Exercises — Module 02: Tools and tool use

2.1 — Description surgery

Find five tool descriptions in open-source harnesses. Rewrite them according to the rules in §2 and §3. For each, run 20 queries where the tool should be used and 20 where it should not. Measure the precision/recall before and after.

Deliverable: a table of before/after precision and recall, plus the five rewrites.

2.2 — One big vs. many small

Design a database_query capability two ways:

• One tool: database_query(sql: str).
• Many tools: list_tables, describe_table, select_rows, aggregate_by.

Implement both. Run the same 10 questions through each. Which wins? On what axes? Would your answer change for a non-technical user?

Deliverable: side-by-side results and a recommendation.

2.3 — Registry from scratch

Implement a tool registry with:

• JSONSchema validation.
• Per-tool timeouts.
• Structured error responses.
• Parallel dispatch.
• A stress test (100 concurrent calls; 30% of handlers fail).

Deliverable: code + stress-test output.

2.4 — MCP integration

Stand up one MCP server exposing one tool. Integrate it into your harness. Measure the latency overhead vs. in-process.

Deliverable: numbers and a short note on when the overhead is worth it.

2.5 — Anti-pattern hunt

Find a codebase (your own or open-source) that uses tools. Identify at least three anti-patterns from Module 02 §9. For each, propose a fix.

Deliverable: a short review document.