Add AI-assisted triage by renanrodrigo · Pull Request #3 · ubuntu/startriage

renanrodrigo · 2026-06-16T11:59:32Z

This adds the whole AI-assisted functionality to startriage. It's full of abstractions and details but it works well.

Vibe coded
--ai for usual triage second opinion
ai-triage for single bugs

Also make the agent perma prompt part of the package

- make it part of the markdown when that is used - save to separate file if not used - print to stdout for ai-triage command

grayson-wolf

looks good, to me, just a few suggestions / nit-questions

grayson-wolf · 2026-06-16T14:17:01Z

+- **Affected package(s):** source package name(s)
+- **Affected version(s):** package version and Ubuntu release(s)
+- **Symptoms:** what goes wrong (error messages, crashes, incorrect output)
+- **Reproduction steps:** how to trigger the bug (if known)


Suggested change

- **Reproduction steps:** how to trigger the bug (if known)

- **Reproduction steps:** how to trigger the bug (if known). If feasible, these steps should work from a clean LXD container or VM.

I think this should be added because sometimes given reprosteps make a ton of assumptions

I would say get this in, and then refine the prompt later

grayson-wolf · 2026-06-16T19:25:49Z

+#model = "claude-opus-4.8"                        # copilot default; set your own for openrouter
+#github_token = "github_pat_..."                  # or COPILOT_GITHUB_TOKEN / GH_TOKEN env
+#openrouter_api_key = "..."                       # or OPENROUTER_API_KEY env
+#openrouter_base_url = "https://openrouter.ai/api/v1"


are these intended to be commented out?

yep only examples, defaults in the code.

grayson-wolf · 2026-06-16T19:37:51Z

+2. **No hallucinated fixes.** If you cannot produce a fix with confidence that it is correct, set `proposed_fix.kind` to `none`. Do not invent plausible-looking patches.
+3. **No patch application.** Do not generate or apply quilt patches. Do not modify any package source tree as a deliverable. Proposed fixes are returned as a unified diff in the `proposed_fix` field only.
+4. **Read-only external access.** Do not post comments on bugs, change bug statuses, subscribe teams, or modify any external system. Your output is recommendations only; a human engineer will act on them.
+5. **No speculation on internal architecture.** If you don't have enough information about a package's internals, say so rather than guessing.


not sure on the best way to word this as a constraint (or if maybe this should be elsewhere) but an explicit instruction to verify upstream commits actually exist is probably a good idea (the tool has given me completely bogus commits that 404d on github before)

I would say get this in, and then refine the prompt later

TheJJ

coming together pretty nicely. the overall architecture is good, i'd say.

one thing that worries me is the implicit assumption this is run in a secure container now:
how about a text-only analysis without funny containers as the first agent, and when it decides we wanna run shell stuff, it spins up a subagent in an lxd container (using the same mechanism we use for git ubuntu deltarebase resolving - we can hopefully create a shared ai lxd containering library so we don't have to duplicate/reinvent stuff)

otherwise many nits and naming issues, you'll see. great work, overall :)

TheJJ · 2026-06-24T12:34:23Z

+        action="store_true",
+        help=(
+            "Also run AI triage on every bug found. With --markdown the AI report is "
+            "appended to that file; otherwise it is written to autotriage-YYYY-MM-DD.md"


appended? ideally it would just be that file then ->. a complete triage report? :)

TheJJ · 2026-06-24T12:36:12Z

+    ai_triage_p = sp.add_parser(
+        "ai-triage",
+        help="AI-triage one or more Launchpad bugs",
+    )


i think a better name would be an option analyze --ai $what (without --ai it could just show the metadata of these bugs). then we have the option to extend the analysis into fixing with --fix, etc someday.

TheJJ · 2026-06-24T14:18:50Z

    )


+def _build_ai_provider(config: StarTriageConfig):


return should be typed properly

TheJJ · 2026-06-24T14:27:16Z

+async def _ai_triage_results(
+    config: StarTriageConfig,
+    results: Sequence[tuple[str, TriageResult]],
+    provider,


TheJJ · 2026-06-24T14:29:16Z

+def _build_ai_provider(config: StarTriageConfig):
+    """Build the AI provider, printing a friendly hint and returning None on misconfig."""
+    from .ai import build_provider
+
+    try:
+        return build_provider(config.ai)
+    except AIConfigError as exc:
+        print(f"error: {exc}", file=sys.stderr)
+        return None
+
+
+async def _ai_triage_results(
+    config: StarTriageConfig,
+    results: Sequence[tuple[str, TriageResult]],
+    provider,
+    markdown_path: Path | None,
+) -> None:
+    """Run the AI agent over the Launchpad tasks gathered by a normal triage run."""
+    from .ai import payloads_from_tasks, run_agent_on_payloads
+    from .sources.launchpad.triage import LaunchpadTriage
+
+    tasks: list = []
+    for _, result in results:
+        if isinstance(result, LaunchpadTriage):
+            tasks = list(result.tasks.tasks)
+            break
+
+    payloads = await asyncio.to_thread(payloads_from_tasks, tasks)
+    report = await run_agent_on_payloads(config, payloads, provider=provider)
+    if report is None:
+        return
+    _emit_ai_report(report, markdown_path)
+
+
+def _emit_ai_report(report: str, markdown_path: Path | None) -> None:
+    """Persist an AI ``report`` for a ``triage --ai`` run.
+
+    With ``--markdown`` the report is appended (behind a notice) to that file,
+    mirroring how the normal triage markdown is produced. Otherwise it is written
+    to a dated ``autotriage-<date>.md`` file and the path is shown on stdout.
+    """
+    from .ai import append_report, write_report
+
+    if markdown_path is not None:
+        append_report(markdown_path, report)
+        print(f"AI triage appended to {markdown_path}")
+    else:
+        path = write_report(report)
+        print(f"AI triage report written to {path}")
+
+
+async def _run_ai_triage(args: argparse.Namespace, config: StarTriageConfig) -> None:
+    from .ai import gather_user_bug_payloads, run_agent_on_payloads
+
+    provider = _build_ai_provider(config)
+    if provider is None:
+        return
+
+    payloads = await asyncio.to_thread(gather_user_bug_payloads, args.bug)
+    if not payloads:
+        print("No valid bugs to triage.", file=sys.stderr)
+        return
+
+    report = await run_agent_on_payloads(config, payloads, provider=provider)
+    if report is not None:
+        print(report)


this is a lot of code for running, this should rather be moved to the ai/ subdir, since the cli.py is a thin entrypoint for the real logic of available tasks. the idea is that you can run various actions directly over the python api, and cli.py just contains the wrappers that make it cli-runnable.

TheJJ · 2026-06-26T14:14:51Z

+    auth is needed. The token is resolved with config-over-env precedence via
+    :meth:`AIConfig.resolve_token`.
+    """
+    if ai_config.provider is AIProvider.copilot:


could go with a match statement here.

TheJJ · 2026-06-26T14:15:02Z

+    Only OpenRouter (BYOK) contributes here, as an OpenAI-compatible ``provider``
+    block; the Copilot provider authenticates at the client level instead.
+    """
+    if ai_config.provider is AIProvider.openrouter:


could go with a match statement here.

TheJJ · 2026-06-26T14:15:59Z

+        logger.debug("Starting Copilot session (model=%s)", self.model)
+        async with CopilotClient(**build_client_kwargs(self._ai_config)) as client:
+            async with await client.create_session(
+                on_permission_request=PermissionHandler.approve_all,


that should only be done in the container. people should assume running startriage on their hosts doesn't extract precious cat pictures.

TheJJ · 2026-06-26T14:17:43Z

+    try:
+        target.write_text(content, encoding="utf-8")
+        return target
+    except OSError:


that's bad error handling. we should determine a location before even starting analysis, so we're pretty sure writing will work before starting work.

TheJJ · 2026-06-26T14:18:51Z

+        provider = "openrouter"
+        model = "anthropic/claude-3.5-sonnet"
+        openrouter_api_key = "or_secret"
+    """,


indent wrong (doesn't ruff format complain?)

renanrodrigo added 11 commits June 15, 2026 22:29

Add configuration options for the AI-assisted triage

3c5d04f

Add custom to_agent_payload method

ff1d8de

Add provider layer to copilot+openrouter

008d883

Add agent loop and result contract for the AI-assisted triage

3cb5ec6

Also make the agent perma prompt part of the package

Render triage results into the autotriage report

14996da

Add CLI --ai option and ai-triage command

940728a

Package the Copilot runtime and ubuntu-dev-tools

ac1c44a

Add step logging for AI triage observability

3f20940

Remove timeouts and add a spinner to follow AI triage execution

4f1acf2

Reformat AI generated output

5851c2f

- make it part of the markdown when that is used - save to separate file if not used - print to stdout for ai-triage command

Reformat agent_prompts into proper markdown

7041f85

grayson-wolf suggested changes Jun 16, 2026

View reviewed changes

TheJJ requested changes Jun 26, 2026

View reviewed changes

	- Reproduction steps: how to trigger the bug (if known)
	- Reproduction steps: how to trigger the bug (if known). If feasible, these steps should work from a clean LXD container or VM.

Uh oh!

Conversation

renanrodrigo commented Jun 16, 2026

Uh oh!

grayson-wolf 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!

TheJJ 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants