- Old prompt: "Using https://docs.firecrawl.dev, integrate POST /webhooks/agent.started from Firecrawl's API into my codebase and test it out. Write a Python script."
- New prompt: "Write a Python script that listens for when a Firecrawl agent starts running and logs the event."
- Changes:
- Replaced endpoint path with the user's actual goal ("listens for when an agent starts")
- Leads with the deliverable instead of the docs URL
- Old prompt: "Using https://docs.firecrawl.dev, integrate GET /v2/team/token-usage/historical from Firecrawl's API into my codebase and test it out. Write a Python script."
- New prompt: "Write a Python script that pulls my team's historical token usage from Firecrawl and prints a summary."
- Changes:
- "Historical token usage" is how a human thinks about it — not the endpoint path
- Added a concrete output goal ("prints a summary") so the agent knows what "test it" means
- Old prompt: "Using https://docs.firecrawl.dev, integrate DELETE /v2/scrape/{jobId}/interact from Firecrawl's API into my codebase and test it out. Write a Python script."
- New prompt: "Write a Python script that ends an active interaction session on a Firecrawl scrape job. It should start a scrape, interact with it, then close the session and confirm it's gone."
- Changes:
- Describes the full workflow a human would actually need — nobody calls a DELETE in isolation
- "Close the session and confirm it's gone" replaces "test it out" with a verifiable outcome
- Old prompt: "Using https://docs.firecrawl.dev, integrate POST /v2/parse from Firecrawl's API into my codebase and test it out. Write a Python script."
- New prompt: "Write a Python script that uploads a document to Firecrawl and extracts the structured content from it."
- Changes:
- "Uploads a document and extracts structured content" describes what parse does from the user's perspective
- Old prompt: "Using https://docs.firecrawl.dev, integrate POST /v2/map from Firecrawl's API into my codebase and test it out. Write a Python script."
- New prompt: "Write a Python script that takes a URL and uses Firecrawl to discover all the pages on that site. Print the list of URLs it finds."
- Changes:
- "Discover all the pages on a site" is how someone describes mapping — the word "map" is Firecrawl-specific
- Input and output are both specified naturally
Real users don't preemptively hand an agent a docs link, they describe what they want and expect it to work.
Since these prompts are testing docs usability from a real user's perspective, the prompts should reflect how people actually talk to agents.