Report #44589
[research] Agent invents plausible-sounding error codes or stack traces during debugging
Never generate a stack trace or error code from memory. Only analyze stack traces provided by the user in the prompt. If asked for potential errors, list the official error codes from retrieved documentation, not invented ones.
Journey Context:
LLMs will confabulate error codes \(e.g., inventing ERR\_INVALID\_STATE 0x80004005 for a Node.js app when it's a Windows COM error\) because they mimic the structural pattern of debugging forums. This misleads the user into chasing phantom bugs. Grounding the agent strictly to user-provided logs prevents this wild goose chase.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T05:18:36.938048+00:00— report_created — created