Summary

Several critical files have very deep import chains that create tight coupling across subsystems. This makes refactoring risky and increases the blast radius of changes.

Worst Offenders

src/agents/pi-embedded-runner/run/attempt.ts (915 lines, 50+ imports)

This file imports from nearly every subsystem:

• auto-reply (heartbeat)
• config (channel capabilities)
• infra (machine name)
• media (constants)
• plugins (hook runner)
• routing (session key)
• signal (reaction level)
• telegram (inline buttons, reaction level)
• tts (system prompt hint)
• utils (user path, message channel, provider utils)
• Plus 30+ imports from within agents/

src/discord/monitor/message-handler.process.ts — 30 imports

src/discord/monitor/provider.ts — 29 imports

src/discord/monitor/message-handler.preflight.ts — 25 imports

Proposed Solution

Introduce a dependency injection pattern — a RunContext or AgentContext object that aggregates resolved config and services, passed into the runner rather than importing directly from every subsystem.

// Instead of 50 direct imports:
interface RunContext {
  config: ResolvedConfig;
  heartbeat: HeartbeatService;
  channels: ChannelCapabilities;
  media: MediaConfig;
  tts: TtsConfig;
  hooks: HookRunner;
  // ...
}

async function runAttempt(ctx: RunContext, params: AttemptParams) {
  // Use ctx instead of importing from 15 different directories
}

This would:

1. Make dependencies explicit and testable
2. Reduce import depth from 3-4 levels (../../../auto-reply/heartbeat.js)
3. Enable easier mocking in tests
4. Reduce the blast radius of subsystem refactors

Context

Found during a codebase review of v2026.2.3. This is a longer-term architectural improvement that would pair well with the file size refactoring (#11512).