fix(triage): resolve stage comment ids by marker at patch time, harden model injection

.github/workflows/qwen-triage.yml

@@ -339,9 +339,22 @@ jobs:
           OPENAI_MODEL: '${{ vars.QWEN_TRIAGE_MODEL || vars.QWEN_PR_REVIEW_MODEL }}'
         run: |-
           set -euo pipefail
-          if [ -n "${OPENAI_MODEL:-}" ]; then
-            sed -i "s/qwen3\.7-max/${OPENAI_MODEL}/g" .qwen/skills/triage/references/pr-workflow.md
+          if [ -z "${OPENAI_MODEL:-}" ]; then
+            exit 0
+          fi
+          TARGET=.qwen/skills/triage/references/pr-workflow.md
+          # The default signature literal must still exist in the skill, or
+          # this injection silently no-ops and every staged comment ships the
+          # wrong model name. Fail loudly so a skill edit that renames the
+          # default surfaces here instead of in production comments.
+          if ! grep -q 'qwen3\.7-max' "$TARGET"; then
+            echo "::error::Signature literal 'qwen3.7-max' not found in $TARGET — update this step's sed target to match the skill."
+            exit 1
           fi
+          # Escape sed replacement metacharacters (& / \) so a model name
+          # containing them cannot corrupt the skill text.
+          ESCAPED=$(printf '%s' "$OPENAI_MODEL" | sed -e 's/[&/\]/\\&/g')
+          sed -i "s/qwen3\.7-max/${ESCAPED}/g" "$TARGET"
 
       - name: 'Run Qwen Triage'
         id: 'triage'
@@ -626,8 +639,8 @@ jobs:
           set -uo pipefail
           MARKER='<!-- qwen-triage stage=status -->'
           if [ "${TRIAGE_OUTCOME:-}" = 'success' ]; then
-            EN="✅ **Qwen Triage finished** — [view run]($RUN_URL). See the stage comments above for the result."
-            ZH="✅ **Qwen Triage 已完成** —— [查看运行]($RUN_URL)。结果见上方各阶段评论。"
+            EN="✅ **Qwen Triage finished** — [view run]($RUN_URL). See the stage comments in this thread for the result."
+            ZH="✅ **Qwen Triage 已完成** —— [查看运行]($RUN_URL)。结果见本线程中的各阶段评论。"
           else
             EN="⚠️ **Qwen Triage ended early** — [view run]($RUN_URL). It stopped before finishing; check the run log."
             ZH="⚠️ **Qwen Triage 提前结束** —— [查看运行]($RUN_URL)。未跑完,请查看运行日志。"

.qwen/skills/triage/references/pr-workflow.md

@@ -25,12 +25,21 @@ or Stage 1c direction escalation), submit exactly one `CHANGES_REQUESTED`
 review and stop. Do not also post or update a Stage 1 issue comment, and do not
 continue to Stage 2, Stage 3, or approval.
 
-**Re-runs:** if the triage runs again on the same PR, update each comment in place:
+**Re-runs:** if the triage runs again on the same PR, update each comment in place. **Resolve the comment id by its stage marker AT PATCH TIME — never from memory, list position, or an earlier stage's bookkeeping.** On a re-run the thread holds four or more bot comments whose list order is not the stage order, and a wrong id silently overwrites another stage's comment (observed on a real re-run: the stage=3 comment clobbered with stage=1 content mid-run). The author filter matters too — the marker is public text anyone can paste into a comment, and the bot PAT may be able to edit other users' comments:
 
 ```bash
-gh api -X PATCH "/repos/$REPO/issues/comments/$COMMENT_ID" -F body=@/tmp/stage-N-updated.md
+BOT_LOGIN=$(gh api user --jq '.login')
+stage_comment_id() { # $1 = stage number (1, 2, 3) or "status"
+  gh api "repos/$REPO/issues/$PR_NUMBER/comments" --method GET --paginate -F per_page=100 |
+    jq -rs --arg bot "$BOT_LOGIN" --arg m "<!-- qwen-triage stage=$1 -->" \
+      '[.[][] | select(.user.login == $bot) | select(.body | startswith($m))] | last | .id // empty'
+}
+CID=$(stage_comment_id 2)   # re-resolve immediately before EACH patch
+[ -n "$CID" ] && gh api -X PATCH "/repos/$REPO/issues/comments/$CID" -F body=@/tmp/stage-2-updated.md
 ```
 
+`startswith`, not `contains`: stage comments carry their marker as the first line, and a substring match would also hit a comment that merely quotes the marker. An empty `CID` means that stage has no comment yet — POST a new one instead of patching.
+
 Never create duplicates. For terminal-exit reviews (submitted via
 `gh pr review --request-changes`), the GitHub API does not support editing PR
 reviews. On re-run: check if a `CHANGES_REQUESTED` review from the bot already

scripts/tests/qwen-triage-workflow.test.js

@@ -5,7 +5,15 @@
  */
 
 import { spawnSync } from 'node:child_process';
-import { readFileSync } from 'node:fs';
+import {
+  mkdirSync,
+  mkdtempSync,
+  readFileSync,
+  rmSync,
+  writeFileSync,
+} from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
 import { describe, expect, it } from 'vitest';
 
 const workflow = readFileSync('.github/workflows/qwen-triage.yml', 'utf8');
@@ -306,4 +314,50 @@ describe('qwen-triage tmux workflow', () => {
       workflow.indexOf("- name: 'Install tmux runner tools'"),
     ).toBeLessThan(workflow.indexOf("- name: 'Checkout PR merge ref'"));
   });
+
+  it('escapes injected model names and fails loudly when the signature literal is gone', () => {
+    const injectStep = step('Inject model name into triage signature');
+    const body = injectStep.match(/run: \|-\n([\s\S]*)$/)?.[1];
+    expect(body).toBeTruthy();
+    const script = body.replace(/^ {10}/gm, '');
+
+    const run = (model, content) => {
+      const dir = mkdtempSync(join(tmpdir(), 'triage-inject-'));
+      try {
+        const target = join(dir, '.qwen/skills/triage/references');
+        mkdirSync(target, { recursive: true });
+        writeFileSync(join(target, 'pr-workflow.md'), content);
+        const proc = spawnSync('bash', ['-c', script], {
+          cwd: dir,
+          env: { ...process.env, OPENAI_MODEL: model },
+          encoding: 'utf8',
+        });
+        return {
+          status: proc.status,
+          out: readFileSync(join(target, 'pr-workflow.md'), 'utf8'),
+          log: `${proc.stdout}${proc.stderr}`,
+        };
+      } finally {
+        rmSync(dir, { recursive: true, force: true });
+      }
+    };
+
+    // A model name carrying every sed replacement metacharacter (/ & \) must
+    // land verbatim — the old unescaped sed corrupted the skill text on these.
+    const meta = run('m1/pre&post\\x', 'sig: qwen3.7-max end');
+    expect(meta.status).toBe(0);
+    expect(meta.out).toBe('sig: m1/pre&post\\x end');
+
+    // The signature literal disappearing from the skill must fail the step —
+    // the old silent no-op shipped the wrong model name in every comment.
+    const missing = run('m2', 'sig: some-other-model end');
+    expect(missing.status).not.toBe(0);
+    expect(missing.log).toContain('Signature literal');
+    expect(missing.out).toBe('sig: some-other-model end');
+
+    // No model configured → file untouched, step succeeds.
+    const empty = run('', 'sig: qwen3.7-max end');
+    expect(empty.status).toBe(0);
+    expect(empty.out).toBe('sig: qwen3.7-max end');
+  });
 });