fix(routing): route short greetings to simple tier when tools attached

[brunobuddy]

Summary

• The scorer's short-message fast path was gated on !hasTools, so personal AI agents like OpenClaw (which always send a tools array) skipped it entirely and fell into full scoring.
• Full scoring returns a tiny raw score for a one-word prompt, but applyMomentum assigns a very high weight to very short messages (~0.56 for a 4-char prompt), so any session that recently saw complex traffic pulled hi / hola up to complex.
• Dropping the !hasTools gate lets short, non-technical prompts short-circuit to simple before momentum can inflate them. The hasComplexSignal check added in routing: short non-tool prompts under 50 chars bypass scoring and always route simple #1172 still forces full scoring for short technical prompts like Debug this function.

Test plan

• npx jest src/scoring — 268 passed
• npx jest (full backend unit suite) — 3535 passed
• npx tsc --noEmit (backend) — clean
• npm run test:e2e --workspace=packages/backend — 107 passed (updated routing-flow.e2e-spec.ts tools-floor case to use a long message so it still exercises applyTierFloors)
• npm test --workspace=packages/frontend — 2110 passed
• npx tsc --noEmit (frontend) — clean
• scoring/index.ts stays at 100% line and branch coverage
• Manual: send hola through OpenClaw in cloud mode, confirm the message log shows simple / short_message instead of complex / momentum

Notes

• packages/backend/src/routing/resolve.service.spec.ts and packages/backend/test/routing-flow.e2e-spec.ts had tests that explicitly asserted tools force non-simple on a short message. Both were switched to a long message so they still exercise the applyTierFloors tools-floor branch instead of the fast path.
• Non-English greetings (hola, bonjour, etc.) are intentionally not added to simpleIndicators — the fix here is in the fast-path gate, not the keyword list.
• Broad codeGeneration / technicalTerms keyword lists are untouched.

---

Summary by cubic

Route short greetings to the simple tier even when tools are attached, preventing momentum from promoting one‑word prompts to complex. Short technical prompts still go through full scoring via the existing complex‑signal check.

• Bug Fixes
  • Removed the !hasTools gate from the short-message fast path in scoreRequest, so non-technical short prompts short-circuit to simple.
  • Updated unit/e2e tests to cover the new behavior and use longer messages when testing the tools-floor path.

^{Written for commit 43590ad. Summary will update on new commits.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.94%. Comparing base (bd60731) to head (43590ad).
⚠️ Report is 6 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1561   +/-   ##
=======================================
  Coverage   97.94%   97.94%           
=======================================
  Files         118      118           
  Lines        8614     8614           
  Branches     3240     3240           
=======================================
  Hits         8437     8437           
  Misses        175      175           
  Partials        2        2           

Flag	Coverage Δ
frontend	97.93% <ø> (ø)
shared	100.00% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[cubic-dev-ai]

No issues found across 6 files