Help
Troubleshooting
Symptom-first index of the things that go wrong, and where each is fixed.
Find the symptom, follow the link.
| Symptom | Cause and fix |
|---|
| No runs at all | Agent is Inactive, still in Pending validation (a copy or template agent nobody has run yet), or the trigger was never configured — triggers |
| Runs happen, nothing arrives in Slack | Delivery unset ("Framework default") or delivery failed — delivery |
| Output is generic | The integration hasn't synced, or the prompt doesn't name your sources — integration troubleshooting |
| Output includes things it shouldn't | Missing exclusions in the instructions or the skill's When to Use — skills and tools |
| Everything is skipped | Filter too narrow. Read one skip reason — runs and debugging |
| Costs jumped | Model change or a trigger firing more often — credits and usage |
| Worked yesterday, fails today | Expired or revoked key — integration troubleshooting |
| Nobody reviews the output | No owner and no cadence — validation and review |
| Symptom | Cause and fix |
|---|
| Buyer can't hear you | Microphone permission, or the wrong input device — run a session |
| Echo or self-interruption | You're on speakers. Use a headset |
| Long pauses before replies | Network. Wired connection, close other video calls |
| Session shows Not scored | Too short to evaluate, or scoring still running — how scoring works |
| Buyer gets your product or pricing wrong | Grounding, not the persona — document stores |
| Reps score well in practice, badly on real calls | They're learning the bot. Rebuild scenarios from real lost deals — score real calls |
| Symptom | Cause and fix |
|---|
| Reps haven't started | Nudge from the leaderboard, and put certification in the ramp plan |
| Everyone stalls on the same stage | The bar is wrong or the material never taught what the stage tests — stages and pass gates |
| Team average looks alarming | Check the sample size (n=) before reacting — performance |
| Everyone passes stage 2 first time | The stage no longer discriminates. Retire or harden it |
| Symptom | Cause and fix |
|---|
| Interviews conclude "not enough information" | Wrong contact (champion, not decision maker) or too long after the decision — set up a program |
| Stuck in Awaiting buyer | Outreach is still chasing. Nothing to do |
| Duplicate interviews for one deal | An earlier attempt was re-triggered; the old one is superseded — read the results |
| Nobody reads them | Threshold too low. Interview fewer, bigger deals |
| Symptom | Cause and fix |
|---|
| Connected but no data | Scopes. Each setup guide lists the minimum — troubleshooting |
| Last synced weeks ago | Expired token; use Reconnect |
| Agent can't write back | Write scopes not granted |
| Your tool isn't listed | Request it, or connect an MCP server |
| Symptom | Cause and fix |
|---|
| MCP client can't connect | Wrong URL or missing /sse path — connect a client |
| Tools listed, calls fail | Token expired — re-authenticate |
search_knowledge returns nothing | Nothing indexed yet, or the query is too specific — MCP tools |
Include these four things and most problems are diagnosable on first reply:
- What you expected and what happened.
- The agent name and the run time, if it's an agent — or the session, if it's a roleplay.
- The error text from the run's steps, not a paraphrase (runs and debugging).
- What changed recently. New integration, edited agent, rotated key, changed provider permissions.
