{
  "skill_name": "wf-planning-fusion-optimizer",
  "evals": [
    {
      "id": 1,
      "name": "Slow replication scenario",
      "prompt": "This Fusion scenario replicates records between two Planning workspaces and takes 40 minutes to run. Here's the blueprint export. Why is it so slow?",
      "expected_output": "Runs scripts/analyze_blueprint.py on the blueprint and presents the Markdown report, leading with the highest-impact read-path findings (schema fetch or per-record GET inside nested loops) framed as calls, operations, and run time.",
      "files": [],
      "assertions": [
        "Runs scripts/analyze_blueprint.py against the provided blueprint rather than reviewing it from memory",
        "Presents the script's report with findings in the order returned (highest impact first)",
        "Explains loop nesting as the multiplier on call volume and operations",
        "Does not invent percentages or time savings the script did not produce",
        "Does not modify the blueprint"
      ]
    },
    {
      "id": 2,
      "name": "Rate-limit 429 errors",
      "prompt": "Our Planning sync scenario keeps hitting 429 rate limit errors under load but works fine in test. Can you review the blueprint?",
      "expected_output": "Treats 429s as a symptom of call volume, runs the analyzer, and recommends structural fixes (collapse per-record reads, use bulk writes, remove sleeps, split connections) rather than only adding delays.",
      "files": [],
      "assertions": [
        "Frames 429s as a symptom of call volume and operations, not a separate problem",
        "Runs scripts/analyze_blueprint.py on the blueprint",
        "Recommends reducing calls (bulk writes, read collapsing) rather than only adding sleep or delay modules",
        "Labels each finding with what it improves (calls, operations, run time, reliability, throughput)",
        "Loads references/findings.md for rationale of reported finding codes when explaining fixes"
      ]
    },
    {
      "id": 3,
      "name": "Non-Planning blueprint",
      "prompt": "Can you make this Salesforce to Slack Fusion scenario faster? Blueprint attached.",
      "expected_output": "Runs or recognizes the scope guard, reports that the blueprint has no Workfront Planning modules and lists the connectors found, and does not force a Planning review.",
      "files": [],
      "assertions": [
        "States the skill is specific to the Workfront Planning connector and does not apply to this scenario",
        "Lists the connectors actually found in the blueprint",
        "Does not produce Planning-specific findings for a non-Planning scenario",
        "May offer general Fusion advice separately, clearly outside this skill"
      ]
    },
    {
      "id": 4,
      "name": "Instance-specific enrichment with MCP",
      "prompt": "I have the Workfront Planning MCP connected. Review this blueprint and tell me which record types and fields the flagged reads actually touch.",
      "expected_output": "Runs the analyzer first, then follows references/mcp-enrichment.md to resolve record-type and field ids to names and confirm read-collapse feasibility, using read-only MCP calls.",
      "files": [],
      "assertions": [
        "Runs scripts/analyze_blueprint.py before any enrichment",
        "Loads references/mcp-enrichment.md",
        "Resolves opaque record-type and field ids to real names via read-only MCP calls",
        "Does not write to or modify the workspace or the blueprint"
      ]
    }
  ]
}