"""extend.py: make the capstone your own. Pick ONE extension and ship it (with an eval case).

The capstone agent (../solution/agent.py) already integrates agentic RAG, memory, observability,
cost, reliability, and security. Extend it into a portfolio piece. Ideas, easy to hard:

  1. Add a knowledge document to corpus.DOCS and an eval case in evals.py that needs it.
  2. Add a second SAFE tool (for example get_order_status(order_id)) and use it in an answer.
  3. Add a third hop to the knowledge base (who does the billing lead report to?) and confirm the
     agent searches three times.
  4. Persist memory per user with save_state / load_state (M21) so a returning user is remembered
     across API requests in app.py.
  5. Add a new risky tool (issue_refund) and gate it like send_email (approval + an allowlist of
     amounts), then add an eval case proving an over-limit refund is blocked.

The rule from M26: whatever you add, add an eval case for it first, then make it pass. Run the gate:
    python ../solution/evals.py
"""

print(__doc__)