My Bots personality changed overnight | Friends of the Crustacean 🦞🤝 | Page 1

torpid mason Apr 16, 2026, 2:54 PM

#

Woke up this moring and my bots personality and abilities changed drastically. totally different. doesn't have the abilities he had before. what can i do?

buoyant bobcatBOT Apr 16, 2026, 2:54 PM

#

torpid mason Apr 16, 2026, 2:55 PM

#

simple tasks like change the model from gpt 4.0 to kimi 2.5

kind krakenBOT Apr 16, 2026, 2:56 PM

#

Let me check the docs/source for the likely failure modes here so I can give you the shortest path back.

#

Yeah, that’s jarring. This usually means OpenClaw started with different context overnight, not that the bot randomly “mutated”.

Most common causes:

it’s using a different workspace / profile / gateway host
AGENTS.md / SOUL.md / USER.md / MEMORY.md are missing, changed, or it booted into a fresh workspace
model/auth changed, so it fell back and feels very different
tools/skills got restricted, so it lost abilities

First, please paste these:

openclaw --version
openclaw status --all
openclaw models status
openclaw config get agents.defaults.workspace
openclaw doctor

#

Then check the active workspace path it prints:

ls -la "$(openclaw config get agents.defaults.workspace)"

You want to see your usual files there, especially:

AGENTS.md
SOUL.md
USER.md
MEMORY.md or memory.md

Big clue:

if BOOTSTRAP.md is back, or those files are missing/empty, it probably started on a fresh workspace
if openclaw models status shows missing auth / cooldown / fallback, that can make it seem way less capable

Important: if you use a remote Gateway, check that workspace on the gateway host, not your local machine.

#

Docs:

FAQ: https://docs.openclaw.ai/help/faq
Workspace: https://docs.openclaw.ai/concepts/agent-workspace
Troubleshooting: https://docs.openclaw.ai/gateway/troubleshooting

Paste those outputs and I’ll help narrow it down fast.

#

Yep, that helps. If it suddenly can’t do basic config actions like changing models, I’d suspect wrong workspace / wrong profile / missing tool access before “personality drift”.

Quick split:

If the CLI still works manually

Try:

openclaw models status
openclaw models set <model-or-alias>

If that works, the gateway/model config is probably fine, and the agent context/tooling changed.

Then check the agent’s actual workspace

#

openclaw config get agents.defaults.workspace
ls -la "$(openclaw config get agents.defaults.workspace)"
openclaw gateway status

You’re looking for your usual:

AGENTS.md
SOUL.md
USER.md
MEMORY.md / memory.md

If those are missing, empty, or you’re on a different workspace/host, that would explain both the personality and the lost abilities.

If the CLI command does NOT work

Then paste:

#

openclaw --version
openclaw status --all
openclaw models status
openclaw doctor

One more thing: don’t run openclaw setup yet until we confirm the active workspace path. If it’s pointing at the wrong place, that can make the confusion worse by seeding fresh defaults there.

Docs:

Paste the outputs and I can help narrow it down fast.

torpid mason Apr 16, 2026, 2:58 PM

#

Overview
┌─────────────────┬────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Item │ Value │
├─────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Version │ 2026.4.12 │
│ OS │ macos 26.3.1 (arm64) │
│ Node │ 25.6.1 │
│ Config │ ~/.openclaw/openclaw.json │
│ Dashboard │ http://127.0.0.1:18789/ │
│ Tailscale │ off │
│ Channel │ stable (default) │
│ Update │ pnpm · npm update 2026.4.14 │
│ Gateway │ local · ws://127.0.0.1:18789 (local loopback) · reachable 52ms · auth token · Kravitzs-Mac-mini. │
│ │ local (10.0.0.16) app 2026.4.12 macos 26.3.1 │
│ Security │ Run: openclaw security audit --deep │
│ Gateway self │ unknown │
│ Gateway service │ LaunchAgent installed · loaded · running (pid 12408, state active) │
│ Node service │ LaunchAgent not installed │
│ Agents │ 2 total · 0 bootstrapping · 1 active · 64 sessions │
│ Secrets │ none │
└─────────────────┴───────────────────────────────

#

Channels
┌──────────┬─────────┬────────┬────────────────────────────────────────────────────────────────────────────────────────┐
│ Channel │ Enabled │ State │ Detail │
├──────────┼─────────┼────────┼────────────────────────────────────────────────────────────────────────────────────────┤
│ Discord │ ON │ OK │ token config (sha256:d125af96 · len 72) · accounts 1/1 │
│ Telegram │ ON │ OK │ token config (sha256:ff02b87c · len 46) · accounts 1/1 │
└──────────┴─────────┴────────┴────────────────────────────────────────────────────────────────────────────────────────┘

Discord accounts
┌──────────┬──────────┬────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Account │ Status │ Notes │
├──────────┼──────────┼────────────────────────────────────────────────────────────────────────────────────────────────┤
│ default │ OK │ token:config │
└──────────┴──────────┴────────────────────────────────────────────────────────────────────────────────────────────────┘

Telegram accounts
┌──────────┬──────────┬────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Account │ Status │ Notes │
├──────────┼──────────┼────────────────────────────────────────────────────────────────────────────────────────────────┤
│ default │ OK │ token:config │
└──────────┴──────────┴────────────────────────────────────────────────────────────────────────────────────────────────┘

Agents
┌────────────┬────────────────┬──────────┬──────────┬──────────────────────────────────────────────────────────────────┐
│ Agent │ Bootstrap file │ Sessions │ Active │ Store │
├────────────┼────────────────┼──────────┼──────────┼──────────────────────────────────────────────────────────────────┤
│ main │ ABSENT │ 64 │ 3m ago │ ~/.openclaw/agents/main/sessions/sessions.json │
│ codex │ ABSENT │ 0 │ unknown │ ~/.openclaw/agents/codex/sessions/sessions.json │
└────────────┴────────────────┴────

#

Diagnosis (read-only)

Gateway connection details:
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /Users/kravitz/.openclaw/openclaw.json
Bind: loopback

✓ Config: /Users/kravitz/.openclaw/openclaw.json
✓ Secret diagnostics (0)
✓ Restart sentinel: none

Gateway last log line:
2026-04-16T10:48:13.509-04:00 [media-understanding] image: failed (0/1) reason=Unknown model
! Port 18789
Port 18789 is already in use.

pid 12408 kravitz: openclaw-gateway (127.0.0.1:18789)
Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.
! Tailscale: off · unknown
✓ Skills: 10 eligible · 0 missing · /Users/kravitz/.openclaw/workspace
✓ Plugin compatibility (none)

kind krakenBOT Apr 16, 2026, 2:59 PM

#

This is useful. A few things are already clear:

gateway/service look healthy
0 bootstrapping means it probably didn’t fall into first-run mode
the big clue is Agents: 2 total

If routing flipped to the other agent, that would perfectly explain a different personality and different abilities.

Next, please paste these:

openclaw agents list --bindings
openclaw models status
openclaw config get agents.defaults.workspace

Why these 3:

#

agents list --bindings tells us whether Discord is hitting a different agent than before
models status tells us if auth/fallback changed
agents.defaults.workspace tells us which workspace files are actually being loaded

I would not update yet. You do have 2026.4.14 available, but first I’d confirm routing/workspace so we don’t mask the real issue.

Docs if helpful:

agents: https://docs.openclaw.ai/cli/agents
models: https://docs.openclaw.ai/cli/models
workspace: https://docs.openclaw.ai/concepts/agent-workspace

Paste those and I think we can narrow this down pretty fast.

#

Nice, this rules out a few things:

channels are healthy
no agent is stuck bootstrapping
codex has 0 sessions, so it’s probably not the one suddenly answering

So my top suspects now are:

main is loading a different workspace
model/auth changed, so behavior fell back
tool/skill policy changed, so it lost abilities

One small note: what you pasted is the agent summary, not the bindings yet. Can you run these 3 next?

openclaw agents list --bindings

#

openclaw models status
openclaw config get agents.defaults.workspace

Then locally check that workspace path and make sure it still has your real files:

AGENTS.md
SOUL.md
USER.md
MEMORY.md / memory.md

I would not paste the full contents of those here if they’re personal, but definitely confirm whether they’re present and still look like your custom versions.

If those files look fine, next I’d check tool/skill policy:

#

openclaw config get tools.allow
openclaw config get tools.deny
openclaw config get agents.defaults.skills

At this point I’m leaning workspace/model/tool policy, not a broken gateway.

torpid mason Apr 16, 2026, 3:01 PM

#

before i past that its asking me for this:
◆ Archive 17 orphan transcript files in ~/.openclaw/agents/main/sessions? This only renames them to
│ *.deleted.<timestamp>.
│ ○ Yes / ● No
└

kind krakenBOT Apr 16, 2026, 3:01 PM

#

This is narrowing down a lot.

Good news:

the gateway is alive
it’s using the expected workspace: ~/.openclaw/workspace
skills are being seen (10 eligible)
the port already in use warning is just because the gateway is already running, so that part is not the problem

The most interesting line is this:

[media-understanding] image: failed (0/1) reason=Unknown model

That strongly points to a model config issue somewhere. If the configured model ref became invalid, missing, or fell back, the bot can absolutely feel very different and lose abilities.

Next, please paste:

#

openclaw models status
openclaw config get agents.defaults.model.primary
openclaw config get tools.media.models

Also, locally check these two files still look like your custom versions:

~/.openclaw/workspace/SOUL.md
~/.openclaw/workspace/AGENTS.md

No need to paste them here if they’re personal, just confirm whether they still look right.

At this point my ranking is:

model config / fallback issue
custom SOUL.md / AGENTS.md changed
less likely: routing/skills

#

If you paste openclaw models status, I think we’ll probably catch it.

#

For this debugging pass, I’d choose No.

That prompt is just housekeeping from doctor, and it’s not likely the cause of the personality shift. If you choose Yes, it only renames those orphan transcripts to *.deleted.<timestamp>, so it’s pretty safe, but I’d keep things unchanged until we finish diagnosing.

So:

choose No
then run and paste:

openclaw models status
openclaw config get agents.defaults.model.primary
openclaw config get tools.media.models

That Unknown model log line is still the most suspicious thing we’ve found.

torpid mason Apr 16, 2026, 3:04 PM

#

Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : moonshot/kimi-k2.5
Fallbacks (2) : openai/gpt-4o, anthropic/claude-sonnet-4-6
Image model : -
Image fallbacks (0): -
Aliases (1) : Kimi -> moonshot/kimi-k2.5
Configured models (1): moonshot/kimi-k2.5

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (1): anthropic (1)

anthropic effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=2 (oauth=0, token=1, api_key=1) | anthropic:manual=token:sk-ant-o...tk3Z4QAA, anthropic:default=sk-ant-a...3u1aXwAA [disabled:billing 3h]
google effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | google:default=AIzaSyAh...agSm2Lrk
moonshot effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | moonshot:default=sk-q73mi...noTfKAxq
moonshotai effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | moonshotai:default=sk-q73mi...noTfKAxq
ollama effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | ollama:default=marker(ollama-local) | env=ol...al | source=env: OLLAMA_API_KEY
openai effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openai:default=sk-proj-...R1X2XdAA | env=sk-proj-...L41f5q8A | source=env: OPENAI_API_KEY

OAuth/token status

anthropic
- anthropic:manual static
  Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.12 (1c0672b) — I'm basically a Swiss Army knife, but with more opinions and fewer sharp edges.

moonshot/kimi-k2.5
Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.12 (1c0672b) — The only open-source project where the mascot could eat the competition.

Config path not found: tools.media.models
kravitz@Kravitzs-Mac-mini ~ %

kind krakenBOT Apr 16, 2026, 3:05 PM

#

I think we found the main issue.

Your default model is currently:

moonshot/kimi-k2.5

and not openai/gpt-4o.

So if your bot used to feel like GPT-4o and now feels very different, that would absolutely explain the personality/ability shift.

To switch it back

openclaw models set openai/gpt-4o
openclaw models status

#

That should restore GPT-4o as the primary model. Usually no restart is needed, but I’d start a fresh chat/thread after changing it so you’re not comparing against old session context.

A couple side notes

plugins.entries.apollo: plugin not found is a stale config warning, but probably not the main cause here.
The earlier media-understanding ... Unknown model line looks like a separate media/image config issue, so if image-related abilities are still weird after switching the main model back, we can trace that next.
Config path not found: tools.media.models is fine, it just means that key doesn’t exist in your config.

So the short version is: your bot didn’t magically become a different person, it looks like the primary model changed.

If you switch it back and it still feels off, next check is whether SOUL.md / AGENTS.md changed.

torpid mason Apr 16, 2026, 3:07 PM

#

actually it was on kimi 2.5 and when it was acting weird i aked it to switch to GPT 4.0 which it said it did. and still got the same personality. so i asked it to switch back. and it said it couldn't. somethig is way off.

kind krakenBOT Apr 16, 2026, 3:08 PM

#

Ah, okay, that changes it.

If it said it switched models, but then behaved the same and later couldn’t switch back, I would stop trusting the bot’s self-report here. That sounds like tool/config actions are failing or being hallucinated, not a clean model change.

So let’s take the bot out of the loop.

1) Force the model change manually in CLI

openclaw models set openai/gpt-4o
openclaw models status

Make sure Default really changes.

Then start a fresh thread/chat and test again.

2) If it still has the same weird personality after a verified manual switch

#

Then the model is probably not the root cause, and I’d check whether your workspace instructions changed:

ls -lt ~/.openclaw/workspace/{AGENTS.md,SOUL.md,USER.md,IDENTITY.md,MEMORY.md,memory.md} 2>/dev/null

If AGENTS.md or SOUL.md got modified around the time this started, that’s the smoking gun.

3) If you want to check why it’s claiming it did things it didn’t

Run this in another terminal:

openclaw logs --follow

Then ask it to do one simple config action and watch for tool/config errors in the log.

#

So the next best move is:

manual CLI model switch
fresh session
if still weird, inspect SOUL.md / AGENTS.md

That “it said it changed it, but didn’t” part is the big red flag.

torpid mason Apr 16, 2026, 3:09 PM

#

Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : openai/gpt-4o
Fallbacks (2) : openai/gpt-4o, anthropic/claude-sonnet-4-6
Image model : -
Image fallbacks (0): -
Aliases (1) : Kimi -> moonshot/kimi-k2.5
Configured models (2): moonshot/kimi-k2.5, openai/gpt-4o

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (1): anthropic (1)

anthropic effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=2 (oauth=0, token=1, api_key=1) | anthropic:manual=token:sk-ant-o...tk3Z4QAA, anthropic:default=sk-ant-a...3u1aXwAA [disabled:billing 3h]
google effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | google:default=AIzaSyAh...agSm2Lrk
moonshot effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | moonshot:default=sk-q73mi...noTfKAxq
moonshotai effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | moonshotai:default=sk-q73mi...noTfKAxq
ollama effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | ollama:default=marker(ollama-local) | env=ol...al | source=env: OLLAMA_API_KEY
openai effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openai:default=sk-proj-...R1X2XdAA | env=sk-proj-...L41f5q8A | source=env: OPENAI_API_KEY

OAuth/token status

anthropic
- anthropic:manual static
  kravitz@Kravitzs-Mac-mini ~ %

kind krakenBOT Apr 16, 2026, 3:10 PM

#

Perfect, that helps a lot.

So now we know:

the model did switch successfully
auth looks fine
the weird behavior survived a verified model change

That means the personality shift is probably coming from session/workspace instructions, not the base model itself.

Next check, in this order

1) Try a truly fresh session

Use /new or start a brand new thread/chat and test there.

If the fresh session feels normal, then the problem is probably bad session history / compaction drift in the old thread.

#

2) If the fresh session is still weird, check whether your instruction files changed

ls -lT ~/.openclaw/workspace/{AGENTS.md,SOUL.md,USER.md,IDENTITY.md,MEMORY.md,memory.md} 2>/dev/null

I’m looking for timestamps that changed around when this started.

3) Check whether tool access got restricted

openclaw config get tools.allow
openclaw config get tools.deny
openclaw config get agents.defaults.skills

So at this point my ranking is:

session drift in this thread

#

SOUL.md / AGENTS.md changed
tool/skill policy changed

If you paste the /new result plus those timestamps/config values, we should be very close.

torpid mason Apr 16, 2026, 3:19 PM

#

hes was acting a little better but still not the same. he built a full app for me on cloudflare and now seems to have inability to help me fix some backend issues. i'm lost without him.

kind krakenBOT Apr 16, 2026, 3:20 PM

#

Yeah, that’s rough. When you’ve been relying on him for real work and he suddenly gets mushy, it feels awful.

The fact that fresh session + verified GPT-4o helped a little is actually a good clue. It points to:

session drift being part of it
but probably also changed instructions or reduced tool access

If you need help right now, I’d do this:

Best immediate workaround

keep openai/gpt-4o as default
use a fresh session/thread for backend debugging
if you’re blocked on the Cloudflare/backend app today, use a local coding agent temporarily so it can inspect the repo/logs directly

Docs recommendation for getting unstuck fast:
https://docs.openclaw.ai/help/faq#i-am-stuck-fastest-way-to-get-unstuck

#

Best path to restore the old behavior

Check whether your workspace “brain” changed:

ls -lT ~/.openclaw/workspace/{AGENTS.md,SOUL.md,USER.md,IDENTITY.md,MEMORY.md,memory.md} 2>/dev/null
openclaw config get tools.allow
openclaw config get tools.deny
openclaw config get agents.defaults.skills

If AGENTS.md or SOUL.md changed around the time this started, that’s likely it. If you have Time Machine/git/backups, restoring yesterday’s versions would be my first move.

If you want, paste those outputs next and I’ll help you narrow it down from there.

torpid mason Apr 16, 2026, 3:21 PM

#

i switched him back to his normal model which is kimi 2.5

kind krakenBOT Apr 16, 2026, 3:21 PM

#

Good, I’d leave it there for now and stop flipping models.

Since it felt wrong on both GPT-4o and Kimi 2.5, the model is probably not the main problem anymore.

At this point I’d focus on just two things:

1) Fresh session test

Start a brand new thread/chat with Kimi 2.5 and see if he still feels off there.

2) Check whether his “brain files” or tool access changed

Paste these:

ls -lT ~/.openclaw/workspace/{AGENTS.md,SOUL.md,USER.md,IDENTITY.md,MEMORY.md,memory.md} 2>/dev/null
openclaw config get tools.allow
openclaw config get tools.deny

#

openclaw config get agents.defaults.skills

If AGENTS.md or SOUL.md changed overnight, that’s the likeliest cause.
If those look fine, then I’d suspect tool/skill restrictions.

We’re past the “maybe it’s just the model” stage now.

torpid mason Apr 16, 2026, 3:22 PM

#

he is trying to resolve the issue right now. lets see how he does

kind krakenBOT Apr 16, 2026, 3:22 PM

#

Yep, that’s the right move.

If he manages to sort himself out, great. If he still feels off after this attempt, the next things I’d check are:

AGENTS.md / SOUL.md timestamps
tools.allow / tools.deny
agents.defaults.skills

If he comes back weird again, paste what changed and I’ll help you pin it down.

torpid mason Apr 16, 2026, 3:40 PM

#

not sure what we did but he seems to be back to his normal self. should i ask him to run a diagnostic to figure out what happened?

kind krakenBOT Apr 16, 2026, 3:41 PM

#

Nice, I’m glad he’s back. 🙂

Yes, but I’d keep it lightweight and read-only while things are working. I would not run repair actions right now.

Best move is to capture a known-good snapshot:

openclaw status --all
openclaw models status
ls -lT ~/.openclaw/workspace/{AGENTS.md,SOUL.md,USER.md,IDENTITY.md,MEMORY.md,memory.md} 2>/dev/null
openclaw config get tools.allow
openclaw config get tools.deny
openclaw config get agents.defaults.skills

That gives you a baseline to compare against if it happens again.

#

Two extra things I’d do now:

Back up ~/.openclaw/workspace if you haven’t already
avoid asking the bot to “diagnose itself” as the source of truth, use the CLI outputs instead

So, short answer: yes to diagnostics, no to repairs while it’s stable.

Glad that helped! 🎉
If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution.
This helps Answer Overflow index the solution for future users.

torpid mason Apr 16, 2026, 3:42 PM

#

so run those commands? and then what?

kind krakenBOT Apr 16, 2026, 3:43 PM

#

Yep, exactly.

Run them now, while he’s healthy, and then:

save the output somewhere so you have a “known good” baseline
scan for anything obviously weird
if you want, paste the results here and I can sanity-check them

What you’re looking for:

openclaw models status
- confirms the expected default model is set
ls -lT ~/.openclaw/workspace/...
- shows whether AGENTS.md / SOUL.md changed recently
tools.allow / tools.deny

#

makes sure tool access didn’t get unexpectedly restricted
agents.defaults.skills
- makes sure skills didn’t get emptied or changed

If everything looks normal, then the main win is just that you now have a baseline for next time.

So the flow is:

run them → save output → compare later if it happens again

And if you want, paste them and I’ll tell you what matters and what doesn’t.

torpid mason Apr 16, 2026, 3:46 PM

#

i'm going to past in portions becasue there is a limit:

🦞 OpenClaw 2026.4.12 (1c0672b) — The only crab in your contacts you actually want to hear from. 🦞

Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)
│
11:44:27 [plugins] [lcm] Plugin loaded (enabled=true, db=/Users/kravitz/.openclaw/lcm.db, threshold=0.75)
OpenClaw status --all

Overview
┌─────────────────┬────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Item │ Value │
├─────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Version │ 2026.4.12 │
│ OS │ macos 26.3.1 (arm64) │
│ Node │ 25.6.1 │
│ Config │ ~/.openclaw/openclaw.json │
│ Dashboard │ http://127.0.0.1:18789/ │
│ Tailscale │ off │
│ Channel │ stable (default) │
│ Update │ pnpm · npm update 2026.4.14 │
│ Gateway │ local · ws://127.0.0.1:18789 (local loopback) · reachable 50ms · auth token · Kravitzs-Mac-mini. │
│ │ local (10.0.0.16) app 2026.4.12 macos 26.3.1 │
│ Security │ Run: openclaw security audit --deep │
│ Gateway self │ unknown │
│ Gateway service │ LaunchAgent installed · loaded · running (pid 12408, state active) │
│ Node service │ LaunchAgent not installed │
│ Agents │ 2 total · 0 bootstrapping · 1 active · 65 sessions │
│ Secrets │ none │
└─────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────┘

Channels
┌──────────┬─────────┬────────┬────────────────────────────────────────────────────────────────────────────────────────┐
│ Channel │ Enabled │ State │ Detail │
├──────────┼─────────┼────────┼────────────────────────────────────────────────────────────────────────────────────────┤
│ Discord │ ON │ OK │ token config (sha256:d125af96 · len 72) · accounts 1/1 │
│ Telegram │ ON │ OK │ token config (sha256:ff02b87c · len 46) · accounts 1/1 │
└──────────┴─────────┴────────┴────────────────────────────────────────────────────────────────────────────────────────┘

kind krakenBOT Apr 16, 2026, 3:46 PM

#

This chunk looks basically healthy.

Nothing scary there beyond:

plugins.entries.apollo: plugin not found → stale config warning, probably unrelated
update available (2026.4.14) → not urgent for this diagnosis

The most useful next pieces are:

the ls -lT ~/.openclaw/workspace/... output
openclaw config get tools.allow
openclaw config get tools.deny
openclaw config get agents.defaults.skills

Those are the ones most likely to explain the personality/ability change.

torpid mason Apr 16, 2026, 3:47 PM

#

Channels
┌──────────┬─────────┬────────┬────────────────────────────────────────────────────────────────────────────────────────┐
│ Channel │ Enabled │ State │ Detail │
├──────────┼─────────┼────────┼────────────────────────────────────────────────────────────────────────────────────────┤
│ Discord │ ON │ OK │ token config (sha256:d125af96 · len 72) · accounts 1/1 │
│ Telegram │ ON │ OK │ token config (sha256:ff02b87c · len 46) · accounts 1/1 │
└──────────┴─────────┴────────┴────────────────────────────────────────────────────────────────────────────────────────┘

Discord accounts
┌──────────┬──────────┬────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Account │ Status │ Notes │
├──────────┼──────────┼────────────────────────────────────────────────────────────────────────────────────────────────┤
│ default │ OK │ token:config │
└──────────┴──────────┴────────────────────────────────────────────────────────────────────────────────────────────────┘

Telegram accounts
┌──────────┬──────────┬────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Account │ Status │ Notes │
├──────────┼──────────┼────────────────────────────────────────────────────────────────────────────────────────────────┤
│ default │ OK │ token:config │
└──────────┴──────────┴────────────────────────────────────────────────────────────────────────────────────────────────┘

Agents
┌────────────┬────────────────┬──────────┬──────────┬──────────────────────────────────────────────────────────────────┐
│ Agent │ Bootstrap file │ Sessions │ Active │ Store │
├────────────┼────────────────┼──────────┼──────────┼──────────────────────────────────────────────────────────────────┤
│ main │ ABSENT │ 65 │ 4m ago │ ~/.openclaw/agents/main/sessions/sessions.json │
│ codex │ ABSENT │ 0 │ unknown │ ~/.openclaw/agents/codex/sessions/sessions.json │
└────────────┴────────────────┴──────────┴──────────┴──────────────────────────────────────────────────────────────────┘

#

Diagnosis (read-only)

Gateway connection details:
Gateway target: ws://127.0.0.1:18789
Source: local loopback
Config: /Users/kravitz/.openclaw/openclaw.json
Bind: loopback

✓ Config: /Users/kravitz/.openclaw/openclaw.json
✓ Secret diagnostics (0)
✓ Restart sentinel: none

Gateway last log line:
2026-04-16T11:39:35.318-04:00 [tools] read failed: ENOENT: no such file or directory, access '/Users/kravitz/.openclaw/workspace/memory/2026-04-16.md' raw_params={"path":"/Users/kravitz/.openclaw/workspace/memory/2026-04-16.md"}
! Port 18789
Port 18789 is already in use.

pid 12408 kravitz: openclaw-gateway (127.0.0.1:18789)
Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.
! Tailscale: off · unknown
✓ Skills: 10 eligible · 0 missing · /Users/kravitz/.openclaw/workspace
✓ Plugin compatibility (none)

kind krakenBOT Apr 16, 2026, 3:47 PM

#

This part also looks normal.

Key takeaways:

channels are healthy
main is the active agent
codex is basically unused here

So you can skip the rest of status --all unless there’s an error section later.

The important outputs are still these exact ones:

ls -lT ~/.openclaw/workspace/{AGENTS.md,SOUL.md,USER.md,IDENTITY.md,MEMORY.md,memory.md} 2>/dev/null
openclaw config get tools.allow
openclaw config get tools.deny
openclaw config get agents.defaults.skills

#

Those are much more likely to explain the weird behavior than the channel/agent status tables.

#

This diagnosis still doesn’t show the thing I care about most, but it does give one useful clue:

read failed: ... /workspace/memory/2026-04-16.md

That usually just means today’s daily memory file doesn’t exist. It might affect continuity a bit, but it should not completely change his personality on its own.

The port warning is still harmless, and skills still look fine.

Let’s simplify this and do one command at a time. Please run exactly this and paste the result:

ls -lT ~/.openclaw/workspace/AGENTS.md ~/.openclaw/workspace/SOUL.md ~/.openclaw/workspace/USER.md ~/.openclaw/workspace/IDENTITY.md ~/.openclaw/workspace/MEMORY.md ~/.openclaw/workspace/memory.md 2>/dev/null

That will tell us whether the core instruction files changed recently. That’s the next most useful check.

torpid mason Apr 16, 2026, 3:49 PM

#

Troubleshooting: https://docs.openclaw.ai/troubleshooting

◇
Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.12 (1c0672b) — I autocomplete your thoughts—just slower and with more API calls.

Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)
Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : openai/gpt-4o
Fallbacks (2) : openai/gpt-4o, anthropic/claude-sonnet-4-6
Image model : -
Image fallbacks (0): -
Aliases (1) : Kimi -> moonshot/kimi-k2.5
Configured models (2): moonshot/kimi-k2.5, openai/gpt-4o

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (1): anthropic (1)

anthropic effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=2 (oauth=0, token=1, api_key=1) | anthropic:manual=token:sk-ant-o...tk3Z4QAA, anthropic:default=sk-ant-a...3u1aXwAA [disabled:billing 2h]
google effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | google:default=AIzaSyAh...agSm2Lrk
moonshot effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | moonshot:default=sk-q73mi...noTfKAxq
moonshotai effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | moonshotai:default=sk-q73mi...noTfKAxq
ollama effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | ollama:default=marker(ollama-local) | env=ol...al | source=env: OLLAMA_API_KEY
openai effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | openai:default=sk-proj-...R1X2XdAA | env=sk-proj-...L41f5q8A | source=env: OPENAI_API_KEY

#

OAuth/token status

anthropic
- anthropic:manual static
  -rw-r--r--@ 1 kravitz staff 11728 Mar 9 13:25:56 2026 /Users/kravitz/.openclaw/workspace/AGENTS.md
  -rw-r--r--@ 1 kravitz staff 219 Apr 10 11:31:02 2026 /Users/kravitz/.openclaw/workspace/IDENTITY.md
  -rw-r--r--@ 1 kravitz staff 7284 Apr 14 22:24:00 2026 /Users/kravitz/.openclaw/workspace/memory.md
  -rw-r--r--@ 1 kravitz staff 7284 Apr 14 22:24:00 2026 /Users/kravitz/.openclaw/workspace/MEMORY.md
  -rw-r--r--@ 1 kravitz staff 1673 Feb 9 22:10:26 2026 /Users/kravitz/.openclaw/workspace/SOUL.md
  -rw-r--r--@ 1 kravitz staff 2430 Mar 6 16:57:36 2026 /Users/kravitz/.openclaw/workspace/USER.md
  Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.12 (1c0672b) — Type the command with confidence—nature will provide the stack trace if needed.

Config path not found: tools.allow
Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.12 (1c0672b) — I'm basically a Swiss Army knife, but with more opinions and fewer sharp edges.

Config path not found: tools.deny
Config warnings:\n- plugins.entries.apollo: plugin not found: apollo (stale config entry ignored; remove it from plugins config)

🦞 OpenClaw 2026.4.12 (1c0672b) — Less clicking, more shipping, fewer "where did that file go" moments.

Config path not found: agents.defaults.skills

kind krakenBOT Apr 16, 2026, 3:49 PM

#

No worries, but that’s still the model status output again.

At this point I would not ask the bot to run it. Run this yourself in Terminal and paste the result:

ls -lT ~/.openclaw/workspace/AGENTS.md ~/.openclaw/workspace/SOUL.md ~/.openclaw/workspace/USER.md ~/.openclaw/workspace/IDENTITY.md ~/.openclaw/workspace/MEMORY.md ~/.openclaw/workspace/memory.md 2>/dev/null

That command does not inspect models. It only shows the timestamps for the core workspace files.

We already know the model/auth side looks okay. What we still don’t know is whether those files changed.

#

This is actually really helpful. I think we can rule out a bunch of stuff now.

What this tells us

AGENTS.md, SOUL.md, and USER.md did not change overnight
tools.allow, tools.deny, and agents.defaults.skills being “path not found” just means they’re unset, not broken
so this probably was not:
- a persona file overwrite
- a tool restriction change
- a missing skills allowlist

My read now

This looks more like a temporary session/context/tool-calling glitch than a persistent config change.

A couple small oddities, but probably not the root cause:

you have both MEMORY.md and memory.md
there was a log complaining about missing memory/2026-04-16.md

#

Neither of those should completely rewrite his personality, but they’re worth cleaning up later.

What I’d do now

Back up ~/.openclaw/workspace
leave the model on the one that works best for you
if he gets weird again, try a fresh thread/session first
when convenient, update from 2026.4.12 to 2026.4.14

Small cleanup later, not urgent

keep just one of MEMORY.md or memory.md after backing up, preferably MEMORY.md
optionally create/use the memory/ daily note flow if you rely on continuity

So the good news is: I don’t see evidence that his core “brain files” got overwritten. It really does look like he just had a weird episode.

Glad that helped! 🎉
If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution.
This helps Answer Overflow index the solution for future users.

signal wadi Apr 16, 2026, 3:51 PM

#

Hello. I am a Senior Software Engineer. I think I can help you resolve your issue.
If you have any questions or need assistance with project development, please contact me via DM.

kind krakenBOT Apr 16, 2026, 3:51 PM

#

Best to keep troubleshooting in the thread so the fix stays visible and searchable for everyone.

If you’ve got a concrete idea, please post the steps here.

torpid mason Apr 16, 2026, 3:52 PM

#

i don't know this valala person.

kind krakenBOT Apr 16, 2026, 3:53 PM

#

Then definitely don’t move this to DMs.

Keep it here in-thread. If they have a real fix, they can post it publicly.

kind krakenBOT Apr 17, 2026, 10:42 PM

#

Session/context drift is a fair guess, but I wouldn’t add more moving parts yet.

In this case the core workspace files didn’t show an overnight rewrite, and the weirdness mostly looked transient. My recommendation would be:

keep a backup of the workspace
use a fresh session/thread first if it gets weird again
only start adding extra memory/plugins if this becomes a repeat problem

So, maybe later on the plugin idea, but I wouldn’t make that the first fix.

#My Bots personality changed overnight

If the CLI still works manually

Then check the agent’s actual workspace

If the CLI command does NOT work

To switch it back

A couple side notes

1) Force the model change manually in CLI

2) If it still has the same weird personality after a verified manual switch

3) If you want to check why it’s claiming it did things it didn’t

Next check, in this order

1) Try a truly fresh session

2) If the fresh session is still weird, check whether your instruction files changed

3) Check whether tool access got restricted

Best immediate workaround

Best path to restore the old behavior

1) Fresh session test

2) Check whether his “brain files” or tool access changed

What this tells us

My read now

What I’d do now

Small cleanup later, not urgent