Expose session working directory in daemon /list endpoint

[lucharo]

Summary

The daemon's /list endpoint only returns startedBy, happySessionId, and pid for each session. The working directory is available in memory (TrackedSession.spawnOptions.directory) but not exposed in the response.

Why

External tools (TUIs, scripts, dashboards) that list sessions need the working directory to:

1. Resume sessions correctly — claude --resume must run from the original directory since session files are stored per-project (~/.claude/projects/<encoded-cwd>/)
2. Display useful context — showing which project a session is working in
3. Avoid brittle workarounds — currently requires reading /proc/{pid}/cwd (Linux-only, fails for dead processes) or reverse-engineering Claude's project directory encoding

Current state

controlServer.ts /list response:

children: z.array(z.object({
  startedBy: z.string(),
  happySessionId: z.string(),
  pid: z.number()
}))

TrackedSession already has:

spawnOptions?: SpawnSessionOptions;  // contains directory: string

And the /session-started webhook receives metadata which may also contain the path.

Suggested change

Add directory (optional) to the /list response:

children: z.array(z.object({
  startedBy: z.string(),
  happySessionId: z.string(),
  pid: z.number(),
  directory: z.string().optional(),
}))

Populated from child.spawnOptions?.directory for daemon-spawned sessions, and from webhook metadata for terminal-started sessions.

🤖 Generated with Claude Code