Debug Agents with Weaver
Weaver gives you built-in tools to find, fix, and prevent errors without leaving chat or the Canvas. This guide walks you from quick fixes to deeper diagnosis, then prevention.
Spot Problems Quickly
Component border colors show status during a run:
Border Colour | Meaning | What To Do |
---|---|---|
🟢 Green | Success | No action needed |
🟡 Yellow | Waiting for upstream | Inspect earlier components |
🟠 Orange | Missing required input | Connect the correct input or fill a field |
⚪ Grey | Still running | Let it complete |
🔴 Red | Failed | Start diagnosis |
Diagnose with Inputs and Outputs
Click any red component to open its inspector. You will see the inputs the block received and the output or error it produced. Look for:
- Empty or malformed inputs
- 500-level API responses
- Quota or rate-limit errors
- JSON parsing failures
Example diagnosis:
Fix with AI
When the issue is prompt or parameter related, use Fix with AI on the failing block. Provide a short, direct instruction:
Weaver updates the component’s prompt or parameters, reruns the block, and shows you what changed. If it stays red, refine your instruction and try again.
Handle Cascading Failures
Select multiple components with Cmd/Ctrl + click, then rerun them in sequence. This helps when a bad output flows downstream and several blocks need to be re-executed together.
Common Error Recipes
Symptom | Likely Fix |
---|---|
JSON parse error | In the upstream LLM or transformer, constrain output to strict JSON and validate before passing forward. |
Quota or rate limit | Insert a Sleep component between retries. You can pass a variable like Retry-After to wait exactly as required. |
Empty input | Check Canvas wiring and ensure the expected upstream output is connected. |
Unexpected API format | Normalize with a JSON Filter or transformer before the next step. |
Prevent Tomorrow’s Bugs
- Keep workflows modular so each step is easy to test in isolation.
- Save real test inputs and reuse them while iterating.
- Lean on linting to catch prompt ambiguities, weak types, or missing auth early.
- Throttle retries with Sleep to respect API limits and backoff policies.
Use Debug Mode and Logs
Turn on Debug Mode from the toolbar when you need deeper inspection. You will see:
- Execution timeline per component
- Request and response payloads
- Headers and runtime metadata
- Parsed inputs and structured outputs
- Model prompts and replies
- Click-through trace from a log entry to the block on Canvas
Use the Agent Activity Logs to compare expected versus actual results across runs and to spot slow or flaky steps.
When You Need Extra Help
- Export your agent and test it locally with the SmythOS runtime.
- Share minimal repro steps and payloads in the Discord debugging channel.
- Keep logs handy so others can help you pinpoint the failure quickly.
Next Steps
- See real prompts that trigger these patterns in the Prompt Library.
- If you are new to Weaver, try the Weaver Quickstart.
- Browse the Component Library for Sleep, JSON tools, and other helpers.
- Review your run history in Agent Logs whenever something feels off.