Debug Log Evidence

  With logging.level: "debug", consecutive Discord-triggered runs for the same channel show different sessionIds:

  20:27:17 embedded run start sessionId=0730aeb3 (channel:1471935020200493120)
  20:27:25 embedded run start sessionId=3d14dac8 (channel:1471935020200493120)  ← 8 sec later, NEW
  20:27:58 embedded run start sessionId=e3987442 (channel:1471935020200493120)  ← 33 sec later, NEW
  20:29:16 embedded run start sessionId=6649c869 (channel:1471935020200493120)  ← NEW
  20:31:03 embedded run start sessionId=6649c869 (channel:1471935020200493120)  ← held once
  20:36:57 embedded run start sessionId=50abf3c2 (channel:1471935020200493120)  ← NEW
  20:37:03 embedded run start sessionId=a95d5e45 (channel:1471935020200493120)  ← 6 sec later, NEW
  20:37:14 embedded run start sessionId=77bcf5f9 (channel:1471935020200493120)  ← 11 sec later, NEW

  All diagnostic entries show sessionKey=unknown, suggesting the session key isn't being resolved during the run
  pipeline.

  What Has Been Ruled Out

  Theory: Gateway process restarting
  How Tested: Confirmed single stable PID across checks
  Result: Not the cause
  ────────────────────────────────────────
  Theory: Multiple gateway instances
  How Tested: gateway probe → 1 target, 0 beacons
  Result: Not the cause
  ────────────────────────────────────────
  Theory: Session store file corruption
  How Tested: sessions.json always valid (0/1000 parse failures at 25ms polling during active writes)
  Result: Not the cause
  ────────────────────────────────────────
  Theory: Session store not persisting
  How Tested: File size stable, LastWriteTime updating
  Result: Not the cause
  ────────────────────────────────────────
  Theory: Session cache race
  How Tested: OPENCLAW_SESSION_CACHE_TTL_MS=0
  Result: Still rotates
  ────────────────────────────────────────
  Theory: contextPruning
  How Tested: Set contextPruning.mode: "off"
  Result: Still rotates
  ────────────────────────────────────────
  Theory: Session reset config
  How Tested: Added session.resetByType.group: { mode: "idle", idleMinutes: 10080 }
  Result: Still rotates
  ────────────────────────────────────────
  Theory: Hooks/cron interference
  How Tested: All 4 hooks bundled, 8 cron jobs all sessionTarget: "isolated", none target the channel
  Result: Not the cause
  ────────────────────────────────────────
  Theory: Scheduled Task lifecycle
  How Tested: Fixed all settings (logon mode, battery, timeout) — was causing gateway kills but fixing it didn't resolve

    session rotation
  Result: Separate issue, fixed but not the cause

  Session Store Entry (sample)

  The session entry in sessions.json looks correct and is present:

  {
    "key": "agent:main:discord:channel:1471935020200493120",
    "kind": "group",
    "sessionId": "0730aeb3-38ca-406b-9543-d5ac378470f1",
    "sessionFile": "...\\sessions\\0730aeb3-38ca-406b-9543-d5ac378470f1.jsonl",
    "compactionCount": 0,
    "chatType": "channel",
    "channel": "discord",
    "groupId": "1471935020200493120"
  }

  The entry exists, the file is valid JSON, but the next run ignores it and mints a new sessionId anyway.

  Suspected Root Cause

  The session resolution path for kind: "group" Discord channel runs is not looking up or reusing the existing session
  entry from the store. By the time embedded run start fires in the debug logs, a new sessionId has already been
  assigned — the decision happens upstream in the session init/resolution code, not in the store read.

  Workaround

  None found. WSL2 has not been tested but may exhibit the same behavior since the parse-failure theory (non-atomic
  Windows file writes) was disproven.

  Config (redacted)

  {
    "channels": {
      "discord": {
        "enabled": true,
        "groupPolicy": "allowlist",
        "guilds": {
          "<guild_id>": {
            "requireMention": false,
            "users": ["<user_id>"],
            "channels": {
              "<channel_id>": { "allow": true }
            }
          }
        },
        "intents": {}
      }
    },
    "agents": {
      "defaults": {
        "contextPruning": { "mode": "cache-ttl", "ttl": "30m", "keepLastAssistants": 3 },
        "compaction": { "mode": "safeguard", "memoryFlush": { "enabled": true } }
      }
    },
    "session": {
      "resetByType": { "group": { "mode": "idle", "idleMinutes": 10080 } }
    }
  }