git:20260522.817c2c7 to git:20260618.80fba74

2 added, 2 removed. Audit A to A.

---

name: pan-resources

description: "Show RAM usage by agents, conversations, and system processes — model breakdown, workspace agents, orphan detection"

triggers:

- pan resources

- system resources

- ram usage

- memory usage

- how much ram

- agent memory

- resource usage

allowed-tools:

- Bash

- Read

---

# System Resource Inventory

- Run `pan resources` to get a full breakdown of RAM usage across all Panopticon-managed processes.

+ Run `pan resources` to get a full breakdown of RAM usage across all Overdeck-managed processes.

```bash

pan resources

```

This shows:

- **System RAM/swap** — total, used, available, with color-coded severity

- **Model breakdown** — count and total RAM per AI model (claude-opus, sonnet, kimi, mimo, etc.)

- **Workspace agents** — each running agent with issue ID, role (work/planning/review/test/merge), model, RAM, and start time

- **Conversations** — count, total RAM, oldest, and archival recommendation for old conversations

- - **Orphaned processes** — Claude processes NOT tracked by Panopticon (the real bug signal)

+ - **Orphaned processes** — Claude processes NOT tracked by Overdeck (the real bug signal)

- **Other heavy processes** — Chrome, Vite, Java, TTS, Playwright, etc. over 100 MB

## JSON output

For machine consumption:

```bash

pan resources --json

```

## Interpreting results

- **Orphaned processes** (red warning) mean a Claude process is running but not tracked by any conversation or workspace agent. This indicates a bug in lifecycle management.

- **Archival recommendation** shows how many conversations are older than today and how much RAM archiving them would free.

- The 3-4 agent ceiling on 64 GB RAM means you should keep workspace agents + conversations below ~15 GB total.