refactor(libsy): source advisor-gate guards from shared ToolSignals counts

[eric-liu-nvidia]

Summary

Followup committed in the #371 review thread: the advisor gate's request-side guards (gate_stall_turns, gate_min_tool_results) now read the shared ToolSignals extraction instead of bespoke counters — one definition of tool/turn counting across algorithms.

• ToolSignals gains two buckets: tool_result_count (per ToolResult block, including empty-content results, matching the gate's previous counting) and assistant_turn_count (Role::Assistant messages; documented in contrast to turn_depth, which counts all messages regardless of role).
• AdvisorGate::route_inner calls ToolSignals::from_request once and reads both counts; the bespoke count_tool_results / assistant_turns helpers in advisor_gate/turn.rs are deleted.
• No behavior change: the two new gate e2e tests were written first and passed against the old helpers; after the swap, both suites re-ran green with zero test modifications.

Tests

• ToolSignals unit tests pin counting parity on both message shapes: one ToolResult per user message (OpenAI-chat normalized role: tool) and several blocks batched in one message (Anthropic) — plus empty-content results still counting, and the assistant_turn_count vs turn_depth contrast.
• Gate e2e: batched result blocks satisfy gate_min_tool_results per block; the stall checkpoint counts assistant turns, not messages.

Additive pub-struct fields only (no out-of-crate consumers). The trigger side (buffered-response inspection: has_tool_use, visible_text) is intentionally untouched and left for a separate followup.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Improved advisor gating accuracy when processing tool results and assistant responses.
  • Batched messages now count each tool result individually, including empty results.
  • Stall detection now counts assistant turns correctly rather than counting all conversation messages.
  • Threshold-based safeguards now respond more reliably to conversations with multiple tool calls or results.

• Tests

  • Added coverage for batched tool results, empty results, and varied conversation formats.

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://NVIDIA-NeMo.github.io/Switchyard/pr-preview/pr-591/
Built to branch gh-pages at 2026-09-01 20:41 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 9e2dc465-4128-462e-b33e-eb1da59db666

📥 Commits

Reviewing files that changed from the base of the PR and between 035d702 and 28c14bd.

📒 Files selected for processing (4)

• crates/libsy/src/algorithms/advisor_gate.rs
• crates/libsy/src/algorithms/advisor_gate/tests.rs
• crates/libsy/src/algorithms/advisor_gate/turn.rs
• crates/libsy/src/algorithms/util/tool_signals.rs

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

Walkthrough

ToolSignals now extracts assistant-message and raw tool-result counts. AdvisorGate::route_inner uses these shared signals for stall and minimum-tool-result thresholds. Tests cover batched and empty results, and assistant-turn counting.

Changes

Advisor gate signal extraction

Layer / File(s)	Summary
Conversation-shape signal extraction crates/libsy/src/algorithms/util/tool_signals.rs	ToolSignals now counts assistant messages and every ToolResult block, including batched and empty results. Tests cover the new counts.
AdvisorGate threshold integration crates/libsy/src/algorithms/advisor_gate.rs, crates/libsy/src/algorithms/advisor_gate/tests.rs, crates/libsy/src/algorithms/advisor_gate/turn.rs	route_inner uses shared counts for stall and minimum-tool-result checks. Tests cover both thresholds. Obsolete local helpers and imports are removed.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Merge Risk: ⚪ Minimal · up to 28c14

This localized refactor preserves the existing gating behavior and introduces no actionable merge-blocking risk; it is merge-ready after normal checks and review.

Poem

A rabbit reads each line,
The patch grows clear beneath the moon,
Small changes hop in place,
Tests guard the garden path,
Reviews bloom before the dawn.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change: refactoring advisor-gate guards to use shared ToolSignals counts.
Docstring Coverage	✅ Passed	Docstring coverage is 86.67% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 15 functions across 4 files.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

_{Comment @coderabbitai help to get the list of available commands.}