Bug Triage Assistant
Convert scattered bug notes, Slack messages, or user complaints into structured engineering tasks with reproduction steps, severity, and root cause hypothesis.
Answer questions using your KB content directly — citing the source, staying within documented scope, and refusing to fill gaps with guesses.
KB search returns documents; users need answers. When someone searches your internal docs and finds a 12-page article, they scan it, don't find what they need, and give up or ask a colleague. The result is an assistant that answers in your organization's voice, cites the specific KB section it's drawing from, and responds 'not found' rather than generating a plausible-sounding guess.
Define the KB scope
Specify which topics, systems, or processes the assistant should cover. Out-of-scope questions should be redirected, not guessed.
Paste the relevant content
Include the KB articles or sections relevant to the expected question set. The assistant's quality is bounded by what you provide.
Set the escalation path
Specify where to direct questions the assistant can't answer — a team, a Slack channel, a ticketing system.
Test against real questions
Run the 10 most common questions your KB currently gets asked. Gaps in coverage are easier to fix before deployment than after.
It's told to say so, not guess. Rule 3 returns an explicit "I don't have documentation on this. You may need to contact [team or channel] or check [location]" — you fill the escalation path in step 3. The system-prompt-generator produces this prompt; it also bans answering from general knowledge, but the assistant only refuses reliably for content you actually pasted in.
Citation is built into the answering steps: rule 2 requires "Per [Article Title / Section Name]: ..." before every answer drawn from your pasted KB content. That's what makes an answer verifiable when the KB later changes. You still keep the pasted sections current — notFor warns stale KB content produces wrong answers even with citations.
Because organizational policy isn't a generality. The Rules block bans "typically" or "usually" when referencing your specific policies — the assistant must be definitive from the pasted content or cite the gap, never hedge toward a plausible-sounding norm. This pairs with the no-general-knowledge rule to keep answers traceable to a documented section rather than the model's priors.
Convert scattered bug notes, Slack messages, or user complaints into structured engineering tasks with reproduction steps, severity, and root cause hypothesis.
Turn AI into a structured pull request reviewer that catches risky changes, flags maintainability issues, and suggests missing test coverage.
Configure AI to answer support questions within your actual policy boundaries — not generic best-guess answers.
A complete 'act as a product manager' role prompt — the perspective, responsibilities, and decision criteria a real PM brings, not just the job title.
A reusable AI agent task template with variables for objective, context, available tools, constraints, success criteria, failure handling, and output format.
Formats fuzzy agent instructions into a structured prompt with objective, available tools, constraints, success criteria, and failure handling.
Build structured system prompts from role, tone, constraints, and model target.