./eng/pr-risk-scan.mjs \
+            --files changed-files.txt \
+            --output-json pr-risk-results/results.json \
+            --output-md pr-risk-results/report.md
+
+      - name: Save metadata
+        run: |
+          echo "${{ github.event.pull_request.number }}" > pr-risk-results/pr-number.txt
+
+      - name: Upload scan artifact
+        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7
+        with:
+          name: pr-risk-scan-results
+          path: pr-risk-results/
+          retention-days: 1
diff --git a/eng/pr-risk-scan.mjs b/eng/pr-risk-scan.mjs
new file mode 100644
index 000000000..d3ca7ec03
--- /dev/null
+++ b/eng/pr-risk-scan.mjs
@@ -0,0 +1,402 @@
+#!/usr/bin/env node
+
+import fs from "fs";
+import path from "path";
+
+const SCRIPT_EXTENSIONS = new Set([
+  ".sh",
+  ".bash",
+  ".ps1",
+  ".py",
+  ".js",
+  ".mjs",
+  ".ts",
+]);
+
+function isLikelyAbsolutePath(value) {
+  if (!value) {
+    return false;
+  }
+
+  // POSIX absolute (/foo), UNC (//server/share), Windows drive paths (C:/foo).
+  return (
+    value.startsWith("/") ||
+    value.startsWith("//") ||
+    /^[A-Za-z]:\//.test(value)
+  );
+}
+
+function isPathWithinRoot(rootPath, targetPath) {
+  const relative = path.relative(rootPath, targetPath);
+  return (
+    relative === "" ||
+    (!relative.startsWith("..") && !path.isAbsolute(relative))
+  );
+}
+
+function hasUnpinnedVersionIndicator(line) {
+  const trimmed = line.trim();
+
+  if (!trimmed) {
+    return false;
+  }
+
+  // Command contexts where floating versions are risky.
+  if (
+    /\b(npm|pnpm|yarn|bun|npx|uvx|pip|pipx)\b[^\n]*(?:@latest\b|\blatest\b)/i.test(
+      trimmed
+    )
+  ) {
+    return true;
+  }
+
+  // package.json/yaml style dependency entries with floating ranges.
+  if (
+    /["'][^"']+["']\s*:\s*["'](\^|~|\*|latest\b)[^"']*["']/i.test(trimmed)
+  ) {
+    return true;
+  }
+
+  // pyproject/requirements style entries with broad lower-bound only specs.
+  if (
+    /\b[A-Za-z0-9_.-]+\s*(>=|>|~=)\s*\d+(?:\.\d+){0,2}\b(?!\s*,\s*<)/.test(
+      trimmed
+    )
+  ) {
+    return true;
+  }
+
+  return false;
+}
+
+const severityLevels = {
+  high: "high",
+  medium: "medium",
+  info: "info",
+};
+
+const LINE_RULES = [
+  {
+    rule_id: "guardrail-bypass-language",
+    severity: severityLevels.high,
+    regex:
+      /\b(ignore (all|any|previous) (guardrails?|rules?|instructions?)|bypass (the )?(guardrails?|safety|policy)|disable (safety|guardrails?)|do not ask (for )?(confirmation|consent)|without prompting (the )?user)\b/i,
+    reason: "Language suggests bypassing policy or confirmation controls.",
+    suggested_fix:
+      "Require explicit policy adherence and user-confirmation steps for risky actions.",
+  },
+  {
+    rule_id: "remote-shell-execution",
+    severity: severityLevels.high,
+    regex: /\b(curl|wget)\b[^\n|]*\|\s*(sh|bash|zsh|pwsh|powershell)\b/i,
+    reason: "Piping remote content directly to a shell is high-risk.",
+    suggested_fix:
+      "Download, verify integrity/signature, and run from a reviewed local file.",
+  },
+  {
+    rule_id: "autoyes-package-exec",
+    severity: severityLevels.high,
+    regex:
+      /\b(npx|npm\s+exec|pnpm\s+dlx|uvx|pipx\s+run)\b[^\n]*\s(-y|--yes)\b/i,
+    reason:
+      "Auto-yes execution can bypass human review of package/runtime prompts.",
+    suggested_fix:
+      "Remove automatic consent flags and require explicit reviewer-approved invocation.",
+  },
+  {
+    rule_id: "package-exec-command",
+    severity: severityLevels.medium,
+    regex: /\b(npx|npm\s+exec|pnpm\s+dlx|uvx|pipx\s+run|uv\s+tool\s+run)\b/i,
+    reason: "Dynamic package/runtime execution introduces supply-chain risk.",
+    suggested_fix:
+      "Pin exact versions and document manual confirmation controls.",
+  },
+  {
+    rule_id: "unpinned-version-indicator",
+    severity: severityLevels.medium,
+    reason: "Unpinned dependencies can change behavior between runs.",
+    suggested_fix: "Use exact immutable versions or commit hashes.",
+    matcher: (line) => hasUnpinnedVersionIndicator(line),
+  },
+];
+
+function parseArgs(argv) {
+  const args = {};
+  for (let i = 0; i < argv.length; i += 1) {
+    const key = argv[i];
+    if (!key.startsWith("--")) {
+      continue;
+    }
+
+    args[key.slice(2)] = argv[i + 1];
+    i += 1;
+  }
+  return args;
+}
+
+function ensureParentDir(filePath) {
+  const directory = path.dirname(filePath);
+  fs.mkdirSync(directory, { recursive: true });
+}
+
+function normalizeRelativePath(value) {
+  const cleaned = String(value || "")
+    .trim()
+    .replace(/\\/g, "/")
+    .replace(/^\.\/+/, "");
+  if (!cleaned) {
+    return "";
+  }
+
+  if (/(^|\/)\.\.(\/|$)/.test(cleaned)) {
+    throw new Error(`Unsafe relative path in changed files list: ${value}`);
+  }
+
+  if (isLikelyAbsolutePath(cleaned)) {
+    throw new Error(`Absolute paths are not allowed in changed files list: ${value}`);
+  }
+
+  return cleaned;
+}
+
+function isPotentialText(contentBuffer) {
+  const nullByte = contentBuffer.includes(0x00);
+  return !nullByte;
+}
+
+function addFinding(findings, finding) {
+  findings.push({
+    rule_id: finding.rule_id,
+    severity: finding.severity,
+    file: finding.file,
+    line: finding.line,
+    match: finding.match.slice(0, 180),
+    reason: finding.reason,
+    suggested_fix: finding.suggested_fix,
+  });
+}
+
+function scanLineRules(filePath, content, findings) {
+  const lines = content.split(/\r?\n/);
+  for (let index = 0; index < lines.length; index += 1) {
+    const line = lines[index];
+    for (const rule of LINE_RULES) {
+      if (typeof rule.shouldApply === "function" && !rule.shouldApply(line)) {
+        continue;
+      }
+
+      const matchedByRegex = rule.regex ? rule.regex.test(line) : false;
+      const matchedByFunction =
+        typeof rule.matcher === "function" ? rule.matcher(line) : false;
+      if (!matchedByRegex && !matchedByFunction) {
+        continue;
+      }
+
+      addFinding(findings, {
+        rule_id: rule.rule_id,
+        severity: rule.severity,
+        file: filePath,
+        line: index + 1,
+        match: line.trim(),
+        reason: rule.reason,
+        suggested_fix: rule.suggested_fix,
+      });
+    }
+  }
+}
+
+function scanSkillScriptPath(filePath, findings) {
+  const normalized = filePath.replace(/\\/g, "/");
+  if (!normalized.startsWith("skills/")) {
+    return;
+  }
+
+  const extension = path.extname(normalized).toLowerCase();
+  if (!SCRIPT_EXTENSIONS.has(extension)) {
+    return;
+  }
+
+  addFinding(findings, {
+    rule_id: "skill-script-touched",
+    severity: severityLevels.info,
+    file: normalized,
+    line: 1,
+    match: normalized,
+    reason:
+      "Script asset under a skill may require external runtime/dependencies.",
+    suggested_fix:
+      "Document dependencies, pin versions, and avoid implicit network installs.",
+  });
+}
+
+function severityCounts(findings) {
+  return findings.reduce(
+    (acc, finding) => {
+      acc[finding.severity] = (acc[finding.severity] || 0) + 1;
+      return acc;
+    },
+    { high: 0, medium: 0, info: 0 }
+  );
+}
+
+function toMarkdownReport(findings, scannedFiles, skippedFiles) {
+  const marker = "<!-- pr-risk-scan-results -->";
+  const counts = severityCounts(findings);
+  const summary = [
+    marker,
+    "## 🔒 PR Risk Scan Results",
+    "",
+    `Scanned **${scannedFiles.length}** changed file(s).`,
+    "",
+    "| Severity | Count |",
+    "|---|---:|",
+    `| 🔴 High | ${counts.high} |`,
+    `| 🟠 Medium | ${counts.medium} |`,
+    `| ℹ️ Info | ${counts.info} |`,
+    "",
+  ];
+
+  if (findings.length === 0) {
+    summary.push(
+      "✅ No matching risk patterns were detected in changed files."
+    );
+  } else {
+    summary.push("| Severity | Rule | File | Line | Match |");
+    summary.push("|---|---|---|---:|---|");
+    for (const finding of findings.slice(0, 100)) {
+      const severity =
+        finding.severity === severityLevels.high
+          ? "🔴"
+          : finding.severity === severityLevels.medium
+          ? "🟠"
+          : "ℹ️";
+      const matchText = finding.match
+        .replace(/\\/g, "\\\\")
+        .replace(/</g, "&lt;")
+        .replace(/>/g, "&gt;")
+        .replace(/\|/g, "\\|")
+        .replace(/@/g, "@\u200b");
+      const backtickRuns = matchText.match(/`+/g);
+      const fenceLength = backtickRuns
+        ? Math.max(...backtickRuns.map((run) => run.length)) + 1
+        : 1;
+      const fence = "`".repeat(fenceLength);
+      const match = `${fence}${matchText}${fence}`;
+      summary.push(
+        `| ${severity} | \`${finding.rule_id}\` | \`${finding.file}\` | ${finding.line} | ${match} |`
+      );
+    }
+
+    if (findings.length > 100) {
+      summary.push(
+        "",
+        `_${findings.length - 100} additional finding(s) omitted from table._`
+      );
+    }
+  }
+
+  if (skippedFiles.length > 0) {
+    summary.push(
+      "",
+      "<details>",
+      "<summary>Skipped non-text or missing files</summary>",
+      ""
+    );
+    summary.push(skippedFiles.map((filePath) => `- ${filePath}`).join("\n"));
+    summary.push("", "</details>");
+  }
+
+  summary.push(
+    "",
+    "> This is an automated soft-gate report. Findings indicate review targets and do not block merge by themselves."
+  );
+
+  return `${summary.join("\n")}\n`;
+}
+
+function main() {
+  const args = parseArgs(process.argv.slice(2));
+  if (!args.files || !args["output-json"] || !args["output-md"]) {
+    throw new Error(
+      "Usage: node ./eng/pr-risk-scan.mjs --files <changed-files.txt> --output-json <results.json> --output-md <report.md>"
+    );
+  }
+
+  const changedFilesPath = path.resolve(args.files);
+  const outputJsonPath = path.resolve(args["output-json"]);
+  const outputMarkdownPath = path.resolve(args["output-md"]);
+  const repoRootPath = process.cwd();
+
+  const changedFiles = fs
+    .readFileSync(changedFilesPath, "utf8")
+    .split(/\r?\n/)
+    .map(normalizeRelativePath)
+    .filter(Boolean);
+
+  const findings = [];
+  const scannedFiles = [];
+  const skippedFiles = [];
+
+  for (const relativePath of changedFiles) {
+    const absolutePath = path.resolve(repoRootPath, relativePath);
+    if (!isPathWithinRoot(repoRootPath, absolutePath)) {
+      throw new Error(`Path escapes repository root: ${relativePath}`);
+    }
+
+    scanSkillScriptPath(relativePath, findings);
+
+    if (!fs.existsSync(absolutePath)) {
+      skippedFiles.push(relativePath);
+      continue;
+    }
+
+    const stat = fs.lstatSync(absolutePath);
+    if (stat.isSymbolicLink()) {
+      skippedFiles.push(`${relativePath} (skipped: symbolic link)`);
+      continue;
+    }
+    if (!stat.isFile()) {
+      skippedFiles.push(relativePath);
+      continue;
+    }
+
+    if (stat.size > 1024 * 1024) {
+      skippedFiles.push(`${relativePath} (skipped: file too large)`);
+      continue;
+    }
+
+    const contentBuffer = fs.readFileSync(absolutePath);
+    if (!isPotentialText(contentBuffer)) {
+      skippedFiles.push(relativePath);
+      continue;
+    }
+
+    const content = contentBuffer.toString("utf8");
+    scanLineRules(relativePath, content, findings);
+    scannedFiles.push(relativePath);
+  }
+
+  const results = {
+    generated_at: new Date().toISOString(),
+    scanned_files: scannedFiles,
+    skipped_files: skippedFiles,
+    finding_count: findings.length,
+    severity_counts: severityCounts(findings),
+    findings,
+  };
+
+  ensureParentDir(outputJsonPath);
+  ensureParentDir(outputMarkdownPath);
+  fs.writeFileSync(outputJsonPath, `${JSON.stringify(results, null, 2)}\n`);
+  fs.writeFileSync(
+    outputMarkdownPath,
+    toMarkdownReport(findings, scannedFiles, skippedFiles)
+  );
+}
+
+try {
+  main();
+} catch (error) {
+  console.error(error.message);
+  process.exit(1);
+}