Triage: #42 "Export to CSV crashes"
Recommendation: category bug, state needs-info. The issue is unlabeled now. The normal path is needs-triage first, but I think it can go straight to needs-info. Either way, the label change and comment wait for your go-ahead.
What I checked
- Prior rejection: there is no
.out-of-scope/directory, so there is no earlier rejection to surface. - Redundancy: this is a crash report, not a feature request, so an existing implementation doesn't apply.
- Codebase: I couldn't explore it. The working directory isn't a git repo, and I only have a file reader here, with no way to list directories or search. I didn't want to guess at paths. If you tell me where the reports page and its export handler live, I'll trace the code path.
- Reproduction: not attempted. There's no running app or repro data, and the report has no steps I could follow.
Why needs-info
The report is too thin to reproduce or to write a durable agent brief from:
- "Spins forever, then the page goes blank" could be a hung request or a server timeout. It could also be an unhandled client-side exception or a memory blowup on a large dataset.
- "Happens on my laptop" gives no browser, OS or app version.
- It doesn't say whether the export works for smaller reports, other reports or other users.
Draft comment (not posted)
> *This was generated by AI during triage.*
## Triage Notes
**What we've established so far:**
- Clicking Export on the reports page shows a spinner that never finishes, then the page goes blank.
- Reported on the reporter's laptop only; we haven't seen it elsewhere yet.
**What we still need from you (@sam):**
- Which report were you exporting, and roughly how many rows does it have? Does a smaller report or a narrower date range export fine?
- Which browser and version are you using, and which OS?
- Open the browser dev tools (Console and Network tabs) and click Export again. Is there a red error in the Console? Does the export request show a status code, or stay "pending"? Please paste what you see.
- Does it fail every time, or only sometimes? Does it also fail for a teammate or on another machine?
- Which app version or deploy are you on?
Next step: should I apply bug + needs-info and post that comment? Or would you rather go to needs-triage first, or give me the code location so I can try to verify the crash from the source before asking @sam anything?
How it was tried: I installed triage together with the grilling and domain-modeling skills it can call, in a clean temporary folder, and invoked it as a slash command with one pasted issue and no issue tracker connected. Only Skill and Read tools were allowed, so nothing was posted, labelled or closed, the codebase could not be explored and no bug was reproduced. It shows only the recommend step and a drafted comment, and it stops to wait for your go-ahead.
What it does
A workflow for maintainers. It moves each issue (and each external pull request, treated as an issue with attached code) through two category roles (bug, enhancement) and five state roles (needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix).
How it works
- Show what needs attention: unlabeled,
needs-triage, andneeds-infoitems with new replies. - For one item: read it in full, check whether the behavior is already implemented, and check the
.out-of-scope/notes for a prior rejection. - Recommend a category and state, then wait for your direction.
- Verify the claim: reproduce the bug, or confirm a PR does what it says.
- If it needs shaping, run a
grilling+domain-modelingsession. - Apply the outcome: an agent brief, triage notes, or a close with an explanation. Bundled guides cover how to write durable agent briefs and the
.out-of-scope/knowledge base.
Good for
Project maintainers with a busy issue tracker. It posts comments and can close issues, so review each action.
High risk: It acts on your issue tracker: it applies labels, posts comments (each starts with an "AI-generated" disclaimer) and can close issues. For a bug or an external PR it may reproduce the bug or check out the PR and run its tests, which runs code you have not reviewed, so use a sandbox or a clean clone for untrusted PRs. It also writes files in the repo (`.out-of-scope/`, glossary or ADR updates). It is written to recommend first and wait for your direction before applying, but read each proposed action before approving. Needs your tracker label mapping and the `grilling` and `domain-modeling` skills. Tried once on a single pasted issue with no tracker connected and no file-writing or shell tools: only the recommendation step and a drafted comment were observed. Nothing was posted, labelled or closed, no bug was reproduced and no PR was checked out.