fix: hermesRun command and tester verdict via llmExtract

- Fix hermes invocation: 'hermes -q' → 'hermes chat -q' with proper flags
- Replace fragile string.includes('PASS') with llmExtract judge
  (previous false positive: matched '--pass-session-id' in usage text)

小橘 🍊（NEKO Team）

nerve.yaml

@@ -10,7 +10,12 @@ senses:
     throttle: 10s
     timeout: 15s
     grace_period: null
-  unknown-sense:
+  linux-tcp-socket-stats:
+    group: system
+    throttle: 15s
+    timeout: 10s
+    grace_period: null
+  disk-usage-mounts:
     group: system
     throttle: 10s
     timeout: 15s
@@ -29,5 +34,8 @@ reflexes:
     sense: linux-system-health
     interval: 30s
   - kind: sense
-    sense: unknown-sense
-    interval: 60s
+    sense: linux-tcp-socket-stats
+    interval: 1m
+  - kind: sense
+    sense: disk-usage-mounts
+    interval: 1m

senses/disk-usage-mounts/index.js

@@ -0,0 +1,131 @@
+import { execSync } from "node:child_process";
+import { diskUsageMounts } from "./schema.ts";
+
+const DF_CMD =
+  "df -B1 --output=source,target,fstype,size,used,avail,pcent";
+
+/** fstype-based exclusions to avoid pseudo / volatile filesystem noise */
+const EXCLUDED_FSTYPES = new Set([
+  "tmpfs",
+  "devtmpfs",
+  "proc",
+  "sysfs",
+  "cgroup2",
+  "cgroup",
+]);
+
+function round2(n) {
+  return Math.round(n * 100) / 100;
+}
+
+function parseUIntField(s) {
+  if (s === "-") return null;
+  const n = Number.parseInt(String(s), 10);
+  if (!Number.isFinite(n) || n < 0) return null;
+  return n;
+}
+
+function parsePcent(tok) {
+  if (!tok || typeof tok !== "string") return null;
+  const t = tok.trim();
+  if (!/^[\d.]+%$/.test(t)) return null;
+  const raw = Number.parseFloat(t.replace("%", ""));
+  return Number.isFinite(raw) ? round2(raw) : null;
+}
+
+/**
+ * Parse one `df --output=source,target,fstype,size,used,avail,pcent` data line.
+ * Mount may contain spaces; last five logical columns are fixed.
+ */
+function parseDfLine(line) {
+  const parts = line.trim().split(/\s+/);
+  if (parts.length < 7) return null;
+
+  const pcentTok = parts[parts.length - 1];
+  const availBytes = parseUIntField(parts[parts.length - 2]);
+  const usedBytes = parseUIntField(parts[parts.length - 3]);
+  const totalBytes = parseUIntField(parts[parts.length - 4]);
+  const fstype = parts[parts.length - 5];
+  const device = parts[0];
+  const mount = parts.slice(1, parts.length - 5).join(" ");
+
+  if (!device || !mount || !fstype) return null;
+  if (totalBytes === null || usedBytes === null || availBytes === null)
+    return null;
+
+  const pcentFromDf = parsePcent(pcentTok);
+  const computed =
+    totalBytes > 0 ? round2((usedBytes / totalBytes) * 100) : 0;
+  let usedPercent = computed;
+  if (pcentFromDf !== null) {
+    const diff = Math.abs(computed - pcentFromDf);
+    usedPercent = diff > 1 ? computed : pcentFromDf;
+  }
+
+  return {
+    device,
+    mount,
+    fstype,
+    totalBytes,
+    usedBytes,
+    availBytes,
+    usedPercent,
+  };
+}
+
+function parseDfOutput(text) {
+  const rows = [];
+  const lines = text.split("\n");
+  for (const line of lines) {
+    const trimmed = line.trim();
+    if (!trimmed) continue;
+    if (/^Filesystem\s+/.test(trimmed) || trimmed.startsWith("Filesystem"))
+      continue;
+    const row = parseDfLine(line);
+    if (row && !EXCLUDED_FSTYPES.has(row.fstype)) rows.push(row);
+  }
+  return rows;
+}
+
+export async function compute(db, _peers) {
+  const ts = Date.now();
+  let mounts = [];
+
+  try {
+    const out = execSync(DF_CMD, {
+      encoding: "utf-8",
+      maxBuffer: 10 * 1024 * 1024,
+    });
+    mounts = parseDfOutput(out);
+  } catch {
+    mounts = [];
+  }
+
+  if (mounts.length > 0) {
+    await db.insert(diskUsageMounts).values(
+      mounts.map((m) => ({
+        ts,
+        device: m.device,
+        mount: m.mount,
+        fstype: m.fstype,
+        totalBytes: m.totalBytes,
+        usedBytes: m.usedBytes,
+        availBytes: m.availBytes,
+        usedPercent: m.usedPercent,
+      })),
+    );
+  }
+
+  return {
+    ts,
+    mounts: mounts.map((m) => ({
+      device: m.device,
+      mount: m.mount,
+      fstype: m.fstype,
+      totalBytes: m.totalBytes,
+      usedBytes: m.usedBytes,
+      availBytes: m.availBytes,
+      usedPercent: m.usedPercent,
+    })),
+  };
+}

senses/disk-usage-mounts/migrations/0001_init.sql

@@ -0,0 +1,14 @@
+-- Migration: 0001_init
+-- Creates the disk_usage_mounts table for disk-usage-mounts sense.
+
+CREATE TABLE IF NOT EXISTS disk_usage_mounts (
+  id            INTEGER PRIMARY KEY AUTOINCREMENT,
+  ts            INTEGER NOT NULL,
+  device        TEXT    NOT NULL,
+  mount         TEXT    NOT NULL,
+  fstype        TEXT    NOT NULL,
+  total_bytes   INTEGER NOT NULL,
+  used_bytes    INTEGER NOT NULL,
+  avail_bytes   INTEGER NOT NULL,
+  used_percent  REAL    NOT NULL
+);

senses/disk-usage-mounts/schema.ts

@@ -0,0 +1,13 @@
+import { integer, real, sqliteTable, text } from "drizzle-orm/sqlite-core";
+
+export const diskUsageMounts = sqliteTable("disk_usage_mounts", {
+  id: integer("id").primaryKey({ autoIncrement: true }),
+  ts: integer("ts").notNull(),
+  device: text("device").notNull(),
+  mount: text("mount").notNull(),
+  fstype: text("fstype").notNull(),
+  totalBytes: integer("total_bytes").notNull(),
+  usedBytes: integer("used_bytes").notNull(),
+  availBytes: integer("avail_bytes").notNull(),
+  usedPercent: real("used_percent").notNull(),
+});

senses/linux-tcp-socket-stats/index.js

@@ -0,0 +1,120 @@
+import { readFile } from "node:fs/promises";
+import { linuxTcpSocketStats } from "./schema.ts";
+
+const SOCKSTAT_PATH = "/proc/net/sockstat";
+const RAW_MAX = 4096;
+
+function parseInt10(s) {
+  const n = Number.parseInt(String(s), 10);
+  return Number.isFinite(n) ? Math.trunc(n) : NaN;
+}
+
+function parseSockstat(content) {
+  let socketsUsed = 0;
+  let tcpInuse = 0;
+  let tcpOrphan = 0;
+  let tcpTw = 0;
+  let tcpAlloc = 0;
+  let tcpMemPages = 0;
+  let parseOk = 1;
+
+  let socketsOk = false;
+  let tcpOk = false;
+
+  for (const line of content.split("\n")) {
+    const trimmed = line.trim();
+    if (trimmed.startsWith("sockets:")) {
+      const parts = trimmed.split(/\s+/);
+      const usedIdx = parts.indexOf("used");
+      if (usedIdx !== -1 && usedIdx + 1 < parts.length) {
+        const v = parseInt10(parts[usedIdx + 1]);
+        if (Number.isFinite(v)) {
+          socketsUsed = v;
+          socketsOk = true;
+        }
+      }
+    } else if (trimmed.startsWith("TCP:")) {
+      const parts = trimmed.split(/\s+/);
+      const map = {};
+      for (let i = 1; i + 1 < parts.length; i += 2) {
+        map[parts[i]] = parseInt10(parts[i + 1]);
+      }
+      const keys = ["inuse", "orphan", "tw", "alloc", "mem"];
+      if (keys.every((k) => Number.isFinite(map[k]))) {
+        tcpInuse = map.inuse;
+        tcpOrphan = map.orphan;
+        tcpTw = map.tw;
+        tcpAlloc = map.alloc;
+        tcpMemPages = map.mem;
+        tcpOk = true;
+      }
+    }
+  }
+
+  if (!socketsOk || !tcpOk) {
+    parseOk = 0;
+    socketsUsed = 0;
+    tcpInuse = 0;
+    tcpOrphan = 0;
+    tcpTw = 0;
+    tcpAlloc = 0;
+    tcpMemPages = 0;
+  }
+
+  return {
+    socketsUsed,
+    tcpInuse,
+    tcpOrphan,
+    tcpTw,
+    tcpAlloc,
+    tcpMemPages,
+    parseOk,
+  };
+}
+
+export async function compute(db, _peers) {
+  const ts = Date.now();
+  let rawSockstat = "";
+  let row = {
+    socketsUsed: 0,
+    tcpInuse: 0,
+    tcpOrphan: 0,
+    tcpTw: 0,
+    tcpAlloc: 0,
+    tcpMemPages: 0,
+    parseOk: 0,
+  };
+
+  try {
+    const content = await readFile(SOCKSTAT_PATH, "utf8");
+    rawSockstat =
+      content.length > RAW_MAX ? content.slice(0, RAW_MAX) : content;
+    row = parseSockstat(content);
+  } catch {
+    rawSockstat = "";
+  }
+
+  await db.insert(linuxTcpSocketStats).values({
+    ts,
+    socketsUsed: row.socketsUsed,
+    tcpInuse: row.tcpInuse,
+    tcpOrphan: row.tcpOrphan,
+    tcpTw: row.tcpTw,
+    tcpAlloc: row.tcpAlloc,
+    tcpMemPages: row.tcpMemPages,
+    parseOk: row.parseOk,
+    rawSockstat,
+  });
+
+  return {
+    ts,
+    socketsUsed: row.socketsUsed,
+    tcpInuse: row.tcpInuse,
+    tcpOrphan: row.tcpOrphan,
+    tcpTw: row.tcpTw,
+    tcpAlloc: row.tcpAlloc,
+    tcpMemPages: row.tcpMemPages,
+    parseOk: row.parseOk,
+    rawSockstat,
+  };
+}

senses/linux-tcp-socket-stats/migrations/0001_init.sql

@@ -0,0 +1,14 @@
+-- Migration: 0001_init
+-- Creates the linux_tcp_socket_stats table for linux-tcp-socket-stats sense.
+
+CREATE TABLE IF NOT EXISTS linux_tcp_socket_stats (
+  ts               INTEGER PRIMARY KEY,
+  sockets_used     INTEGER NOT NULL,
+  tcp_inuse        INTEGER NOT NULL,
+  tcp_orphan       INTEGER NOT NULL,
+  tcp_tw           INTEGER NOT NULL,
+  tcp_alloc        INTEGER NOT NULL,
+  tcp_mem_pages    INTEGER NOT NULL,
+  parse_ok         INTEGER NOT NULL,
+  raw_sockstat     TEXT    NOT NULL
+);

senses/linux-tcp-socket-stats/schema.ts

@@ -0,0 +1,13 @@
+import { integer, sqliteTable, text } from "drizzle-orm/sqlite-core";
+
+export const linuxTcpSocketStats = sqliteTable("linux_tcp_socket_stats", {
+  ts: integer("ts").primaryKey(),
+  socketsUsed: integer("sockets_used").notNull(),
+  tcpInuse: integer("tcp_inuse").notNull(),
+  tcpOrphan: integer("tcp_orphan").notNull(),
+  tcpTw: integer("tcp_tw").notNull(),
+  tcpAlloc: integer("tcp_alloc").notNull(),
+  tcpMemPages: integer("tcp_mem_pages").notNull(),
+  parseOk: integer("parse_ok").notNull(),
+  rawSockstat: text("raw_sockstat").notNull(),
+});

workflows/sense-generator/index.ts

@@ -64,7 +64,7 @@ function cursorAgent(prompt: string, mode: "plan" | "ask" | "default", cwd: stri
 function hermesRun(prompt: string): string {
   const escaped = prompt.replace(/'/g, "'\\''");
   return run(
-    `hermes -q '${escaped}' --model anthropic/claude-sonnet-4 --no-memory 2>&1 || true`,
+    `hermes chat -q '${escaped}' --model anthropic/claude-sonnet-4 -t terminal --yolo 2>&1 || true`,
   );
 }
 
@@ -281,17 +281,32 @@ Reply with either:
 - "FAIL: <error details>" if it doesn't`;
 
         const result = hermesRun(testPrompt);
-        ctx.log(`tester: result — ${result.substring(0, 200)}`);
+        ctx.log(`tester: raw result — ${result.substring(0, 300)}`);
 
-        const passed = result.toUpperCase().includes("PASS");
-        if (passed) {
-          return { type: "test_passed", senseName, result };
+        // Use LLM to judge pass/fail instead of fragile string matching
+        const verdict = llmExtract<{ passed: boolean; reason: string }>(
+          `Test output for sense "${senseName}":\n\n${result.substring(0, 4000)}`,
+          "judge_test_result",
+          "Determine whether the test passed or failed based on the output",
+          {
+            type: "object",
+            properties: {
+              passed: { type: "boolean", description: "true if the sense was successfully triggered and returned valid data" },
+              reason: { type: "string", description: "Brief explanation of why it passed or failed" },
+            },
+            required: ["passed", "reason"],
+          },
+        );
+        ctx.log(`tester: verdict — passed=${verdict.passed}, reason="${verdict.reason}"`);
+
+        if (verdict.passed) {
+          return { type: "test_passed", senseName, result: verdict.reason };
         }
 
         return {
           type: "test_failed",
           senseName,
-          reason: result,
+          reason: verdict.reason,
           attempt,
         };
       },