// Use when about to state a conclusion, diagnosis, recommendation, or any technical assertion — requires actively arguing against your own position before committing to it, to catch first-match bias, confirmation bias, and shallow analysis.
Use when explicitly asked to discover and document a project's development workflow by indexing the repository
Use when about to make any factual claim — API behavior, library features, configuration options, platform specifics, version details, or any technical assertion — without a verifiable reference link. Requires searching online resources and citing authoritative sources before stating facts.
Dispatches long-running commands to terminal multiplexers (tmux, screen, zellij) instead of blocking the agent's execution tool. Use when about to execute a command that runs indefinitely or for extended periods, such as dev servers, file watchers, log tails, or long test suites.
Prevents piping command output through tail or head which permanently discards data. Use when about to execute shell commands, especially builds, tests, or any command with potentially large output.
| name | are-you-sure |
| description | Use when about to state a conclusion, diagnosis, recommendation, or any technical assertion — requires actively arguing against your own position before committing to it, to catch first-match bias, confirmation bias, and shallow analysis. |
Before stating any conclusion, argue against it. If it survives, state it. If it doesn't, investigate further.
Apply the devil's advocate pattern before any claim:
# BAD — first match, no challenge
"The timeout is caused by the database query. I'll add an index."
# GOOD — devil's advocate applied
"My initial conclusion is the timeout is caused by the database query —
it takes 3.2s in the slow query log.
However, the network trace shows 1.8s of that is connection pool
exhaustion, not query execution. Adding an index wouldn't help if the
pool is starved. The connection pool is sized at 5 but there are 20
concurrent requests.
Revised conclusion: the root cause is connection pool sizing, not
query performance. I'll increase the pool size and re-measure before
considering an index."
| Rationalization | Reality |
|---|---|
| "This is obvious" | Obvious conclusions are where blind spots hide. Challenge it anyway. |
| "I already considered alternatives" | If you didn't state them, you didn't consider them rigorously. Make it visible. |
| "It'll slow me down" | A wrong conclusion wastes more time than 30 seconds of self-challenge. |
| "I found strong evidence" | Strong evidence for one theory doesn't rule out other theories. Look for disconfirming evidence. |
| "There's only one possible explanation" | That belief is itself the bias. There is almost never only one explanation. |
| "I'll verify after implementing" | Implementing the wrong fix and then discovering it wastes far more time. Challenge now. |
| "The user wants a quick answer" | The user wants a correct answer. Speed without accuracy is waste. |
| "I checked multiple sources" | Multiple sources agreeing doesn't mean you asked the right question. Did you search for why you might be wrong? |