发布

  • [OPIK-7051] [DOCS] Add metrics-instrumentation agent skill (#7251)

    frostbyte_neo 发布于 2026-07-02 11:28:41 +00:00

    • [OPIK-7051] [DOCS] Add metrics-instrumentation agent skill

    Captures the end-to-end pattern used for online scoring: per-stage OpenTelemetry
    counters and native histograms with workspace labels read from the reactive
    context, a reactive (non-self-subscribing) enqueue path, and a flow-ordered
    Grafana dashboard with a per-customer funnel, ClickHouse in-query name resolution,
    and a workspace filter. Indexed in .agents/skills/README.md.

    Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

    • [OPIK-7051] [DOCS] Rewrite metrics-instrumentation skill as a normative spec
    • Restructure as a spec: numbered MUST/SHOULD requirements per section (Model,
      Backend metrics, Dashboard, Name resolution, Validation, Constraints, Delivery).
    • Drop the narrative 'gotchas / hard-won' framing; the rules are stated as
      normative constraints (section 6).
    • Remove internal infrastructure references (repo names, prod DB, hosts, channels)
      so the public-repo skill stays generic; the dashboard is described as living in
      the deployment/monitoring configuration outside this repo.

    Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

    • [OPIK-7051] [DOCS] Refine metrics-instrumentation skill (emission-only scope)

    Scope the skill to backend metric emission and align it with the live
    online-scoring reference:

    • Expand §1.2 into a full instrument-selection guide (counter / native
      histogram / gauge / UpDownCounter — when to use and when not to), with
      RED + USE framing; add upDownCounterBuilder to the §2.1 example.
    • State the patterns applied (producer/consumer/error/success/queue-time/
      e2e-latency/saturation/volume) instead of listing classes, and point to
      metric-name families rather than class paths (which move).
    • Drop the workspace_name "unknown" fallback wording (name falls back to
      workspace_id).
    • Move all dashboard-authoring content (layout, PromQL/ClickHouse query
      contracts, per-customer name resolution, dependency panels, validation)
      out to a separate dashboard skill; keep only backend constraints here.

    Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com


    Co-authored-by: Claude Opus 4.8 (1M context) noreply@anthropic.com

    下载附件