--- name: bugbug-feedback-reporting description: Collects problems with BugBug itself during an agent session (errors, missing features, confusing tool behavior, documentation gaps, workarounds) and sends one support-ready report per problem with bugbug_send_feedback. Use when a BugBug tool fails unexpectedly, the goal cannot be met with BugBug, or a workaround was needed. --- # BugBug Feedback Reporting ## Preflight 1. Confirm the problem is in BugBug or its MCP tools, not in the user's application under test or in the user's own test logic. 2. Confirm `bugbug_send_feedback` is available. If it is missing, or it reports that telemetry is disabled, do not retry; tell the user feedback was not sent. 3. Decide whether the problem is resolved, worked around, or blocking. Send the report once that is clear, not at the first error. ## Inputs Gather from the session; do not ask the user unless a detail is missing and cannot be recovered from the session. - What the user was trying to achieve, in product terms. - Every BugBug tool called, in order, and the outcome of each. - Exact error codes and messages returned by BugBug. - Identifiers involved: project, test, suite, run, or step IDs. IDs are not secrets and let support find the case. - The workaround used, or why none worked. - What would have solved the problem, if known. ## Safety - Never include secrets, credentials, tokens, cookies, variable values, or personal data. Redact them from error text before sending. - Do not paste page content, full responses, or logs; summarize and quote only the error line. - Send one report per distinct problem. Do not send duplicates for the same problem or retry a failed send. - Report problems with BugBug only. Do not report failures of the tested application, flaky selectors the user can fix, or user mistakes. - Do not change BugBug data as part of reporting. ## Workflow 1. Classify the problem: `missing_feature`, `confusing_behavior`, `documentation_gap`, `performance`, or `other`. 2. Rate impact: `blocker` (goal not achieved), `major` (achieved only through a workaround or significant extra effort), `minor` (small friction). 3. Write `summary` as one sentence that names the problem, not the symptom, e.g. "No way to bulk-update a selector across tests". 4. Fill the optional fields so support needs no follow-up: - `goal`: the user's objective and the project ID. - `attempted`: tools and approaches tried, in order, with outcomes and the test, suite, or run IDs involved. Add steps to reproduce when known. - `errorDetails`: exact error codes and messages, redacted. - `suggestion`: what would have solved it. - `relatedTools`: names of the BugBug tools involved. 5. Call `bugbug_send_feedback` once per problem. Continue the task afterwards; mention the report to the user only if they ask or the problem blocked them. ## Output - One `bugbug_send_feedback` call per distinct problem. - When the user asks, a short statement of what was sent and that it went to the BugBug team. If it was not sent, say so and why. ## Resources Resolve bundled resource paths relative to this skill directory. - `assets/bugbug-icon.png` is the OpenAI UI icon for this skill.