Debugging prompts
Find root causes instead of patching symptoms: stack traces, flaky tests, prod-only bugs and performance.
Root-cause a bug, don't patch symptoms
Strong reasoning model recommended
Variables: expectedactualstepserror_outputcode_or_files
Help me find the root cause of this bug. Don't propose a fix until we've confirmed the cause.
What I expected: {{expected}}
What happened: {{actual}}
Steps to reproduce: {{steps}}
Error / logs:
```
{{error_output}}
```
Relevant code: {{code_or_files}}
Process:
1. List 3-5 plausible hypotheses, ranked by likelihood, each with the evidence for and against.
2. For the top hypothesis, give me the single quickest experiment (log line, breakpoint, command) to confirm or rule it out.
3. Wait for my result before continuing. Explain this error like I'm new to the stack
Fine on small local models (7B+)
Variables: error_messagelanguage_or_framework
Explain this error message in plain English:
```
{{error_message}}
```
Context: I'm using {{language_or_framework}}.
Tell me: what it literally means, the 3 most common causes, how to check which one applies to me, and the usual fix for each. Keep it under 250 words. Stack trace triage
Paste the full trace, not a screenshot
Variables: languagestack_trace
Here is a stack trace from {{language}}:
```
{{stack_trace}}
```
1. Identify the frame where the error originates in MY code (not library code) and explain why that's the likely culprit.
2. Explain the chain of calls that led there.
3. List what information is missing to be sure (variable values, inputs).
4. Suggest the exact log statements to add, with file and line, to capture that information. Works locally, fails in production
Any model
Variables: environmentsymptomlocal_setupprod_setuplogs
My app works locally but fails in {{environment}}.
Symptom: {{symptom}}
Local setup: {{local_setup}}
Production setup: {{prod_setup}}
Logs from production: {{logs}}
Walk through the usual local-vs-prod differences one by one (env vars, build mode, file paths and case sensitivity, network/CORS, timeouts, memory, dependency versions, time zones, permissions, caching/CDN) and for each say whether it fits my symptom and how to check it. Finish with the top 3 suspects. Flaky test investigator
Any model
Variables: test_codefailure_output
This test passes sometimes and fails sometimes:
```
{{test_code}}
```
Failure output: {{failure_output}}
List the likely sources of flakiness in this specific test (timing/async, shared state, test order, randomness, dates/time zones, network, unseeded data). For each, point to the exact line that makes it vulnerable and show a deterministic rewrite. Performance bottleneck hunt
Give real measurements if you have them
Variables: what_is_slowcurrent_timetarget_timecodestack
This is slow: {{what_is_slow}} (currently ~{{current_time}}, target {{target_time}}).
Code / query:
```
{{code}}
```
1. Point out the likely hotspots and estimate their cost (big-O or I/O round trips).
2. Tell me how to measure before changing anything (profiler, EXPLAIN, timing logs) for {{stack}}.
3. Suggest fixes in order of effort vs impact. Don't sacrifice readability for micro-optimisations. Rubber-duck mode
Great for small local models
Variables: code_and_explanation
Be my rubber duck. I'll explain what my code is supposed to do, line by line. Your job:
- Ask one short clarifying question at a time.
- Point out any place where what I say and what the code does don't match.
- Don't give me the answer unless I ask "just tell me".
Here's the code and my explanation:
{{code_and_explanation}} Bisect a regression
Any model
Variables: good_versionbad_versionsymptom
Something broke between {{good_version}} and {{bad_version}}. Symptom: {{symptom}}.
Walk me through using `git bisect` to find the commit: the exact commands, how to write an automated test script for `git bisect run` for this symptom, and what to do when I find the commit (how to read it, revert vs fix-forward). Get all 209 prompts + agent workflows
Every prompt as Markdown and JSON, 10 multi-step agent workflows, and 12 ready-made slash commands for Cursor and Claude Code. Instant zip download, one-time purchase.
New prompts every week
One short email with the best new prompts and agent workflows we've tested. Free; unsubscribe any time.