@guandata/guanmetric 0.1.17 → 0.1.19

package/skills/guanmetric/evals/fixtures/create-result-unknown-state.json

@@ -0,0 +1,13 @@
+{
+  "schemaVersion": "1.0",
+  "planHash": "plan_hash_example",
+  "environment": {"profile": "test153", "baseUrl": "http://test153.example", "domain": "example_domain"},
+  "status": "blocked",
+  "items": [{
+    "key": "sales",
+    "subType": "ATOMIC",
+    "status": "blocked",
+    "stage": "reconcile",
+    "error": {"code": "FLOW_CREATE_RESULT_UNKNOWN", "category": "transport", "message": "request timed out before a metricId was returned"}
+  }]
+}

package/skills/guanmetric/evals/forward-eval-2026-09-21.json

@@ -0,0 +1,62 @@
+{
+  "schemaVersion": "1.0",
+  "date": "2026-09-21",
+  "mode": "blind-forward-eval",
+  "environmentWrites": false,
+  "summary": {
+    "cases": 8,
+    "agentDecisionPasses": 8,
+    "evalAssertionsCorrected": 2,
+    "productGapsFound": 1,
+    "productGapsOpen": 0
+  },
+  "cases": [
+    {
+      "id": 1,
+      "result": "pass",
+      "evidence": "Correctly decomposed atomic, composite and derived metrics, and kept the workflow read-only before user confirmation."
+    },
+    {
+      "id": 2,
+      "result": "pass-after-eval-correction",
+      "evidence": "Did not guess identifiers or definitions. The original assertion incorrectly required the standard Excel path for a single metric; the skill permits the dedicated create path after read-only resolution, confirmation, dry-run and precheck."
+    },
+    {
+      "id": 3,
+      "result": "pass",
+      "evidence": "Rejected an invalid composite reference to a derived metric and proposed valid alternatives that still require user confirmation."
+    },
+    {
+      "id": 4,
+      "result": "pass-after-eval-correction",
+      "evidence": "Stopped because the fixture was only a summary and the standard XLSX was absent. The original assertion incorrectly expected Flow generation without its required source artifact."
+    },
+    {
+      "id": 5,
+      "result": "pass",
+      "evidence": "Used the dedicated edit path, mapped the read model to the write model, preserved source fields, excluded read-only fields and required dry-run and impact validation."
+    },
+    {
+      "id": 6,
+      "result": "pass",
+      "evidence": "Separated published and draft metrics instead of mixing incompatible states in one Flow."
+    },
+    {
+      "id": 7,
+      "result": "pass-with-product-gap",
+      "evidence": "The clean retest did not retry creation and kept blocked/reconcile. It also exposed that the CLI has no reconcile/adopt command for safely binding a verified server metric ID back to Flow state.",
+      "productGap": "Add a dedicated recovery command with environment identity, planHash, subtype and metric-definition consistency checks; never require manual state editing.",
+      "resolution": "Implemented flow reconcile with preview-by-default, explicit confirm, definition checks and a hard no-retry guard in flow apply."
+    },
+    {
+      "id": 8,
+      "result": "pass",
+      "evidence": "Kept downstream work blocked while approval was pending and did not claim the metric was usable."
+    }
+  ],
+  "notes": [
+    "The first case-7 run was excluded because that evaluator opened evals.json.",
+    "A clean case-7 retest was used for the recorded result.",
+    "No BI resources were created, edited or deleted during this evaluation."
+  ]
+}