Correct system-architect specs from live 2026-10-05 inspection

- .13: KDE Plasma 6 (not Niri); RAM 15GB; disk letters corrected
  (root is /dev/sda2, storage sdb2, data sdc2, ubuntu_storage sdd1);
  add sda3 swap (81% full)
- .13 GPU: GTX 760 dead, no driver bound, software rendering
- .27: 4 cores/8 threads, Quadro P620 + HD 630, add swap + NTFS disk
- add Resource Pressure on .13 section and must-stay-on list
- note dead stacks (airflow 24GB logs, engram container) and caveats

agents/image-maker.md

@@ -68,13 +68,22 @@ The image comes back as base64 in `choices[0].message.images[0].image_url.url`.
 
 ## API Key
 
-`OPENROUTER_API_KEY` lives in **`/home/sam/.secrets`** (a plain `KEY=value` file). Load it with:
+`OPENROUTER_API_KEY` must be present as an **environment variable**. Do not assume a fixed file path — the canonical location differs per machine.
+
+**Canonical source on all machines:** `~/.config/environment.d/10-secrets.conf` (plain `KEY=value`, no `export`). This is the only location that **systemd user services read**, which is what matters when this subagent runs under Paperclip rather than in an interactive shell.
+
+Check availability first:
 
 ```bash
-set -a; . /home/sam/.secrets; set +a
 echo "${OPENROUTER_API_KEY:+key loaded}"
 ```
 
-It is NOT in `~/.config/environment.d/10-secrets.conf` — older versions of this file claimed that and the call failed.
+If it is empty, source it for the current shell:
+
+```bash
+set -a; . ~/.config/environment.d/10-secrets.conf; set +a
+```
+
+**Do not hardcode other paths.** `.13` additionally keeps keys in `~/.secrets` (an `export`-style file), but `.27` has no `~/.secrets` at all — anything that sources that path fails there. Prefer the environment variable; fall back to `~/.secrets` only if the variable is unset *and* the file exists.
 
 Write detailed, specific prompts. Save images to the user's current working directory or a specified path. Tell the user where you saved the file.

models.json

@@ -63,6 +63,17 @@
       "api": "omni-prompt-tools",
       "apiKey": "sk-fff5c0d55c0bc0bf-b58ed0-76da5716",
       "models": [
+        {
+          "id": "default-opencode-go-ds-flash",
+          "name": "Default Opencode Go Ds Flash",
+          "api": "omni-prompt-tools",
+          "input": [
+            "text"
+          ],
+          "contextWindow": 1000000,
+          "maxTokens": 384000,
+          "reasoning": true
+        },
         {
           "id": "aug/claude-haiku-4.5",
           "name": "Claude Haiku 4.5",
@@ -642,17 +653,6 @@
           "maxTokens": 8192,
           "reasoning": true
         },
-        {
-          "id": "default-opencode-go-ds-flash",
-          "name": "Default Opencode Go Ds Flash",
-          "api": "omni-prompt-tools",
-          "input": [
-            "text"
-          ],
-          "contextWindow": 1000000,
-          "maxTokens": 384000,
-          "reasoning": true
-        },
         {
           "id": "deepseek/deepseek-v4-flash",
           "name": "V4 Flash",
@@ -662,7 +662,13 @@
           ],
           "contextWindow": 1000000,
           "maxTokens": 384000,
-          "reasoning": true
+          "reasoning": true,
+          "cost": {
+            "input": 0.07,
+            "output": 0.28,
+            "cacheRead": 0.014,
+            "cacheWrite": 0.07
+          }
         },
         {
           "id": "deepseek/deepseek-v4-pro",
@@ -673,7 +679,13 @@
           ],
           "contextWindow": 1000000,
           "maxTokens": 384000,
-          "reasoning": true
+          "reasoning": true,
+          "cost": {
+            "input": 0.435,
+            "output": 0.87,
+            "cacheRead": 0.0036,
+            "cacheWrite": 0.435
+          }
         },
         {
           "id": "ds/deepseek-v4-flash",
@@ -1975,7 +1987,13 @@
           ],
           "contextWindow": 1048576,
           "maxTokens": 65535,
-          "reasoning": true
+          "reasoning": true,
+          "cost": {
+            "input": 0.3,
+            "output": 2.5,
+            "cacheRead": 0.03,
+            "cacheWrite": 0.083333
+          }
         },
         {
           "id": "openrouter/google/gemini-2.5-flash-image",
@@ -2083,7 +2101,13 @@
           ],
           "contextWindow": 1048576,
           "maxTokens": 65536,
-          "reasoning": true
+          "reasoning": true,
+          "cost": {
+            "input": 0.25,
+            "output": 1.5,
+            "cacheRead": 0.025,
+            "cacheWrite": 0.083333
+          }
         },
         {
           "id": "openrouter/google/gemini-3.1-flash-lite-image",
@@ -5292,4 +5316,4 @@
       ]
     }
   }
-}
+}

scripts/omni-pricing.mjs

@@ -0,0 +1,164 @@
+#!/usr/bin/env node
+/**
+ * Inject per-model pricing into ~/.agents/models.json for the `omni` provider.
+ *
+ * WHY THIS EXISTS
+ * ---------------
+ * `pi --list-models` / the omni extension's /omni sync regenerates models.json
+ * from OmniRoute and writes entries WITHOUT a `cost` field. pi's calculateCost()
+ * then reports $0.00 for every run, so Paperclip's cost events and budgets all
+ * read zero even though real money is being spent.
+ *
+ * The fix is to add { cost: { input, output, cacheRead, cacheWrite } } (USD per
+ * million tokens) to the model entries we actually use.
+ *
+ * Because /omni sync rewrites the whole file, this script must be re-run after
+ * every sync. That is the whole reason it is a script and not a one-off edit.
+ *
+ * PRICE SOURCES (both authoritative, no guessing)
+ *   - deepseek/*      -> OmniRoute /api/pricing, group "deepseek"
+ *   - openrouter/*    -> OpenRouter /api/v1/models (strip the "openrouter/" prefix)
+ *
+ * Model ids that use routing prefixes (aug/, ds/, oc/, tllm/, opencode-go/) do
+ * NOT map cleanly to pricing groups and are deliberately skipped — add them to
+ * MANUAL below if you need them.
+ *
+ * USAGE
+ *   node omni-pricing.mjs --dry-run     # show what would change
+ *   node omni-pricing.mjs               # write models.json
+ */
+
+import fs from "node:fs";
+import path from "node:path";
+
+const MODELS_JSON = path.join(process.env.HOME, ".agents", "models.json");
+const OMNIROUTE_PRICING = "http://192.168.20.13:20128/api/pricing";
+const OPENROUTER_MODELS = "https://openrouter.ai/api/v1/models";
+
+// Models we actually assign to workers. Keep this list short and deliberate.
+const TARGETS = [
+  "deepseek/deepseek-v4-flash",
+  "deepseek/deepseek-v4-pro",
+  "openrouter/google/gemini-3.1-flash-lite",
+  "openrouter/google/gemini-2.5-flash",
+];
+
+// Hand-priced entries: id -> {input, output, cacheRead, cacheWrite}
+const MANUAL = {};
+
+const dryRun = process.argv.includes("--dry-run");
+
+function omniKey() {
+  const cfg = JSON.parse(fs.readFileSync(MODELS_JSON, "utf8"));
+  return cfg.providers?.omni?.apiKey ?? "";
+}
+
+/** Fetch OmniRoute's own price table: { group: { model: {input,output,cached,cache_creation} } } */
+async function fetchOmniRoutePricing() {
+  const res = await fetch(OMNIROUTE_PRICING, {
+    headers: { Authorization: `Bearer ${omniKey()}` },
+  });
+  if (!res.ok) throw new Error(`OmniRoute pricing HTTP ${res.status}`);
+  return res.json();
+}
+
+/** Fetch OpenRouter prices keyed by bare model id (no "openrouter/" prefix). */
+async function fetchOpenRouterPricing() {
+  const res = await fetch(OPENROUTER_MODELS);
+  if (!res.ok) throw new Error(`OpenRouter models HTTP ${res.status}`);
+  const { data } = await res.json();
+  const out = {};
+  for (const m of data) {
+    const p = m.pricing ?? {};
+    out[m.id] = {
+      input: Number(p.prompt ?? 0) * 1e6,
+      output: Number(p.completion ?? 0) * 1e6,
+      cacheRead: Number(p.input_cache_read ?? 0) * 1e6,
+      cacheWrite: Number(p.input_cache_write ?? 0) * 1e6,
+    };
+  }
+  return out;
+}
+
+const r6 = (n) => Math.round(Number(n) * 1e6) / 1e6;
+
+/** Resolve a pi model id to a {input,output,cacheRead,cacheWrite} price, or null. */
+function resolvePrice(id, omniPricing, orPricing) {
+  if (MANUAL[id]) return MANUAL[id];
+
+  // deepseek/<name> -> OmniRoute group "deepseek"
+  if (id.startsWith("deepseek/")) {
+    const name = id.slice("deepseek/".length);
+    const row = omniPricing?.deepseek?.[name];
+    if (row) {
+      return {
+        input: r6(row.input),
+        output: r6(row.output),
+        cacheRead: r6(row.cached ?? 0),
+        cacheWrite: r6(row.cache_creation ?? 0),
+      };
+    }
+  }
+
+  // openrouter/<bare-id> -> OpenRouter catalogue
+  if (id.startsWith("openrouter/")) {
+    const bare = id.slice("openrouter/".length);
+    if (orPricing[bare]) {
+      const p = orPricing[bare];
+      return { input: r6(p.input), output: r6(p.output), cacheRead: r6(p.cacheRead), cacheWrite: r6(p.cacheWrite) };
+    }
+  }
+
+  return null;
+}
+
+const cfg = JSON.parse(fs.readFileSync(MODELS_JSON, "utf8"));
+const models = cfg.providers?.omni?.models;
+if (!Array.isArray(models)) {
+  console.error("models.json: providers.omni.models is not an array — aborting");
+  process.exit(1);
+}
+
+const [omniPricing, orPricing] = await Promise.all([
+  fetchOmniRoutePricing(),
+  fetchOpenRouterPricing(),
+]);
+
+const byId = new Map(models.filter((m) => m && m.id).map((m) => [m.id, m]));
+let changed = 0;
+const missing = [];
+
+for (const id of TARGETS) {
+  const entry = byId.get(id);
+  if (!entry) {
+    missing.push(id);
+    continue;
+  }
+  const price = resolvePrice(id, omniPricing, orPricing);
+  if (!price) {
+    missing.push(id);
+    continue;
+  }
+  const current = JSON.stringify(entry.cost ?? null);
+  const next = JSON.stringify(price);
+  if (current !== next) {
+    entry.cost = price;
+    changed++;
+    console.log(
+      `${dryRun ? "[dry-run] " : ""}${id}  ->  in $${price.input}  out $${price.output}  cacheR $${price.cacheRead}  cacheW $${price.cacheWrite}`,
+    );
+  } else {
+    console.log(`${id}  ->  already priced, unchanged`);
+  }
+}
+
+if (missing.length) console.log(`\nnot found / no price (skipped): ${missing.join(", ")}`);
+
+if (dryRun) {
+  console.log(`\ndry-run: ${changed} entr(y/ies) would change`);
+} else if (changed) {
+  fs.writeFileSync(MODELS_JSON, JSON.stringify(cfg, null, 2) + "\n");
+  console.log(`\nwrote ${MODELS_JSON} (${changed} changed)`);
+} else {
+  console.log("\nnothing to do");
+}

skills/system-architect/SKILL.md

@@ -13,56 +13,83 @@ metadata:
       hardware:
         cpu:
           model: "Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz"
-          cores: 8
+          cores: 4
           threads: 8
         ram:
-          total_gb: 62.7
+          total_gb: 62
         gpu:
-          - i915
+          - Intel HD Graphics 630 (i915) — drives the 4 display outputs
-          - nvidia
+          - NVIDIA Quadro P620 2GB (nvidia 580.173.02)
         disks:
           - device: /dev/nvme0n1p2
             size: 869G
             mountpoint: /
+            note: "ext4, 66% used (2026-10-05)"
+          - device: /dev/nvme0n1p3
+            size: 69G
+            mountpoint: "[SWAP]"
           - device: efivarfs
             size: 256K
             mountpoint: /sys/firmware/efi/efivars
           - device: /dev/nvme0n1p1
             size: 1022M
             mountpoint: /boot
+          - device: /dev/sda2
+            size: 224G
+            mountpoint: "(not mounted)"
+            note: NTFS
+      verified: 2026-10-05
     - name: nixos-desktop
       type: Desktop
       ip: 192.168.20.13
-      os: "NixOS, Niri"
+      os: "NixOS 26.11 (Zokor), KDE Plasma 6 / KWin Wayland (SDDM)"
-      role: NixOS with Niri. Second dev. Web facing. Network scripts.
+      role: Always-on web-facing server. Docker host + native services. Memory-bound — do not treat as CPU-bound.
-      tech: Docker (Mosquitto, N8N, Pi Hole back up with nebulasync from sam-ubuntu1 pihole, pocketbase, speech_piper, voice_bridge), websites at /var/www/, snapcast, gstreamer, librespot, mopidy
+      tech: Docker (OmniRoute, Outline, n8n, Pi-hole replica + nebulasync, Mosquitto, voice_bridge/voice_whisper, speech_piper, pocketbase, family-home-lab, dsh, worldmonitor, prefect, litellm, headroom/JEV), Caddy :8000 serving /var/www, snapcast, librespot, mopidy, native PostgreSQL :5433 (paperclip), paperclipai, Paseo Daemon, where-woof
       hardware:
         cpu:
           model: "AMD Ryzen 5 5600 6-Core Processor"
-          cores: 12
+          cores: 6
           threads: 12
         ram:
-          total_gb: 15.5
+          total_gb: 15
+          note: "6.8 GiB used; swap 7.1 GiB of 8.8 GiB used (81%) - 2026-10-05"
         gpu:
+          - "NVIDIA GTX 760 (Kepler, PCI 10de:11c2) - DEAD. No driver bound; only /dev/dri/card0 simple-framebuffer. /proc/driver/nvidia absent. Desktop renders via software (llvmpipe). hardware.nvidia.open=true can never work on Kepler (open modules need Turing+). Recommended replacement: AMD RX 6600 (in-kernel amdgpu)."
         disks:
-          - device: /dev/sdb2
+          - device: /dev/sda2
             size: 907G
             mountpoint: /
-          - device: efivarfs
+            note: "ext4, 459G used (54%) - 2026-10-05"
-            size: 128K
+          - device: /dev/sda3
-            mountpoint: /sys/firmware/efi/efivars
+            size: 8.8G
-          - device: /dev/sdb1
+            mountpoint: "[SWAP]"
+            note: "81% full - the main resource problem on this machine"
+          - device: /dev/sda1
             size: 1022M
             mountpoint: /boot
+          - device: /dev/sdb2
+            size: 916G
+            mountpoint: /mnt/storage
+            note: empty spare
+          - device: /dev/sdc2
+            size: 1.8T
+            mountpoint: /mnt/data
+            note: "21% used"
           - device: /dev/sdd1
             size: 2.7T
             mountpoint: /mnt/ubuntu_storage_3TB
-          - device: /dev/sda2
+            note: "USB drive - photo master + Borg backup target for .27 and .13. Letter shifts on replug (has been sdd1/sde1)."
-            size: 1.9T
+          - device: /dev/sde1
-            mountpoint: /mnt/data
+            size: 1.4T
-          - device: /dev/sdc2
+            mountpoint: "(not mounted)"
-            size: 932G
+            note: MaxtorBackup, old migration tarballs
-            mountpoint: /mnt/storage
+      caveats:
+        - "Device letters are NOT stable - NixOS mounts by UUID. Always check lsblk/df before using a device path."
+        - "GPU is dead: no hardware acceleration. KDE session costs ~2.4 GB because of software rendering plus 28-day uptime."
+        - "litellm and headroom are idle and almost entirely swapped out (~1.05 GB of swap combined)."
+        - "Dead stacks to ignore: airflow (7 containers exited, 24 GB of logs in /home/sam/deployment/airflow/logs), engram container (never started - the live engram is a user systemd service on :7437 and its DB is empty), trigger_dev, t3_stack_react, sams-home-network."
+        - "There is a SECOND airflow install at /home/sam/deployment/ai-resume/airflow/ - also not running."
+      verified: 2026-10-05
     - name: sam-ubuntu1
       type: virtual machine
       ip: 192.168.20.35
@@ -184,6 +211,28 @@ Sam's NixOS multi-machine infrastructure reference.
 
 Full hardware specs are in the frontmatter metadata.
 
+## Resource Pressure on .13 (verified 2026-10-05)
+
+`.13` is **memory-bound, not CPU-bound**. Load average is ~0.27 on 12 threads, but the
+8.8 GB swap partition is **81% full**. Before proposing anything that adds load to `.13`,
+check the swap. Full detail: Obsidian → `300 areas/360 Dev-Ops Network Computers/Resource Use — .13 CPU & RAM.md`.
+
+| Consumer | RAM + swap | Notes |
+|---|---|---|
+| KDE Plasma session (~30 processes) | ~2.4 GB | Software rendering (GPU dead) + 28-day uptime |
+| Docker containers (all) | ~3.5 GB RSS | `voice_whisper` 877 MB is the largest |
+| `litellm` + `headroom` | ~1.05 GB *swap* | Idle; both fully paged out; redundant with OmniRoute |
+| Dead Airflow logs | 24 GB *disk* | `/home/sam/deployment/airflow/logs` — nothing depends on it |
+
+**Must stay on `.13`:** Pi-hole `:53`, OmniRoute `:20128/20129` (all pi LLM traffic),
+Mosquitto `:1883` (Home Assistant on `.30`), the Borg backup target on `/mnt/ubuntu_storage_3TB`
+(`.27` pushes daily at 04:00), Caddy `:8000` + `/var/www`.
+
+**Corrections applied 2026-10-05:** `.13` runs **KDE Plasma 6**, not Niri. `.13` disk device
+letters were wrong (root is `/dev/sda2`). Home Assistant is on **`.30`, not `.35`** (`.35` is
+`sam-ubuntu1`, the Caddy/docker app host). Pi-hole primary is **`.35`**; `.13` is the replica
+(confirmed from `nebulasync`: `PRIMARY=http://192.168.20.35`, `REPLICAS=http://192.168.20.13`).
+
 ## Pi Agent (uniform across machines)
 
 - **Binary**: nixpkgs `pi-coding-agent` 0.82.1 in `home.packages` on .27/.13/.51 (Nix-managed; npx wrapper + npm-global install removed 2026-08-05). Updates = nixpkgs flake bump + rebuild.