v4.3.0

What's New

MCP server improvements to reduce agent token overhead, based on benchmark analysis across 25 coding tasks.

Batch Show

• wonk_show now accepts comma-separated names (e.g. name="main,parse,validate") — turns 6 sequential show calls into 1.

Empty Result Diagnostics

• When wonk_show or wonk_sym return 0 results, the response includes diagnostic hints: fuzzy symbol name suggestions, advice to remove filters, and fallback to wonk_search.

Budget Defaults

• wonk_search default budget: 2000 → 4000 tokens (fits ~40-50 results)
• wonk_show default budget: unbounded → 4000 tokens (prevents runaway returns)

Truncation Hints

• All truncated responses now include an actionable hint field (e.g. "use Read(file, line, 80) for full file content") instead of a raw budget_limit number.

Task-Oriented Tool Descriptions

• Rewrote descriptions for all 10 core tools to guide agent discovery:
  • wonk_callpath: "Trace how symbol A reaches symbol B..."
  • wonk_context: "Get everything about a symbol in ONE call... Replaces chaining wonk_sym + wonk_callers + wonk_callees + wonk_ref."
  • wonk_flows: "Trace execution from an entry point through callees..."

Fuzzy Path Matching

• wonk_search paths parameter now accepts partial directory fragments (e.g. paths=["task"] resolves to src/task/ if it exists).

Full Changelog: v4.2.1...v4.3.0