Lead enrichment
Scores each new sales lead from 0 to 100 from its website, writes a one-line summary, and tells the sales rep once.
- Written by our coding agent
- 4 automatic tests
The job
Read the new leads, open each company’s website, score how well it fits a buyer of developer tools, summarize what it does in one sentence, then send the sales rep one message.
The agent gets this description and the permissions on the right, nothing else. It never sees how “done” will be checked.
How it was built
Written from a task description
One attempt, kept unchanged. The checker leaned yes but wasn’t sure enough, so a person decided.
Review
The first version limited the score as if it were money, which would have let a third lead slip through. The score became a plain number range.
Automatic tests
Each test plays out a run with the agent's moves written in advance, instead of asking an AI model, and checks that capsule allows, stops or refuses each one correctly. They run every time capsule's code changes.
- Normal run
- Stackforge scores 92 and Hearth Bakery 3, each with a summary. The message to the rep waits for you and is allowed.
- Unlisted page
- A pricing page that isn’t on the list is refused. The agent carries on and finishes.
- Bad score
- A score of 150 is refused outright. Nobody is asked.
- Message denied
- The message to the rep waits for you and is denied. Nothing is sent.
What’s proved
These hold for every run, not just the ones above. Each is proved, and the proof is checked by computer.
- A number outside its allowed range is never accepted.
- Combining permissions never turns “ask me first” into “go ahead.”
What this doesn’t show
- This job hasn’t had a real run with an AI model yet. Its tests run on every code change.