Prompt Details
Model
Chat - (gpt-5.6)
Token size
488
Example input
[INCIDENT BRIEF]: Some users report intermittent login failures. Expected successful login; actual occasional 500 responses. Reproduction and recent changes unknown.
[PASTE ERRORS OR LOGS]: Only HTTP 500, no stack trace, request ID or server logs.
[PASTE CODE OR FILES]: None supplied.
[STACK / VERSION / DEPLOYMENT DETAILS]: Web service; stack and version unknown.
Example output
1. Summary of Issue
Observed facts
Some login attempts intermittently return HTTP 500.
Expected behavior:
Plain text
Successful login
Actual behavior:
Plain text
Occasional HTTP 500
Known evidence is limited to:
Plain text
HTTP 500
There is:
no stack trace
no request ID
no server log
no relevant code
no known reproduction pattern
no known recent change
no confirmed application stack
Because of that, the root cause is not established from the supplied evidence.
A 500 only proves that the server encountered an unhandled or internally classified failure somewhere in the login request path. It does not identify whether the failure is in authentication logic, database access, session creation, an external identity provider, infrastructure, or another dependency.
2. Root Cause
Verified cause
Not yet determined.
There is insufficient evidence to identify a primary cause without guessing.
Smallest diagnostic needed
The highest-value next diagnostic is:
Capture one failing login request with its corresponding server-side exception/error log and correlation/request identifier.
Specifically, for one 500, obtain:
Plain text
timestamp
request path
request ID / correlation ID
exception type
stack trace
upstream/dependency error, if any
HTTP status returned
That single artifact will usually separate application-code failures from dependency or infrastructure failures.
Current hypotheses
These are hypotheses only, not confirmed causes.
Because the failure is intermittent, plausible categories include:
Database or connection-pool failure
transient connection exhaustion
timeout
failed user/session lookup
deadlock or database unavailability
Session/token creation failure
cache/session-store outage
signing/key access issue
serialization failure
External authentication dependency
OAuth/OIDC/LDAP/identity-provider timeout
rate limiting
malformed or unexpected upstream response
Unhandled input-dependent exception
only certain users or account states trigger the error
null/invalid profile or credential-related field
Race or concurrency issue
duplicate session creation
concurrent login/update conflict
Infrastructure instability
one unhealthy application instance
load balancer routing to a broken node
memory/resource exhaustion
The evidence does not currently distinguish among these.
Related components
The login path likely involves some subset of:
Plain text
HTTP route/controller
authentication service
user repository/database
password/credential verifier
session or token service
cache/session store
external identity provider
middleware
reverse proxy/load balancer
Which of these is actually affected remains unknown.
3. Impact Assessment
Application impact
Confirmed impact:
some users cannot authenticate successfully
failed requests return a generic 500
The intermittent nature means the service is not completely unavailable.
The percentage of affected users and requests is unknown.
Files affected
No exact files can responsibly be named because no source tree or stack has been supplied.
Any claim such as:
Plain text
auth.js
login.py
session_service.go
would be invented.
The specific file should only be identified after the failing stack trace points to the relevant code path.
Database impact
Unknown.
There is no evidence of:
corruption
failed writes
schema problems
transaction failures
Database involvement is a plausible hypothesis because authentication commonly queries user state, but it is not established.
API impact
Confirmed:
Plain text
login endpoint occasionally returns HTTP 500
Potential downstream effects include:
failed client authentication
retry loops
degraded availability
elevated error rate
No broader API impact is established.
Security implications
There are two relevant concerns.
First, authentication failures on their own do not prove a security vulnerability.
Second, generic 500 responses are preferable to returning raw stack traces or sensitive authentication details to clients. However, server-side diagnostics must still retain sufficient detail for investigation.
Additional security-relevant hypotheses include:
account-specific malformed state
token/signing failures
identity-provider issues
abuse or rate-limit edge cases
None are currently verified.
4. Recommended Fix
A code fix should not be applied yet because the failure mechanism is unknown.
Immediate minimal action
Instrument or retrieve diagnostics around the login request path.
At minimum, each login request should emit a correlation ID and structured error record containing:
Plain text
request_id
timestamp
route
application instance
authentication stage
exception class
exception message
dependency name
latency
final status
Sensitive values must not be logged, including:
Plain text
passwords
session tokens
access tokens
refresh tokens
credential hashes
full authorization headers
Minimal investigation sequence
For the next failing request:
Plain text
1. Identify the exact request timestamp/request ID.
2. Locate the server-side exception.
3. Determine the failing stage:
route -> user lookup -> credential verification -> session/token -> response.
4. Check dependency status at that timestamp.
5. Compare with a successful request for the same path.
If there are multiple application instances, also record the instance/container/host. Intermittent failures caused by one unhealthy node often become immediately visible this way.
Smallest safe fix
The smallest safe fix depends on the evidence.
Examples:
If logs show:
Plain text
database connection timeout
the minimal fix may involve database/pool handling.
If logs show:
Plain text
null reference in session creation
the change should be confined to that specific session path.
If all failures originate from one application node:
Plain text
instance-3
the immediate remediation may be operational rather than a source-code change.
Until one of those is established, changing authentication code would violate the minimal-change requirement and risks masking the real issue.
5. Testing Strategy
Because the root cause is still unknown, testing should first focus on reproducing and isolating the failure.
Diagnostic reproduction matrix
Test login attempts across:
Plain text
same user repeatedly
multiple affected users
known-good user
correct credentials
incorrect credentials
new session
existing session
different application instances, if visible
Record whether failures correlate with:
Plain text
specific users
specific nodes
specific timestamps
request volume
database latency
external identity-provider latency
session-store state
Regression testing
Once the actual cause is found, add a regression test reproducing that exact failure condition.
For example, if the cause turns out to be a transient dependency timeout, the test should simulate that dependency condition and verify the intended behavior.
If the cause is malformed user state, create a fixture matching that state.
Do not create speculative tests for unrelated causes before the failure mechanism is known.
Monitoring improvements
For production, add or confirm:
Plain text
login success rate
login 5xx rate
5xx count by application instance
authentication latency
database/pool errors
session-store errors
external identity-provider errors
Alerting should distinguish:
Plain text
authentication rejection -> expected 4xx
internal authentication failure -> unexpected 5xx
Verification criteria
A fix is only verified when:
Plain text
the original failing condition can be reproduced before the fix,
the specific failure no longer occurs after the fix,
normal successful login still works,
expected authentication failures still return the correct non-500 status,
and no sensitive information is exposed in responses or logs.
No tests have been executed here.
At this point, the correct engineering conclusion is: the incident is confirmed, but the root cause is not yet diagnosable from the available evidence. The next required artifact is one correlated server-side error/stack trace from a failing login request.
Get this prompt + 9 more with Select$14 your first month, then $19/mo$14 first month, then $19/mo
View SelectBy purchasing this prompt, you agree to our terms of service
GPT-5.6
Investigate software defects using supplied code, logs, observed behaviour and environment details. Get an issue summary, evidence-based root-cause assessment, impact analysis, a minimal correction and regression-test strategy. Separates hypotheses from findings and requests missing evidence instead of inventing a diagnosis. Includes four worked debugging examples covering JavaScript, Python and incomplete incident reports.
...more
Added 2 weeks ago
