SWEny, YAML workflows for AI agents I'm running in prod (triage, E2E)

Minor Changes

• 5053263: Add optional onEvent streaming callback to coding agent provider configs.

  • ClaudeCodeConfig, GoogleGeminiConfig, OpenAICodexConfig, MockCodingAgentConfig
    all accept onEvent?: AgentEventHandler
  • Claude provider uses --output-format stream-json for structured events (tool calls,
    tool results, text deltas, thinking blocks)
  • Gemini and Codex providers emit text lines as { type: "text" } events
  • New AgentEvent and AgentEventHandler types exported from @sweny-ai/providers/coding-agent
  • No breaking changes — onEvent is optional; omitting it preserves existing behaviour exactly

• 474589e: New observability providers: Prometheus and PagerDuty.

Patch Changes

• 6a71f2a: Harden coding agent providers: stderr capture, timeout support, error logging.
  • Extract spawnLines() helper to shared.ts — eliminates duplicate spawn code across
    Claude, Gemini, and Codex providers
  • spawnLines captures stderr and forwards each line to the logger (previously discarded)
  • Event handler errors are logged via the logger instead of silently swallowed
  • Process killed by signal resolves with exit code -1 and logs the signal name
  • Add timeoutMs to CodingAgentRunOptions — passed through to both event and non-event paths
  • Add timeoutMs to execCommand for non-event mode too
  • Mock agent now lets onEvent handler errors propagate (aids test debugging)