Step 6 of 8 in Zero to First Company All tips
Deployment & Opsbeginnercurrenttested against paperclip 0.9

Reading agent logs and activity

Use the logs to find the real failure point before you touch prompts, budgets, or org structure.


Read logs in order

Check the run in this order:

  1. task accepted
  2. context loaded
  3. tool calls attempted
  4. output produced

That shows where the chain actually broke.

Common early failures

  • missing files in context
  • bad adapter credentials
  • tool permissions mismatch
  • tasks too vague for the agent role

Good debugging habit

Do not rewrite the prompt after one failure. First confirm whether the agent even reached the reasoning step.

If it failed before the model did any real work, prompt tuning is the wrong fix.