Prospectus Agent 1.0.0

[hannesvdc]

v1.0.0 A fully autonomous prospectus agent

Prospectus-Agent crosses from an agent you need to babysit to a fully autonomous daily outreach engine. Version 1.0.0 discovers prospects, writes tailored emails, sends them across multiple businesses, and threads its own follow-ups completely on its own. It still prints the emails, so you can review just a daily sample. Version 1.0 marks that it's stable enough to run in production, unattended. Run at your own risk.

One-command daily cycle

The complete agent can be run with just two commands:

prospectus-agent --runall && prospectus-agent --deliver --live

Explanation:
--deliver sends the entire queue: both new-prospect initials and due follow-ups, across every profile, in one call. --profile X scopes to one business;
--deliver --followup restricts to follow-ups; dry-run by default, --live to send.

• Every {profile, followup, deliver, live} combination now does something meaningful, or fails with a clear message.
• Follow-ups are uncapped (they go to people you've already contacted); the daily cap now applies only to cold initials.

You can run this command daily, or just create a bash script (loop) that automates even that.

Runs unattended without going off the rails

• Phase-isolated pipelin. a flaky API call or one bad company drops just that item; the run finishes, the follow-up sweep still happens, and the digest always gets written. No more a single hiccup stranding the whole batch.
• Follow-up digest fix — followups.md is rebuilt from the complete currently-due set, so --runall always produces it (previously it silently omitted
  follow-ups drafted on an earlier day).

Code health (1.0 cleanup)

• Single-source version: lives once in init.py; pyproject.toml reads it dynamically. (Fixes --version having been stuck at 0.2.0.)
• De-duplication — one Gmail-creds predicate, one vendor list + error message, shared outbox render helpers, and the follow-up state machine consolidated
  into one function.
• Type annotations across 9 modules, following the existing db.py convention.
• Dead code removed.

Quality

211 offline tests, all green. Sends only from the configured Workspace inbox, dry-run by default, with a no-double-send guard.

Next Steps

I'm thinking about creating a standalone application with GUI. Is this something you might find useful?

---

This discussion was created from the release Prospectus Agent 1.0.0.