Triage a raw bug report into something a team can act on — clean repro steps, a defensible severity/priority, environment, likely area/owner, and duplicate check.
SkillScore87Maintained
Automatic
◆QA & Test Engineering pack
Last verified 8/31/2026
What it changes
Passed 20 of 23 test cases with the skill, and 14 of 23 without it.
Without skill61%
With skill87%
Checks it passed that the model alone missed
Refrains from fabricating specific navigation actions.
Evaluates impact, frequency, and workaround factors in the rationale.
Includes both Severity rating and Priority rating in the top header line.
and 5 more in the full report
6 fixed14 passed either way3 still failing0 broke
Pass-rate lift
+26 pp
Tokens
+21%
Agent turns
0%
Response time
+2%
Model gemini-3.6-flashMeasured 8/31/2026v120 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 6
Prompt
Here is a raw report from our Slack support channel: 'Hey, whenever a manager tries to export the monthly payroll CSV on v2.4 in Chrome, the page just freezes completely and they lose their unsaved filters. There is no workaround currently.' Please run this through a full triage pass. I need the output formatted with structured reproduction steps showing expected versus actual outcomes, distinct severity and priority ratings with explicit justification, a summary table for environment metadata, routing recommendations for component and team queue, a duplicate check against existing issues, and a list of any missing details we still need from the reporter.
What was checked, without and with the skill
Checked
Without skill
With skill
Includes both Severity rating and Priority rating in the top header line.
Missed
Passed
Justifies severity and priority using an explicit calculation of impact, frequency, and workaround factors.
Missed
Passed
Includes an environment summary table containing Version, Platform, User/role, First seen, and Frequency columns.
Missed
Passed
case-01 · gemini-3.6-flash · 8/31/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md70 linessingle file
Bug Triage Pack
A bug backlog is only as good as its triage — and most bugs arrive as "it's broken" with no repro, a panicked severity, and no idea who should look. This turns a raw report into a triaged item: reproducible steps, a severity and priority you can defend (not vibes), the environment that matters, a suspected area and owner queue, and a check for whether it's a duplicate — asking for what's missing instead of inventing it.
What This Skill Produces
Normalized repro — clear, numbered steps; expected vs. actual; the minimal path to reproduce
Severity & priority — each with reasoning (impact × frequency × workaround availability), not a gut number
Environment & metadata — version, platform, browser/OS, user/role, first-seen, frequency
Suspected area & owner queue — where it likely lives and who should take it next
Duplicate/related check — is this the same as a known issue, and what to link
Missing-info flags — what the reporter must supply before it's actionable
Required Inputs
Ask for these if not provided:
The raw report — whatever came in (a Slack message, a customer ticket, a screenshot description)
Your severity/priority scale — what P0–P3 / S1–S4 mean here (else a sensible default is used and labelled)
Environment details — version, platform, who hit it, how often
Known issues (optional) — a list to dedupe against
Framework: Triage You Can Defend
Repro or it's not a bug yet. No reliable steps → the first action is "needs repro," not a severity.
Severity ≠ priority. Severity is how bad the impact is; priority is when we fix it (a rare-but-catastrophic bug and a constant-but-cosmetic one differ on both axes).
Score, don't feel. Impact (data loss > broken flow > cosmetic) × frequency (all users > edge case) × workaround (none > easy) → a defensible level.
Route it. Suspected component and the queue/owner it should go to — triage that doesn't route just moves the pile.
Dedupe. Check against known issues; a linked duplicate is worth more than a fifth copy.
Flag gaps, don't guess. Missing environment or repro is a request to the reporter, not an assumption.
Output Format
Bug: [title] · Severity [Sx] · Priority [Px]
Why this level: impact [x] × frequency [y] × workaround [z].
Repro
… → Expected: … · Actual: …
Environment
| Version | Platform | User/role | First seen | Frequency |
Pressure-test a business or side-hustle idea before you sink money and months into it — find the real demand, the risky assumptions, and the cheapest way to test them. Use when asked to validate my business idea, is this a good business idea, test my side hustle, or should I start this. Produces a read on the core assumptions the idea depends on, who the customer really is and whether the pain is real, the cheapest experiments to test demand before building, a rough viability check (market, comp