Skip to main content

Testim failure research

Dispatch the testim-researcher agent (agents/testim-researcher.md) — do not research inline in the main loop.

Preflight​

  1. A Testim API key must be resolvable at call time from your secrets manager - see testim-researcher "Setup / Auth". Never persist it to settings.json, a shell profile, or any tracked file. If it cannot be resolved, stop and tell the user which read failed; the key comes from Testim → account settings → API keys, and EU orgs also need TESTIM_API_URL=https://api.eu.testim.io.
  2. Extract from the user's message: the full Testim URL (or test/result IDs) plus any extra context they gave (e.g. "this started after X").

Dispatch​

Pass the agent: the URL/IDs verbatim, and nothing about how to call the API (the agent knows the verified endpoint surface; extra instructions cause endpoint-probing regressions).

Output contract — enforce on relay​

The agent returns a fixed 7-section report. Scope ends at section 6 (code hook):

  • No root-cause analysis, no fullstack trace, no fix — if the user wants those next, that is a separate task starting from the code hook.
  • If the agent's report contains causal analysis ("because", "most likely", backend tracing), trim it before relaying and note the scope.
  • If the user wants to document a bug from this finding, feed the report into identify-bug (Testim research is a valid input there, alongside JAM links).
  • If the traversed app URLs cross into an iframe-embedded app (e.g. a flow in superco-consumer that embeds management-webapp or facility-rentals), the code hook may live in that embedded app's repo, not the host page's. Check repo-map before relaying the code hook if the failing element looks like it's inside an iframe.

Notes​

  • Testim API quota is 2000 requests/month shared org-wide; a single research run uses ≤10. Don't re-dispatch for the same result — continue the same agent via SendMessage.
  • Screenshot URLs in reports expire ~24h after fetch.