feat: add live session signals

[ccf]

Summary

• add live local session signal analysis for friction, satisfaction, and abandonment risk
• expose a new MCP tool for polling live session signals during an active session
• mark the roadmap live-session-signals item shipped

Testing

• PYTHONPATH=src pytest tests/test_live_session_signals.py tests/test_mcp_tools.py tests/test_mcp_server.py -q
• ruff check src/primer/common/schemas.py src/primer/server/services/live_session_signal_service.py src/primer/mcp/tools.py src/primer/mcp/server.py tests/test_live_session_signals.py tests/test_mcp_tools.py tests/test_mcp_server.py
• ruff format --check src/primer/common/schemas.py src/primer/server/services/live_session_signal_service.py src/primer/mcp/tools.py src/primer/mcp/server.py tests/test_live_session_signals.py tests/test_mcp_tools.py tests/test_mcp_server.py

---

Note

Medium Risk
Adds a new MCP-exposed capability that analyzes in-progress local transcripts and derives risk/sentiment heuristics, which could affect user-facing guidance if the scoring or session selection is wrong. Changes are self-contained and covered by new unit tests, with no auth or persistence changes.

Overview
Adds live local session signal analysis that summarizes in-progress sessions for friction, sentiment, and abandonment risk, returning a LiveSessionSignalsResponse with structured signals, risk_level, and satisfaction_signal.

Exposes this via a new MCP tool live_session_signals (wired through primer_live_session_signals) that formats the live summary for repeated polling during a session, and updates tests/roadmap to reflect the shipped capability.

^{Written by Cursor Bugbot for commit 0fcccb6. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is ON, but it could not run because the spend limit has been reached. To enable Bugbot Autofix, raise your spend limit in the Cursor dashboard.}

[cursor]

Duplicated _EXECUTION_TOOL_NAMES risks silent divergence

Medium Severity

_EXECUTION_TOOL_NAMES is copy-pasted from execution_evidence_service.py into live_session_signal_service.py. The new file already imports _classify_status and extract_execution_evidence from that module, so it could import this constant too. If a new tool name is added to one copy but not the other, _count_tool_errors would silently drift from the execution-evidence logic, miscounting errors for the forgotten tool.