メニュー
Rank open and unresolved Sentry issues by severity, event volume, users affected, and regression or first-seen status, then produce a prioritized action list. Read-only. Use when user asks "triage Sentry", "what should I fix first", or "rank our open errors".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Atlas architect: boot the workspace. Verify claude-mem and context-mode, scan the project, recommend tooling (confirm first), wire hooks, seed the docs/ SSOT.
Measure atlas's own run health, audit context/asset waste, and mine session transcripts. Three lenses: RUN HEALTH reads the observability DB for metrics (wall-clock, dispatches, verifier coverage) and baseline->target goals; ASSET AUDIT scores each context-loaded skill/agent/plugin by token cost and relevance to flag disable/relocate; SESSION FORENSICS mines transcripts for unused tools and unverified claims into rules. No args reports trends.
Use to map a codebase into feature-grouped flowcharts, find architectural duplication across features, and propose the simplest unified architecture - discovery-first and zero-arg. Runs as a Workflow that fans out one explorer per feature, hunts duplication, and synthesizes a unification proposal with file:line evidence. Use before a refactor, to "find the ideal path," or to unify duplicated systems.
Generate or improve a company-specific data analysis skill by extracting tribal knowledge from analysts. BOOTSTRAP MODE - Triggers: "Create a data context skill", "Set up data analysis for our warehouse", "Help me create a skill for our database", "Generate a data skill for [company]" -> Discovers schemas, asks key questions, generates initial skill with reference files ITERATION MODE - Triggers: "Add context about [domain]", "The skill needs more info about [topic]", "Update the data skill with [metrics/tables/terminology]", "Improve the [domain] reference" -> Loads existing skill, asks targeted questions, appends/updates reference files Use when data analysts want Claude to understand their company's specific data warehouse, terminology, metrics definitions, and common query patterns.
QA an analysis before sharing with stakeholders - methodology checks, accuracy verification, and bias detection. Use when reviewing an analysis for errors, checking for survivorship bias, validating aggregation logic, or preparing documentation for reproducibility.
Build self-contained interactive HTML dashboards with Chart.js, dropdown filters, and professional styling. Use when creating dashboards, building interactive reports, or generating shareable HTML files with charts and filters that work without a server.
| name | sentry-issue-triage |
| description | Rank open and unresolved Sentry issues by severity, event volume, users affected, and regression or first-seen status, then produce a prioritized action list. Read-only. Use when user asks "triage Sentry", "what should I fix first", or "rank our open errors". |
Convert a noisy stream of unresolved Sentry issues into a ranked, actionable list. This skill is read-only: it never changes issue state. It scores each issue on impact signals and tells the user which issues deserve attention first and why.
find_organizations to get the organization slug, then find_projects to get the project slug(s). If the user named an org or project, confirm it against these results instead of guessing.search_issues using a query that restricts to unresolved work, for example is:unresolved. Add is:regression or firstSeen:-24h as additional passes when the user asks about regressions or new errors.search_events scoped to each issue to read recent event counts and the affected-user count over the chosen time window (default last 24h, widen to 14d on request).get_sentry_resource rather than inferring.A prioritized triage brief containing:
update_issue. update_issue is VISIBLE-TO-OTHERS: it mutates Sentry state the whole team sees, so this skill only proposes status changes for a human to apply.search_events returns nothing for an issue, report it as "no events in window" rather than guessing.