EvalArc / Trace WorkbenchSource & documentation

Golden cases · skill deliveries · evaluator results

Follow the run.
Keep the evidence.

RECORDED Actual local stdio MCP instruction delivery, collected by scripts/record-load.mjs using an authored skill. No model or AWS evaluation; evaluator results intentionally absent.

Inspect each case against its declared acceptance rules. Missing results, skipped evaluations and assessed zero scores remain distinct.

Imported evaluator judgments and caller-declared trace coverage. Identity and consistency checks do not authenticate the producer, rerun a judge, enforce permissions or independently verify task completion.

0Accepted cases
0Rejected cases
1Incomplete cases
Download original inputDownload review JSON
Review your own AgentCore export locally

Wrap the saved spans and Evaluate responses with your versioned golden cases and rubrics, following the input contract. The CLI runs offline and creates this report.

evalarc trace-import input.json --output runs/review-001
evalarc trace-import current.json --baseline baseline.json --output runs/compare-001
evalarc trace-verify runs/review-001
Input contract and preparation guide

Follow each golden case

load-evidence-review

INCOMPLETE

Load the authored evidence-review skill through MCP.

1 exported spans · Skill observation declared complete

Expected skills

  • evidence-review MATCHED
Imported evaluator results
Evaluator / targetValueConfigured gateExplanation
goal-reviewnot-runSession—MISSINGNo result for this declared target.
Builtin.SkillInstructionFollowingnot-run481d88f854ae4e7c—MISSINGNo result for this declared target.
Skill deliveries and recording identity
evidence-review — bundle matched
{
  "name": "evidence-review",
  "trace_id": "7419e7e182a245919fc3c49b1c02b7a7",
  "span_id": "481d88f854ae4e7c",
  "receipt": {
    "schema": "skills-anywhere-load-1",
    "load_id": "dff92916-a748-4b3b-aca9-293d0043c6c1",
    "loaded_at": "2026-09-15T16:44:48.892Z",
    "provider": "dsh-skills-anywhere",
    "provider_version": "0.12.0",
    "name": "evidence-review",
    "skill_sha256": "8e08d5d07d0b7f9683145ef435653e4cfe1d8a312d1255ed79a6d0241e5c7542",
    "content_sha256": "473876db0cd9579138f1cd7905f6037855bea94c4c3740eb2c7f698aef2ddc0b",
    "bundle_sha256": "f93102cb0592a462afb43cc6cf78b09096208aa1772bcaa341e75a20c1deffd0",
    "declared_tools": [
      "Read"
    ],
    "permissions_enforced": false
  },
  "identity": "matched"
}
{
  "session_id": "bbc3c08e-7dc7-434a-9c65-8d745bcc8547",
  "trace_ids": [
    "7419e7e182a245919fc3c49b1c02b7a7"
  ]
}
Frozen configuration, dataset and source identity
{
  "run_id": "mcp-delivery-bbc3c08e-7dc7-434a-9c65-8d745bcc8547",
  "source_sha256": "0c804eaa82da0a065ada91de792edc89de712b056103ef4fa37d0034bbb02662",
  "dataset_sha256": "0f45fb24d56ea9e4ccd4834d4aa414cd0db5a45b5b10a833c1f5fc7c126c3cc4",
  "configuration": {
    "model": "none: local MCP client",
    "model_parameters": {},
    "prompt_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
    "tools_sha256": "2c52bc1da3c4285bb25979f13097557ea48d74d978902a278c9d869ab5ff7aa9",
    "skills": {
      "evidence-review": "f93102cb0592a462afb43cc6cf78b09096208aa1772bcaa341e75a20c1deffd0"
    }
  },
  "evaluators": [
    {
      "id": "goal-review",
      "revision": "not-run",
      "level": "session",
      "rating": {
        "kind": "numeric",
        "min": 0,
        "max": 1,
        "pass_at_least": 1
      }
    },
    {
      "id": "Builtin.SkillInstructionFollowing",
      "revision": "not-run",
      "level": "skill",
      "rating": {
        "kind": "numeric",
        "min": 0,
        "max": 1,
        "pass_at_least": 1
      }
    }
  ]
}