diff --git a/.github/workflows/eval-dispatch.yml b/.github/workflows/eval-dispatch.yml
new file mode 100644
index 00000000..c3be2d19
--- /dev/null
+++ b/.github/workflows/eval-dispatch.yml
@@ -0,0 +1,117 @@
+name: PR Eval Dispatch
+
+# `/eval <products> [cases=<ref>]` PR 评论 → 触发内网评测流水线，报告由内网 bot 回贴。
+# 本 workflow 只在默认分支上下文运行，不 checkout、不执行 PR 代码；
+# maintainer 打出 /eval 即代表已审阅 PR 代码（内网 CI 将构建并执行该 head）。
+# 触发通道与凭证全部经 secrets 注入，文件内不出现任何内网信息。
+
+on:
+  issue_comment:
+    types:
+      - created
+
+permissions: {}
+
+concurrency:
+  group: eval-dispatch-${{ github.event.issue.number }}
+  cancel-in-progress: false
+
+jobs:
+  dispatch:
+    name: Dispatch internal evaluation
+    if: >-
+      github.event.issue.pull_request &&
+      startsWith(github.event.comment.body, '/eval') &&
+      contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association)
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      contents: read
+      issues: write
+      pull-requests: read
+    steps:
+      - name: Check out default branch tooling
+        uses: actions/checkout@v4
+
+      - name: Parse /eval command
+        id: parse
+        continue-on-error: true
+        env:
+          COMMENT_BODY: ${{ github.event.comment.body }}
+        run: python3 scripts/ci/eval_comment_parse.py
+
+      - name: Reply usage on parse failure
+        if: steps.parse.outcome == 'failure'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          PARSE_ERROR: ${{ steps.parse.outputs.error }}
+        run: |
+          body="❌ /eval 命令解析失败：${PARSE_ERROR}"
+          jq -n --arg body "$body" '{body: $body}' | curl --fail --silent --show-error \
+            -X POST \
+            -H "Authorization: Bearer ${GH_TOKEN}" \
+            -H "Accept: application/vnd.github+json" \
+            --data @- \
+            "https://api.github.com/repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" > /dev/null
+          exit 1
+
+      - name: Resolve PR head SHA
+        id: pr
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+        run: |
+          head_sha="$(curl --fail --silent --show-error \
+            -H "Authorization: Bearer ${GH_TOKEN}" \
+            -H "Accept: application/vnd.github+json" \
+            "https://api.github.com/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" | jq -r '.head.sha')"
+          if [ -z "$head_sha" ] || [ "$head_sha" = "null" ]; then
+            echo "Failed to resolve PR head SHA." >&2
+            exit 1
+          fi
+          echo "head_sha=${head_sha}" >> "$GITHUB_OUTPUT"
+
+      - name: Trigger internal evaluation pipeline
+        env:
+          EVAL_TRIGGER_TOKEN: ${{ secrets.EVAL_TRIGGER_TOKEN }}
+          EVAL_TRIGGER_URL: ${{ secrets.EVAL_TRIGGER_URL }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
+          PRODUCTS: ${{ steps.parse.outputs.products }}
+          CASES_REF: ${{ steps.parse.outputs.cases_ref }}
+        run: |
+          if [ -z "$EVAL_TRIGGER_TOKEN" ] || [ -z "$EVAL_TRIGGER_URL" ]; then
+            echo "EVAL_TRIGGER_URL / EVAL_TRIGGER_TOKEN not configured; cannot dispatch." >&2
+            exit 1
+          fi
+          curl --fail --silent --show-error \
+            -X POST \
+            -F "token=${EVAL_TRIGGER_TOKEN}" \
+            -F "ref=main" \
+            -F "variables[PR_NUMBER]=${PR_NUMBER}" \
+            -F "variables[PR_HEAD_SHA]=${PR_HEAD_SHA}" \
+            -F "variables[PRODUCTS]=${PRODUCTS}" \
+            -F "variables[CASES_REF]=${CASES_REF}" \
+            "$EVAL_TRIGGER_URL"
+          echo "Internal evaluation dispatched."
+
+      - name: Acknowledge on PR
+        env:
+          GH_TOKEN: ${{ github.token }}
+          PR_NUMBER: ${{ github.event.issue.number }}
+          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
+          PRODUCTS: ${{ steps.parse.outputs.products }}
+          CASES_REF: ${{ steps.parse.outputs.cases_ref }}
+        run: |
+          cases_note=""
+          if [ -n "$CASES_REF" ]; then
+            cases_note="，用例版本 \`${CASES_REF}\`"
+          fi
+          body="🛰️ /eval 已受理：产品集 \`${PRODUCTS}\`${cases_note}，评测对象 \`${PR_HEAD_SHA}\`。内网评测流水线运行结束后将由 bot 回贴报告（首行为基线对比头条）。"
+          jq -n --arg body "$body" '{body: $body}' | curl --fail --silent --show-error \
+            -X POST \
+            -H "Authorization: Bearer ${GH_TOKEN}" \
+            -H "Accept: application/vnd.github+json" \
+            --data @- \
+            "https://api.github.com/repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" > /dev/null
diff --git a/scripts/ci/eval_comment_parse.py b/scripts/ci/eval_comment_parse.py
new file mode 100644
index 00000000..de4433f1
--- /dev/null
+++ b/scripts/ci/eval_comment_parse.py
@@ -0,0 +1,70 @@
+#!/usr/bin/env python3
+"""解析 PR 评论中的 `/eval <products> [cases=<ref>]` 触发命令。
+
+供 .github/workflows/eval-dispatch.yml 调用：评论体经环境变量 COMMENT_BODY
+传入（避免 argv 注入），解析结果以 GitHub Actions output（products / cases_ref）
+写出；格式非法时把单行错误写入 output `error` 并以非零退出。
+
+产品集显式必填；`cases=<ref>` 为用例仓库版本逃生舱（破坏性变更配对验证）。
+"""
+
+import os
+import re
+import sys
+
+PRODUCT_RE = re.compile(r"^[a-z0-9][a-z0-9-]*$")
+REF_RE = re.compile(r"^[A-Za-z0-9._/-]+$")
+USAGE = "用法: /eval <product>[,<product>...] [cases=<ref>]；示例: /eval drive,doc cases=feat/x"
+
+
+def parse(body: str):
+    lines = [line.strip() for line in (body or "").splitlines() if line.strip()]
+    if not lines:
+        raise ValueError(f"评论为空。{USAGE}")
+    tokens = lines[0].split()
+    if tokens[0] != "/eval":
+        raise ValueError(f"首行必须以 /eval 命令开头。{USAGE}")
+    if len(tokens) < 2:
+        raise ValueError(f"产品集显式必填。{USAGE}")
+
+    products = [p for p in tokens[1].split(",") if p]
+    if not products:
+        raise ValueError(f"产品集显式必填。{USAGE}")
+    for product in products:
+        if not PRODUCT_RE.match(product):
+            raise ValueError(f"产品名非法: {product}。{USAGE}")
+
+    cases_ref = ""
+    for extra in tokens[2:]:
+        if extra.startswith("cases="):
+            cases_ref = extra[len("cases="):]
+            if not cases_ref or not REF_RE.match(cases_ref):
+                raise ValueError(f"cases 引用非法: {extra}。{USAGE}")
+        else:
+            raise ValueError(f"未知参数: {extra}。{USAGE}")
+
+    return ",".join(products), cases_ref
+
+
+def _write_outputs(pairs):
+    lines = [f"{key}={value}" for key, value in pairs]
+    output_path = os.environ.get("GITHUB_OUTPUT", "")
+    if output_path:
+        with open(output_path, "a", encoding="utf-8") as f:
+            f.write("\n".join(lines) + "\n")
+    print("\n".join(lines))
+
+
+def main() -> int:
+    try:
+        products, cases_ref = parse(os.environ.get("COMMENT_BODY", ""))
+    except ValueError as exc:
+        _write_outputs([("error", str(exc))])
+        print(str(exc), file=sys.stderr)
+        return 1
+    _write_outputs([("products", products), ("cases_ref", cases_ref)])
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/scripts/ci/eval_comment_parse_test.py b/scripts/ci/eval_comment_parse_test.py
new file mode 100644
index 00000000..bbf76178
--- /dev/null
+++ b/scripts/ci/eval_comment_parse_test.py
@@ -0,0 +1,50 @@
+import pytest
+
+from eval_comment_parse import parse
+
+
+def test_single_product():
+    assert parse("/eval drive") == ("drive", "")
+
+
+def test_multiple_products():
+    assert parse("/eval drive,doc,edu-app") == ("drive,doc,edu-app", "")
+
+
+def test_cases_ref_escape_hatch():
+    assert parse("/eval drive cases=feat/drive-latest") == ("drive", "feat/drive-latest")
+
+
+def test_extra_lines_after_command_are_ignored():
+    body = "/eval drive\n\n顺便说明：这个 PR 只动了 drive 的 --latest。"
+    assert parse(body) == ("drive", "")
+
+
+def test_missing_products_rejected():
+    with pytest.raises(ValueError, match="产品集显式必填"):
+        parse("/eval")
+
+
+def test_similar_command_prefix_rejected():
+    with pytest.raises(ValueError, match="/eval 命令开头"):
+        parse("/evaluate drive")
+
+
+def test_illegal_product_name_rejected():
+    with pytest.raises(ValueError, match="产品名非法"):
+        parse("/eval drive;rm")
+
+
+def test_unknown_extra_token_rejected():
+    with pytest.raises(ValueError, match="未知参数"):
+        parse("/eval drive --force")
+
+
+def test_illegal_cases_ref_rejected():
+    with pytest.raises(ValueError, match="cases 引用非法"):
+        parse("/eval drive cases=$(whoami)")
+
+
+def test_empty_comment_rejected():
+    with pytest.raises(ValueError, match="评论为空"):
+        parse("   \n  ")