Link Hackbot's Bugzilla comments to their run with front matter by kyoshino · Pull Request #7433 · mozilla/bugbug

kyoshino · 2026-11-09T06:01:32Z

For Bug 2086124.

Important

Deploy this only once mozilla/bmo#2819 is in production and BMO's hackbot_user param names the account hackbot-api posts as. Until then, Bugzilla shows the block as plain text at the top of every Hackbot comment.

hackbot-api now puts a front matter block at the top of every Bugzilla comment it posts:

---
hackbot:
  run: <run ID>
  action: <action ID>
  agent: <agent name>
  url: https://hackbot.moz.tools/runs/<run ID>
---

BMO's Hackbot extension hides the block from readers, bugmail, webhooks and Pulse, and shows the agent and a link to the run in the comment header instead. REST raw_text keeps it. This is what native feedback (bug 2086122) will attach to.

The block is added at apply time in bugzilla_handler.py, so a model can't write or change it, and the Hackbot UI and Slack never show it.
ApplyContext carries the action ID (RunAction.ref). When a field update and a comment are merged into one PUT, it carries the comment's ID.
The block is left out when the action has no ID (summaries recorded before action IDs) or a value isn't an identifier, because BMO would show it as text.
Agent text posted without our block starts on a new line when it begins with ---, so BMO can't read a block the model wrote as metadata. That covers the case above, a comment inside update_bug's changes, attachment comments and new-bug descriptions.
The feedback footer is unchanged; that change belongs to bug 2086122.

Testing

hackbot-api tests pass (424), including new ones for the block, the merged-comment ID, and agent-written blocks.
Pre-commit hooks pass on the changed files.
On a local BMO with the extension, a comment that starts with a newline keeps it, and the block below it isn't read as metadata.

🤖 Generated with Claude Code

hackbot-api now puts a front matter block at the top of every Bugzilla comment it posts: the run, the action, the agent and a link to the run in the Hackbot UI. BMO's Hackbot extension hides the block and shows the agent and the link in the comment header instead (bug 2086124). The block is added at apply time, so a model can't write or change it, and the Hackbot UI and Slack never show it. ApplyContext carries the action ID; a field update merged with a comment carries the comment's ID. The block is left out when the action has no ID (older summaries) or a value isn't an identifier, since BMO would show it as text. Agent text posted without our block (those comments, a comment inside update_bug's changes, attachment comments and new-bug descriptions) starts on a new line when it begins with a rule, so BMO can't read a block the model wrote as metadata.

kyoshino · 2026-11-09T06:20:38Z

It seems that the test failure is related to GitHub’s downtime, in the same way as the pull request to BMO. Please rerun the test once the issue has been resolved.

suhaibmujahid

Thank you! It looks great. I left some simple comments.

suhaibmujahid · 2026-11-12T02:17:46Z

                for member in member_rows
            ]
+            # The group carries at most one comment, which keeps its own ID.
+            comment_id = next(


Suggested change

comment_id = next(

comment_action_id = next(

suhaibmujahid · 2026-11-12T02:19:19Z

            params = resolve_placeholders(row.params, results_by_ref)
            outcome = await _dispatch(
-                run, row.type, params, attachments, hackbot_client
+                run, row.type, params, attachments, hackbot_client, row.ref


We should pass the id instead of the ref:

Suggested change

run, row.type, params, attachments, hackbot_client, row.ref

run, row.type, params, attachments, hackbot_client, row.id

suhaibmujahid · 2026-11-12T02:19:30Z

            ]
+            # The group carries at most one comment, which keeps its own ID.
+            comment_id = next(
+                (m.ref for m in member_rows if m.type == "bugzilla.add_comment"),


Same as https://github.com/mozilla/bugbug/pull/7433/changes#r4545447248

suhaibmujahid · 2026-11-12T02:21:49Z

    params: dict,
    attachments: list[dict],
    hackbot_client: HackbotClient,
+    action_id: str | None,


If we use id instead of ref that would be int instead of str and it will never be None.

suhaibmujahid · 2026-11-12T02:27:33Z

+            body["comment"] = _with_front_matter(params["comment"], ctx)
            body["comment_tags"] = [ctx.agent]
+        elif isinstance(body.get("comment"), dict) and isinstance(
+            body["comment"].get("body"), str
+        ):
+            # A comment the agent put in `changes` itself, which is also posted
+            # from the Hackbot account.
+            body["comment"] = _with_front_matter(body["comment"], ctx)


We could simplify it be checking for body.get("comment") after if params.get("comment"):

if params.get("comment"): body["comment"] = params["comment"] body["comment_tags"] = [ctx.agent] if body.get("comment"): body["comment"] = _with_front_matter(body["comment"], ctx)

suhaibmujahid · 2026-11-12T02:33:30Z

+    return "\n".join(["---", "hackbot:", *lines, f"  url: {url}", "---", "", ""])
+
+
+def _unparsed(text: str) -> str:


Suggested change

def _unparsed(text: str) -> str:

def _escape_front_matter(text: str) -> str:

suhaibmujahid · 2026-11-12T02:38:59Z

+    fields = {"run": ctx.run_id, "action": ctx.action_id, "agent": ctx.agent}
+    invalid = [key for key, value in fields.items() if not _IDENTIFIER.fullmatch(value)]
+    if invalid:
+        log.warning(


So we get a single about this in Sentry:

Suggested change

log.warning(

log.error(

Alternatively, we could raise an exception to fail loudly.

suhaibmujahid · 2026-11-12T02:43:39Z

+# BMO's Hackbot extension reads this block off the top of comments from the
+# Hackbot account, hides it, and shows the agent and a link to the run in the
+# comment header instead. It reads only this shape: `run`, `action` and `agent`
+# must be identifiers and `url` an https URL. Anything else shows as plain text,
+# so a block that would not parse is left out rather than posted.
+_IDENTIFIER = re.compile(r"[A-Za-z0-9._-]+")


Do we really need this? Could we make it dynamic so it accepts whatever we pass in? Or does it solve a technical limitation?

kyoshino requested a review from a team as a code owner November 9, 2026 06:01

Merge branch 'master' into claude/hackbot-feedback-system-66a0ee

b063fb6

suhaibmujahid requested a review from ayoubdiourin7 November 12, 2026 02:04

suhaibmujahid requested changes Nov 12, 2026

View reviewed changes

Navigation Menu

Repository navigation

Link Hackbot's Bugzilla comments to their run with front matter - #7433

Link Hackbot's Bugzilla comments to their run with front matter#7433
kyoshino wants to merge 2 commits into
mozilla:masterfrom
kyoshino:claude/hackbot-feedback-system-66a0ee

kyoshino commented Nov 9, 2026

Uh oh!

kyoshino commented Nov 9, 2026

Uh oh!

suhaibmujahid left a comment

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

suhaibmujahid Nov 12, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	run, row.type, params, attachments, hackbot_client, row.ref
	run, row.type, params, attachments, hackbot_client, row.id

		return "\n".join(["---", "hackbot:", *lines, f" url: {url}", "---", "", ""])


		def _unparsed(text: str) -> str:

	def _unparsed(text: str) -> str:
	def _escape_front_matter(text: str) -> str:

Conversation

kyoshino commented Nov 9, 2026

Testing

Uh oh!

kyoshino commented Nov 9, 2026

Uh oh!

suhaibmujahid left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants