feat(mcp): answer-by-union for homonym symbols in get_answer

[RaghavChamadiya]

What

When a get_answer question names a symbol that has several definitions no qualifier can disambiguate (a helper duplicated across modules, an overloaded name), the tool now runs an exact-name index scan and inlines the union of every definition body in symbol_bodies, instead of returning a low-confidence list of candidate files.

It also adds a guard: a qualified name (Parent.method) that matches no indexed definition returns not-found rather than synthesizing from a same-named symbol in another file.

Why

The low-confidence candidate-file list was the wrong shape for an agent: handed a list of files, it drills back in with get_symbol on each one, or falls back to grep, to find the definition it asked about. For a name with N definitions that costs N follow-up calls. Inlining the bodies answers the question in a single call.

The not-found guard means a precise query (Foo.bar where bar exists only under other classes) never degrades into a confident answer built from the wrong definition.

How it works

• _anchor_symbol_hits now surfaces unresolvable homonyms and qualified-misses alongside the hits it injects.
• build_homonym_union_bodies renders the definition bodies greedily under a char budget in Read format (first always renders); any that overflow are listed as get_symbol pointers with a do-not-Read redirect.
• get_answer returns the union (grounding: "exact_symbol", confidence: "high") or the not-found response before the synthesis gate, so neither path emits the candidate-file list.

Validation

• New unit tests cover the union/qualified-miss split, the char-budget overflow, and the qualified-miss guard; the existing anchor tests are updated for the new return shape. Full MCP unit suite passes (388).
• Verified against a real index: a question naming a 4-definition helper returns all four bodies inlined with no candidate-file list. In a headless agent A/B on the same repo, the change cut total tool calls by 50-60% on homonym questions, with the best case answering in a single get_answer call.

[repowise-bot]

✅ Health: 7.7 (unchanged)

📋 At a glance
4 hotspots touched · 3 new findings introduced · 5 co-change pairs left out · 1 file with recent fix history · 1 dead-code finding.

Files & modules (2)

• packages (3 files)
  • .../tool_answer/config.py
  • .../tool_answer/answer.py
  • .../tool_answer/symbols.py
• tests (1 file)
  • .../mcp/test_answer_calibration.py

⚠️ Change risk: moderate _{(riskier than 52% of this repo's commits · raw 9.0/10)}
This change's risk is driven by:

• more lines added than baseline
• more scattered than baseline

🩹 Review priority _{(files here with the most recent bug-fix history — defects cluster, so review these first)}

• .../tool_answer/answer.py — fixed 2× in the last ~6 months

🔎 More signals (3)

🔥 Hotspots touched (4)

• .../tool_answer/config.py — 4 commits/90d, 5 dependents · primary owner: Swati Ahuja (94%)
• .../tool_answer/answer.py — 8 commits/90d, 2 dependents · primary owner: Swati Ahuja (66%)
• .../mcp/test_answer_calibration.py — 3 commits/90d, 1 dependents · primary owner: Raghav Chamadiya (60%)

1 more

• .../tool_answer/symbols.py — 4 commits/90d, 3 dependents · primary owner: Swati Ahuja (58%)

🔗 Hidden coupling (2 files)

• .../tool_answer/answer.py co-changes with these files (not in this PR):
  • .../mcp_server/tool_symbol.py (3× — 🟢 routine)
  • .../mcp_server/tool_search.py (3× — 🟢 routine)
  • .../mcp_server/_meta.py (2× — 🟢 routine)
  • .../templates/claude_md.j2 (2× — 🟢 routine)
• .../tool_answer/config.py co-changes with .../tool_answer/confidence.py (2× — 🟢 routine) — not in this PR.

💀 Dead code (1 finding)

• 💀 .../tool_answer/answer.py _json_default (confidence 0.65)

---

📊 Full report · ⭐ Star Repowise · 📥 Install bot · Last updated 2026-07-11 04:38 UTC
_{Silence on a single PR with [skip repowise] in the title · Per-repo toggle on repowise.dev/settings?tab=bot}