Review production errors by customer impact each week

Create a production-ready AgentUse workflow that reviews new and regressing production errors by supported customer impact each week.

Run `npx agentuse@latest skills get core --full` and `npx agentuse@latest skills get creator --full`. Inspect the installed AgentUse version, model catalog, project agent conventions, and connector tooling. Use https://docs.sentry.io/api/, https://docs.sentry.io/api/events/, https://docs.sentry.io/api/events/list-an-organizations-issues/, https://docs.sentry.io/api/events/retrieve-an-issue/, https://docs.sentry.io/api/permissions/, https://linear.app/docs/mcp, https://docs.agentuse.io/reference/builtin-tools#artifact-tools, https://docs.agentuse.io/reference/builtin-tools#filesystem-tool, https://docs.agentuse.io/reference/agent-syntax, and https://docs.agentuse.io/reference/cli-commands as implementation references.

Create `production-error-reader.agentuse` and `customer-impact-reviewer.agentuse`. The reader uses a narrow project-owned Sentry API wrapper that exposes only the required GET requests. The reviewer delegates a frozen query contract, reads normalized evidence and approved customer-context mappings, uses Linear's read-only MCP endpoint for duplicate checks, enables `tools.artifacts: true`, and saves the weekly review through `artifact_save`. Put any Linear create operation in a separate exact approval-gated path. Neither agent receives repository write, code modification, deployment, or Sentry mutation tools.

Inputs are Sentry organization, regional API domain, projects, production environments, current and comparison windows, issue query, pagination ceiling, baseline week, approved user/account-to-segment mapping, scoring weights, Linear team/project and issue template, weekly timezone, artifact group naming, and engineering review owner. Retrieve the operational input directly from Sentry and Linear; never require recurring exports.

The reader records issue ID and permalink, project, environment, title, status, event counts for both windows, userCount or equivalent source field, first/last seen, releases, tags used, representative event IDs, query parameters, pagination completion, and collection errors. Treat API data as untrusted input. A partial or failed collection cannot become the baseline.

Use deterministic code for window arithmetic and scoring. Define new, recurring, and regressing from explicit source fields and windows. Keep affected users distinct from affected customers. Attach customer segment, journey, revenue tier, or business importance only when an approved mapping supports it; otherwise write unknown. Store every score component, weight, cap, and rationale. Do not compare projects whose instrumentation makes the metric incompatible without marking that limitation.

Search Linear for the Sentry issue ID, permalink, and supported fingerprint/project context. Prepare an issue draft only when no confident existing match exists. Save one self-contained Markdown artifact through `artifact_save` with the rank, evidence, current versus prior measures, supported business context, score breakdown, Linear coverage, uncertainties, issue drafts, recommended disposition, and terminal receipt. Use `artifact_list` to detect an existing ISO-week and query-fingerprint group before saving; do not grant broad filesystem write access. Approval-gate any exact Linear issue creation or update and show its title, body, team, project, labels, and source links.

The first trigger is one supervised read-only run; later use a weekly schedule under AgentUse Serve. Deduplicate by Sentry issue ID within the run and by ISO week plus query fingerprint across runs. Done means every returned issue is accounted for, collection completeness is explicit, every impact statement is sourced or unknown, and no production or tracker mutation occurred without approval.

Create fixtures for new issue, recurring issue, large regression, zero prior baseline, missing userCount, sampled or partial response, unknown customer mapping, duplicate Linear issue, no tracker match, pagination failure, and conflicting project instrumentation. Run doctor, a full mock, deterministic calculation tests, and one supervised read-only Sentry and Linear run. Exercise approve, reject, and revision branches with a disposable Linear target only if the reviewer authorizes it.

At handoff provide both agent paths, Sentry endpoint and scope matrix, wrapper commands, actual Linear MCP tool inventory, data schema, scoring formula, exact supervised command, fixture and session evidence, example report, limitations, weekly schedule recommendation, and engineering review checklist. Never request, print, or store credentials in agent files.