[sl0thentr0py]

Summary

Adds RubyLLM agent identity and agent-run tracing.

• Tags chats through the shared RubyLLM::Agent.apply_configuration path, covering plain chats and Rails-backed construction/find flows.
• Creates one gen_ai.invoke_agent span per outer operation across complete, generate, run_tools, and step.
• Suppresses duplicate wrappers for the same chat while preserving genuinely nested agents.
• Propagates gen_ai.agent.name to model and tool descendants.
• Finishes spans on success, approval pauses, and errors without keeping runs open across jobs.

Testing

• Covers nested agents, duplicate suppression, model/tool descendants, return values, errors, and scope restoration.
• Validates RubyLLM 2.0.0 and current 2.x with Ruby 3.1 and 4.0.
• Runs the complete sentry-agents specs and lint.

Issues

• Part of: RubyLLM integration #3096
• Part of: RUBY-217

[cursor]

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

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 18541b7. Configure here.}

[cursor]

Agent run depth never resets

High Severity

The ensure in with only restores depth when depth is already positive, so the outer wrapper never clears @sentry_agents_agent_run_depth. After the first run, later calls on the same chat skip gen_ai.invoke_agent spans, including follow-up turns on a reused agent chat.

 

^{Reviewed by Cursor Bugbot for commit 18541b7. Configure here.}

[linear-code]

RUBY-217

[crmne]

Please build this on the new agent.ruby_llm event instead of patching Agent.apply_configuration and Chat. It also fixes two bugs in the current approach: after the first turn on a chat no invoke_agent span is emitted (the depth counter never resets, as Bugbot says), and Rails-backed agents from Agent.create and find get no agent span at all.