/fleet — what are my background agents doing?

🧒 When reporting to a non-technical operator: end with a child-simple "In plain words" recap in their language. (memory eli5-always)

The operator runs an autonomous fleet (Claude Desktop "Cowork" app → many headless claude.exe Code agents) building their Second Brain in the background. This answers: how many, what they're building, anything stuck/looping/burning. READ-ONLY — never kill a process without the operator's explicit go (killing mid-write loses uncommitted work; git holds only committed state).

🖥️ Dashboard first (the operator works visually)

python "$IMPORTS_ROOT/build_fleet_dashboard.py" → open $OBSIDIAN_VAULT/_Dashboards/Fleet-Dashboard.html: KPIs (agents / masters / files in flight / work commits / MCP) + a health flag 🟢/🟡/🔴 + a commit feed (real work vs auto-backups) + what is being written right now. READ-ONLY. The text below is for digging by hand.

Recipe (all read-only)

1. Who's running + the master:

   $cl = Get-CimInstance Win32_Process -Filter "Name='claude.exe'"
   "claude agents: $($cl.Count)"
   $cl.ParentProcessId | Sort-Object -Unique | ForEach-Object { $p = Get-CimInstance Win32_Process -Filter ("ProcessId="+$_) -EA SilentlyContinue; if ($p) { "parent $_ = $($p.Name) | $($p.CommandLine.Substring(0,[Math]::Min(80,$p.CommandLine.Length)))" } }
   

   Many children sharing ONE Claude.exe parent = the Desktop "Cowork" master; an explorer.exe grandparent = the operator launched it from the GUI.
2. What they built (committed): git -C "$OBSIDIAN_VAULT" log -20 --pretty="%cr | %s" — DESCRIPTIVE messages are the fleet's work; terse pre-intervention are auto-backups. Group by theme.
3. Writing right now: git -C "$OBSIDIAN_VAULT" status --short → count = files in flight this second (re-run after ~30s; growing = actively writing).
4. New reusable artifacts: python "$IMPORTS_ROOT/retro_inventory.py" 1 → read its digest (new skills/scripts/notes).
5. MCP load (optional): count python.exe … telegram-mcp … main.py processes (~2 per agent) — high = many live sessions eating memory.
6. Collision lens — who's in which FILE right now (session_monitor.py, merged in here 2026-06-13): python "$USERPROFILE/!CLAUDE-HP17 May26\session_monitor.py" (snapshot) · --watch (live terminal) · --serve → http://127.0.0.1:8765 (browser, auto-refresh). Parses the session .jsonl logs for file-level touches and flags ⚠️ COLLISIONS (one file edited by 2+ sessions within 15 min) + hot files. Steps 1–5 answer "what are they building / stuck / burning"; this answers "are any about to overwrite each other". Run it before any structural / bulk vault edit when sessions run hot. Same fleet, two lenses — /fleet is the front door. (memory session-monitor)

Flags to surface

• 🟢 healthy = commits advancing through DIFFERENT real themes over time.
• 🟡 stuck/looping = many agents but no NEW descriptive commits for a long stretch, or the same commit message repeating.
• 🔴 burn = dozens of lingering agents / MCP procs with no progress → tell the operator; HE decides whether to stop the Desktop app (you don't kill it).

Output

Tight one-liner + detail: "N agents (master PID …) · building: <themes> · writing now: <k> files · flags: 🟢/🟡/🔴 …". Then 🧒 recap. If you spotted something durable they built that isn't captured, mention it (or note /retro / the daily sweep will catch it).

<!--kit-footer-->

---

Like this skill? It is part of second-brain-starter-kit: the second brain we built for ourselves and run every day at Palo Alto AI Research Lab. Install the whole set with npx skills add tonydzi/second-brain-starter-kit. Everything is open source and free, so take what you need.

Flagships worth a look on their own: secondop-panel (a second opinion from a panel of external models), claude-memory-tidy (stop your agent's memory from rotting), telegram-mcp-kit (your own Telegram over MCP in about 15 minutes).

Author: Anton Dziatkovskii, Palo Alto AI Research Lab. Telegram @tonydzi - WhatsApp +1 341 222 9178 - X @Tony_Stef_

Engineers: want to test-drive this setup? Message me. I hand out free starter seeds to engineers who test and report back, and custom skill requests are welcome.