Normalize changelog PR metadata from repo sync

.agents/skills/changelog-draft/SKILL.md

@@ -36,11 +36,11 @@ Record the range as `previous_cut_tag..release_tag`.
 
 ### Step 2 — Fetch PR data
 
-Run the `fetch_prs.py` script to collect all PRs merged in the release range and extract explicit changelog markers:
+Run the `fetch_prs.py` script to collect all public-release PRs merged in the release range and extract explicit changelog markers. Pass the repository that the workflow checked out, not necessarily the public repository. Release workflows run from `warpdotdev/warp-internal`, and the script deterministically resolves `warp-repo-sync[bot]` PRs back to their original public `warpdotdev/warp` PR metadata before emitting JSON. When running from `warpdotdev/warp-internal`, the script intentionally omits PRs that were not authored by the repo-sync bot, because those are private internal changes that must not be exposed to the changelog agent or generated artifacts.
 
 ```bash
 python3 .agents/skills/changelog-draft/scripts/fetch_prs.py \
-  --repo warpdotdev/warp \
+  --repo "${GITHUB_REPOSITORY:-warpdotdev/warp}" \
   --base-ref <previous_tag> \
   --head-ref <release_tag>
 ```
@@ -52,6 +52,7 @@ The script outputs JSON to stdout with this structure:
   "prs": [
     {
       "number": 1234,
+      "url": "https://github.com/warpdotdev/warp/pull/1234",
       "title": "...",
       "author": "username",
       "body": "...",
@@ -61,12 +62,22 @@ The script outputs JSON to stdout with this structure:
         { "category": "NEW-FEATURE", "text": "Added dark mode" }
       ],
       "linked_issues": [5678],
-      "changed_files": ["app/src/ai/agent.rs", "crates/warp_features/src/lib.rs"]
+      "changed_files": ["app/src/ai/agent.rs", "crates/warp_features/src/lib.rs"],
+      "source_repo": "warpdotdev/warp",
+      "internal_pr": {
+        "number": 25712,
+        "url": "https://github.com/warpdotdev/warp-internal/pull/25712",
+        "author": "warp-repo-sync[bot]",
+        "title": "...",
+        "repo": "warpdotdev/warp-internal"
+      }
     }
   ]
 }
 ```
 
+Use the top-level `number`, `url`, `author`, `body`, `labels`, `changed_files`, and `source_repo` fields as the source of truth. `internal_pr` is audit-only and must never be used for contributor attribution or user-facing changelog links. If `url` is empty, omit the PR link from user-facing markdown rather than synthesizing one.
+
 ### Step 3 — Classify contributors
 
 Run the `classify_contributors.py` script with the unique author logins from Step 2:
@@ -174,6 +185,8 @@ Combine explicit entries (Step 2) and inferred entries (Step 6) into the final r
 
 PRs marked with `CHANGELOG-NONE` are explicitly opted out and must never appear in the changelog markdown.
 
+When creating entries, copy `pr_number`, `url`, `author`, `source_repo`, and `internal_pr` from the normalized PR record. The release JSON converter uses `url` directly; do not invent public PR URLs from PR numbers.
+
 ### Step 8 — Write output files
 
 Write two files to `output_dir`:
@@ -219,14 +232,17 @@ The markdown draft must **not** include "Needs Review" or "Skipped PRs" sections
   "entries": [
     {
       "pr_number": 1234,
+      "url": "https://github.com/warpdotdev/warp/pull/1234",
       "category": "NEW-FEATURE",
       "text": "Added dark mode",
       "source": "explicit",
       "author": "external-contributor",
       "is_external": true,
       "confidence": "high",
       "rationale": null,
-      "feature_flag": null
+      "feature_flag": null,
+      "source_repo": "warpdotdev/warp",
+      "internal_pr": null
     }
   ],
   "skipped": [...],

.agents/skills/changelog-draft/scripts/convert_to_release_json.py

@@ -31,23 +31,24 @@
     "IMAGE": "images",
 }
 
-REPO_URL = "https://github.com/warpdotdev/warp"
-
 
 def format_entry(entry: dict) -> str:
     """Format a single changelog entry as a text line with a PR link.
 
     Includes external contributor attribution when applicable.
     """
     text = entry["text"]
-    pr_number = entry["pr_number"]
-    link = f"([#{pr_number}]({REPO_URL}/pull/{pr_number}))"
+    pr_number = entry.get("pr_number") or entry.get("number")
+    url = entry.get("url") or entry.get("pr_url")
+
+    link = ""
+    if url and pr_number:
+        link = f" ([#{pr_number}]({url}))"
 
     attribution = ""
     if entry.get("is_external") and entry.get("author"):
         attribution = f" — @{entry['author']} ✨"
-
-    return f"{text} {link}{attribution}"
+    return f"{text}{link}{attribution}"
 
 
 def convert(draft: dict) -> dict:

.agents/skills/changelog-draft/scripts/fetch_prs.py

@@ -26,6 +26,16 @@
     r"(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?)\s+#(\d+)",
     re.IGNORECASE,
 )
+PUBLIC_REPO = "warpdotdev/warp"
+INTERNAL_REPO = "warpdotdev/warp-internal"
+REPO_SYNC_AUTHORS = frozenset(
+    {
+        "app/warp-repo-sync",
+        "warp-repo-sync",
+        "warp-repo-sync[bot]",
+    }
+)
+PUBLIC_PR_URL_RE = re.compile(r"https://github\.com/warpdotdev/warp/pull/(\d+)")
 
 
 def run(cmd: list[str], *, check: bool = True) -> str:
@@ -90,7 +100,7 @@ def get_merged_commits(sha: str) -> list[str]:
 
 def fetch_pr_data(repo: str, pr_number: int) -> dict | None:
     """Fetch PR metadata and changed file paths via gh CLI."""
-    fields = "number,title,author,body,labels,mergedAt,files"
+    fields = "number,title,author,body,labels,mergedAt,files,url"
     raw = run(
         ["gh", "pr", "view", str(pr_number), "--repo", repo, "--json", fields],
         check=False,
@@ -103,6 +113,137 @@ def fetch_pr_data(repo: str, pr_number: int) -> dict | None:
         return None
 
 
+def fetch_pr_commit_messages(repo: str, pr_number: int) -> list[str]:
+    """Fetch commit messages for a PR via the GitHub API."""
+    raw = run(
+        ["gh", "api", f"repos/{repo}/pulls/{pr_number}/commits"],
+        check=False,
+    )
+    if not raw:
+        return []
+    try:
+        commits = json.loads(raw)
+    except json.JSONDecodeError:
+        return []
+
+    messages = []
+    for commit in commits:
+        if not isinstance(commit, dict):
+            continue
+        commit_data = commit.get("commit")
+        if isinstance(commit_data, dict):
+            message = commit_data.get("message")
+            if message:
+                messages.append(message)
+    return messages
+
+
+def get_author_login(data: dict) -> str:
+    """Extract a GitHub login from a gh PR JSON object."""
+    if isinstance(data.get("author"), dict):
+        return data["author"].get("login", "")
+    if isinstance(data.get("author"), str):
+        return data["author"]
+    return ""
+
+
+def get_label_names(data: dict) -> list[str]:
+    """Extract label names from a gh PR JSON object."""
+    label_names = []
+    for lbl in data.get("labels", []) or []:
+        if isinstance(lbl, dict):
+            label_names.append(lbl.get("name", ""))
+        else:
+            label_names.append(str(lbl))
+    return label_names
+
+
+def get_file_paths(data: dict) -> list[str]:
+    """Extract changed file paths from a gh PR JSON object."""
+    file_paths = []
+    for f in data.get("files", []) or []:
+        if isinstance(f, dict):
+            file_paths.append(f.get("path", ""))
+    return file_paths
+
+
+def is_repo_sync_pr(data: dict) -> bool:
+    """Return whether this PR was created by the public-to-internal repo sync bot."""
+    return get_author_login(data) in REPO_SYNC_AUTHORS
+
+
+def should_include_pr(repo: str, data: dict) -> bool:
+    """Return whether a PR should be exposed to changelog generation.
+
+    Releases are cut from warp-internal, but non-sync-bot PRs merged there are
+    private/internal changes. Do not expose them to the Oz changelog agent or to
+    generated artifacts.
+    """
+    return repo != INTERNAL_REPO or is_repo_sync_pr(data)
+
+
+def extract_public_pr_number(text: str) -> int | None:
+    """Extract a public warpdotdev/warp PR number from text."""
+    if not text:
+        return None
+    m = PUBLIC_PR_URL_RE.search(text)
+    if m:
+        return int(m.group(1))
+    # Repo-sync commits commonly preserve the original public squash-merge
+    # subject, such as "feat: add thing (#1234)".
+    m = re.search(r"\(#(\d+)\)\s*$", text.splitlines()[0] if text else "")
+    if m:
+        return int(m.group(1))
+    return None
+
+
+def resolve_public_pr_number(repo: str, pr_number: int, data: dict) -> int | None:
+    """Resolve a repo-sync PR back to its original public warpdotdev/warp PR."""
+    public_pr_number = extract_public_pr_number(data.get("body", "") or "")
+    if public_pr_number is not None:
+        return public_pr_number
+
+    for message in fetch_pr_commit_messages(repo, pr_number):
+        public_pr_number = extract_public_pr_number(message)
+        if public_pr_number is not None:
+            return public_pr_number
+    return None
+
+
+def pr_reference(repo: str, pr_number: int, data: dict) -> dict:
+    """Build a compact audit reference to a PR."""
+    return {
+        "number": data.get("number", pr_number),
+        "url": data.get("url", ""),
+        "author": get_author_login(data),
+        "title": data.get("title", ""),
+        "repo": repo,
+    }
+
+
+def normalize_pr_data(repo: str, pr_number: int, data: dict) -> tuple[str, dict, dict | None]:
+    """Resolve repo-sync PRs to public PR metadata.
+
+    The release workflow runs from warp-internal, where public PRs are mirrored
+    as warp-repo-sync[bot] PRs with different PR numbers. For changelog output
+    and contributor attribution, use the original public PR metadata when it can
+    be resolved, and keep the internal PR under `internal_pr` for audit only.
+    """
+    internal_pr = pr_reference(repo, pr_number, data) if repo != PUBLIC_REPO else None
+    if repo == PUBLIC_REPO or not is_repo_sync_pr(data):
+        return repo, data, internal_pr
+
+    public_pr_number = resolve_public_pr_number(repo, pr_number, data)
+    if public_pr_number is None:
+        return repo, data, internal_pr
+
+    public_data = fetch_pr_data(PUBLIC_REPO, public_pr_number)
+    if public_data is None:
+        return repo, data, internal_pr
+
+    return PUBLIC_REPO, public_data, internal_pr
+
+
 def extract_linked_issues(body: str) -> list[int]:
     """Extract issue numbers from closing keywords in a PR body."""
     if not body:
@@ -161,42 +302,33 @@ def process_pr(pr_num: int) -> None:
         data = fetch_pr_data(args.repo, pr_num)
         if data is None:
             return
-
-        author_login = ""
-        if isinstance(data.get("author"), dict):
-            author_login = data["author"].get("login", "")
-        elif isinstance(data.get("author"), str):
-            author_login = data["author"]
-
-        label_names = []
-        for lbl in data.get("labels", []) or []:
-            if isinstance(lbl, dict):
-                label_names.append(lbl.get("name", ""))
-            else:
-                label_names.append(str(lbl))
+        if not should_include_pr(args.repo, data):
+            return
+        source_repo, data, internal_pr = normalize_pr_data(args.repo, pr_num, data)
+        author_login = get_author_login(data)
+        label_names = get_label_names(data)
 
         body = data.get("body", "") or ""
         explicit_entries = extract_markers(body)
         linked_issues = extract_linked_issues(body)
-
-        file_paths = []
-        for f in data.get("files", []) or []:
-            if isinstance(f, dict):
-                file_paths.append(f.get("path", ""))
-
-        prs.append(
-            {
-                "number": data.get("number", pr_num),
-                "title": data.get("title", ""),
-                "author": author_login,
-                "body": body,
-                "labels": label_names,
-                "merged_at": data.get("mergedAt", ""),
-                "explicit_entries": explicit_entries,
-                "linked_issues": linked_issues,
-                "changed_files": file_paths,
-            }
-        )
+        file_paths = get_file_paths(data)
+
+        pr = {
+            "number": data.get("number", pr_num),
+            "url": data.get("url", "") if source_repo == PUBLIC_REPO else "",
+            "title": data.get("title", ""),
+            "author": author_login,
+            "body": body,
+            "labels": label_names,
+            "merged_at": data.get("mergedAt", ""),
+            "explicit_entries": explicit_entries,
+            "linked_issues": linked_issues,
+            "changed_files": file_paths,
+            "source_repo": source_repo,
+        }
+        if internal_pr is not None:
+            pr["internal_pr"] = internal_pr
+        prs.append(pr)
 
     for sha in commit_shas:
         pr_num = extract_pr_number(sha)

.github/workflows/changelog_draft.yml

@@ -55,6 +55,7 @@ jobs:
             Output directory: ${{ runner.temp }}/changelog-draft
 
             Follow the workflow in .agents/skills/changelog-draft/SKILL.md exactly.
+            When fetching PR data, pass the checked-out repository ("${{ github.repository }}") to fetch_prs.py and rely on the script's repo-sync normalization to resolve public warpdotdev/warp PR numbers, URLs, and authors. The script intentionally omits non-repo-sync PRs from warp-internal because they are private internal changes. Do not infer or synthesize public PR links manually.
 
             After writing the output files, print the full contents of changelog-draft.md to stdout so it appears in the workflow log.
 

.github/workflows/create_release.yml

@@ -1661,6 +1661,7 @@ jobs:
 
             Follow the workflow in .agents/skills/changelog-draft/SKILL.md exactly.
             Make sure to produce both output files: changelog-draft.md and changelog-draft.json.
+            The release workflow may run from warpdotdev/warp-internal. When fetching PR data, pass the checked-out repository ("${{ github.repository }}") to fetch_prs.py and rely on the script's repo-sync normalization to resolve public warpdotdev/warp PR numbers, URLs, and authors. The script intentionally omits non-repo-sync PRs from warp-internal because they are private internal changes. Do not infer or synthesize public PR links manually.
 
             After writing the output files, print the full contents of changelog-draft.md to stdout so it appears in the workflow log.