Prompt Engineer
AI agents and tool calling questions
40 questions on AI agents and tool calling for prompt engineers, easiest first.
Easy 5
Medium 17
- One huge prompt that failsmedium
- Writing a tool descriptionmedium
- When a tool call failsmedium
- An agent that never stopsmedium
- A refund larger than the ordermedium
- A web page that talks to the agentmedium
- Credentials inside tool resultsmedium
- An agent with more rights than the usermedium
- Making the model always call a toolmedium
- Think, act, observemedium
- Too little detail to answermedium
- A report too long for one replymedium
- Looking up three cities one by onemedium
- Tool output filling the contextmedium
- Choosing among 200 categoriesmedium
- Bad arguments to a toolmedium
- An agent with no finish linemedium
Hard 18
- Agent picks the wrong toolhard
- Injection through a web pagehard
- An agent that deletes datahard
- Agent with 120 toolshard
- Errors flowing down a chainhard
- Long agent run fills contexthard
- The three dangerous ingredientshard
- Poisoned long-term memoryhard
- An agent stuck in a loophard
- Injection spreading between agentshard
- Users approving everythinghard
- Review in the same call or a new onehard
- An order ID nobody gavehard
- An agent losing its placehard
- Per-request tool lists and cachinghard
- Quality drops in long agent sessionshard
- Delegating a messy searchhard
- Where to put tool usage ruleshard