Prospectus Agent 0.3.0

[hannesvdc]

Per-business tuning, send/follow-up tracking from the CLI, and an internal tidy-up.

Per-profile settings

• Profiles can now carry a settings: block that overrides the shared .env
  tunables for that business only (precedence: profile settings → .env → default).
  So one business can run a different cadence than another — e.g. Reaction Studio at
  target_company_count: 5 while Open Numerics stays at 10. Any .env tunable is
  overridable (case-insensitive keys): max_per_sector, fit_score_threshold,
  max_discovery_calls, max_company_size, model/effort settings, etc. The API key
  stays env-only.

New CLI verbs

• prospectus-agent --sent — records that you've sent the drafts: marks every
  drafted company as sent, using each one's draft date as the contact date, so the
  follow-up clock starts correctly (including for drafts sent days ago). Pure database,
  no API calls; idempotent.
• prospectus-agent --followup — standalone follow-up sweep: lists every company
  past the no-reply threshold, drafts a follow-up for each, and writes them to the
  outbox — without running discovery.
• prospectus-status mark / the DB layer can now set an explicit historical contact
  date, not just "today".

Email

• Stronger, value-framed closing call-to-action. The ask is now purposeful
  ("…to see where we could help / create value for your team") instead of limp
  phrasings like "to see whether there's a fit".

Internals

• De-duplicated the codebase: a single strict-tool builder (llm.function_tool), a
  shared token-usage report (llm.usage_summary), and shared entrypoint setup
  (runner.open_session) used by the daily run, refine, and follow-up commands.

---

This discussion was created from the release Prospectus Agent 0.3.0.