Check Result

Verify the expected effect of a step actually happened before moving to the next.

jcaiagent7143-ui Updated

File contents

Check Result

Steps

  1. Read step.expected_effect — what observable thing should be true now?
  2. Probe for it independently: re-read the file, GET the API endpoint, query the DB. Don't trust the call's return value alone.
  3. If observed matches expected → ok=true + cite the evidence.
  4. If not → ok=false with the specific mismatch. Decide: retry this step, roll back, or escalate to the user.

Output shape

{
  "ok": true,
  "evidence": "GET /things/42 returned 200 with name='new-thing'"
}
// or
{
  "ok": false,
  "mismatch": "expected file out.json present; got ENOENT",
  "action": "retry-step"
}

Failure modes

  • Trusting the tool's own success signal — many APIs return 200 + body saying "queued" while the side effect never lands.
  • Re-checking via the same code path that just wrote — use an independent probe.
  • Treating "looks ok" as "is ok"; the verdict is binary.

jcaiagent7143-ui/harnessforge/tree/main/src/harness/blueprints/workflow-agent/skills/check-result commit 611b1e430f

Frequently asked questions

npx skillmds@latest add jcaiagent7143-ui/check-result