Diagnostic checklist
A seven-step review before asking AI to debug an error
The useful middle ground is not simply "paste less." It is a repeatable review that preserves causal evidence, adds facts missing from the log, and removes only context you can justify removing.
Define the failure boundary
Write down the command, route, test, build step, or user action that failed. A log without the triggering action is difficult to interpret.
Verification: Could another developer reproduce the same action?
Find the first cause candidate
Start at the first explicit compiler error, exception, failed assertion, missing module, or non-zero command. Later failures may only be consequences.
Verification: Did you preserve the earliest meaningful failure rather than only the final summary?
Preserve location and causality
Keep the application file, line number, component, job, Docker instruction, and every distinct caused-by exception that connects the symptom to code.
Verification: Can the reader identify where the failure entered application code?
Remove only defensible noise
Reduce exact duplicates, successful setup narration, transfer progress, cache messages, and unrelated warnings. Do not delete context merely because it is long.
Verification: Can you explain why every removed category is irrelevant to this diagnosis?
Add the missing human context
State the expected behavior, environment, recent change, and checks already completed. These facts rarely appear in the stack trace.
Verification: Does the request explain what changed and what should have happened?
Perform a privacy review
Search for credentials, cookies, customer identifiers, private hosts, signed URLs, payloads, and confidential code. Replace values with meaningful placeholders.
Verification: Would publishing this exact excerpt in a public issue be acceptable?
Request one outcome
Ask for the likely root cause, next diagnostic step, smallest safe fix, or reproduction plan. Avoid asking for all of them at once.
Verification: Can the answer be evaluated against one clear task?
Stop and collect more evidence when
- The failure cannot be reproduced consistently.
- The cleaned excerpt no longer contains a source location or causal exception.
- The answer depends on versions, environment variables, generated files, or request data not included in the report.
- The proposed fix involves a broad dependency upgrade, security change, or data migration without supporting evidence.