Troubleshooting

Agent Troubleshooting

Common Agent issues and fast diagnosis workflows for stale values, merge fields, and signing flow errors.

Definition #

Agent Troubleshooting is a structured method for identifying, isolating, and resolving runtime issues in candidate journeys.

Why it matters #

Fast diagnosis reduces candidate drop-off, protects conversion quality, and shortens recovery time during live operations.

How it works #

Use a layered approach:

  1. Reproduce with exact candidate path and step sequence.
  2. Inspect data state (candidate fields, metadata, workflow variables).
  3. Check rendering layer (runner content and merged output).
  4. Validate integration responses (API/webhook/service outcomes).
  5. Confirm persistence (database record creation/update timing).
  6. Apply fix + verify with the same scenario.

Configuration fields #

When investigating, always capture:

  • Candidate id / token
  • Agent id / path id / step id
  • Widget type and configuration snapshot
  • Relevant merge fields and raw source values
  • Integration request/response payload (masked if sensitive)
  • Error code and event timestamp window

Example scenario #

Stale value after option selection #

Symptom: Next step shows old or empty value in confirmation text.
Likely cause: frontend state not refreshed after backend write.
Check: whether step progression returns refresh indicator and re-fetches workflow data.

Merge field appears empty #

Symptom: rendered text shows blank where dynamic field is expected.
Likely cause: wrong key, missing metadata value, or merge processing not applied to that property.
Check: raw candidate data + metadata + template token naming + property merge handling.

Signature flow returns 409 (processing) #

Symptom: signing URL creation fails with “cannot create ceremony, envelope processing.”
Likely cause: race condition between envelope readiness and ceremony request.
Check: retry/backoff behavior, persistence timing for sign-document mapping, webhook order tolerance.

Common mistakes #

  • Testing with inconsistent candidate data between runs
  • Debugging UI only, without checking backend persistence timing
  • Assuming webhook event order is deterministic
  • Not separating unrelated external errors from primary root cause
  • Fixing symptom text without validating data lifecycle

Troubleshooting checklist #

  • Reproduce with one deterministic test candidate
  • Confirm source data written successfully before render step
  • Verify merge token names exactly match available context keys
  • Validate front-end refresh behavior after save/proceed actions
  • Inspect integration status codes and retry behavior
  • Check webhook timing gaps and lookup fallbacks
  • Re-test end-to-end after fix (including back/forward navigation)

Last updated Mar 28, 2026