Prompts for coding agents

Use Firedrill. List my workspaces and run `firedrill status --json`. Tell me the
workspace, credential store and drivers available on this machine. Do not create
anything.

Read this repository and describe the AI agent for a Firedrill simulation. Tell me:
- how it starts and which port or process it uses
- how a user talks to it (HTTP, stdin, a function, a web page or a webhook)
- every external API, SDK client and MCP server it calls
- what state it keeps (memory, files, databases, caches, sessions)
- which configuration names it needs, and which are secrets
Do not change any files.

Write a Firedrill descriptor and local plan for this agent. Pick the driver profile
that matches how users talk to it. Keep both files outside the repository. Show me
both files and every command the plan will start. After I approve, run
`firedrill register` and put the printed digest in the plan.

Write a Firedrill persona, evaluation and scenario for this situation:
<describe the situation, for example "a customer asks for a refund on a damaged
order and objects when asked for a photo">.
The evaluation should check what a good answer does, in plain criteria. Save all
three with `firedrill library save` and show me the version IDs.

List the synthetic tools available with `firedrill tools runnable` and
`firedrill tools catalog`. Match them to the external services this agent calls.
For each match, show me the operations the agent uses and how the local plan will
route the agent to the simulated service. Do not save anything yet.

Save a Firedrill world for this agent with the scenarios we saved. Run each scenario
three times. Then run `firedrill plan --world <world-id>` and show me the review,
including any unresolved requirements and the credit estimate. Stop there.

I approve plan <plan-id>. Run `firedrill apply <plan-id> --approve --follow`, then
run the drills with `firedrill run`. Show me the commands before they start. When
the campaign settles, summarize the report.

Read the Firedrill report for campaign <campaign-id>. For each drill, tell me how it
ended and which evaluations were met, missed or not assessable. For anything missed,
quote the turns and tool calls that explain it. Do not rerun anything.

Based on the Firedrill report for campaign <campaign-id>, propose a fix in this
repository for the failures. After I approve the change, plan and run the same
world again and compare the new report with the old one.