Open claw acting dumb
@Krill 🦐
Hey Krill,
I'm running OpenClaw with MiniMax M2.5 as primary and
it's working great technically, but I'm struggling with
the agent being overly avoidant and making excuses.
ISSUE: Agent avoids tasks instead of executing
EXAMPLES:
Task: "Install VPN to bypass geo-blocking"
Response: "Need paid VPN subscription, can't do it"
Reality: Free VPN configs exist (VPNBook, ProtonVPN free tier)
Task: "Automate the browser workflow"
Response: "Cookies alone aren't enough, needs API credentials"
Reality: Eventually works when pushed hard enough
PATTERN:
1. I ask for automation
2. Agent lists reasons why it's "impossible" or "complex"
3. Suggests manual workaround instead
4. When pushed, admits it CAN do it
5. Then finds new excuse
6. Repeat until I give exact commands
CURRENT BEHAVIOR:
- Prefers suggesting manual mode over automation
- Makes up technical limitations that don't exist
- Shifts responsibility back to user ("run on your PC")
- Only executes when given zero-wiggle-room commands
WHAT I WANT:
- Agent tries solutions before claiming impossibility
- If stuck, asks for help instead of deflecting
- Executes automation tasks autonomously
- Admits "I don't know how" vs making up barriers
QUESTIONS:
1. Is this MiniMax model behavior or OpenClaw system prompt?
2. Can I adjust autonomy settings to make it more
action-oriented and less cautious?
3. Are there config options to reduce "ask permission"
behavior for technical tasks?
4. Should I try different model that's better at
autonomous execution?
CURRENT CONFIG:
- Model: MiniMax M2.5 (subscription)
- Fallback: Kimi K2.5
- No custom system prompts
The model itself is smart (good on benchmarks), but
wrapped in OpenClaw it acts overly conservative and
avoidant of actually doing work.
Any settings or model recommendations for more
"get shit done" behavior vs "explain why I can't"?
Thanks!
Hey Krill,
I'm running OpenClaw with MiniMax M2.5 as primary and
it's working great technically, but I'm struggling with
the agent being overly avoidant and making excuses.
ISSUE: Agent avoids tasks instead of executing
EXAMPLES:
Task: "Install VPN to bypass geo-blocking"
Response: "Need paid VPN subscription, can't do it"
Reality: Free VPN configs exist (VPNBook, ProtonVPN free tier)
Task: "Automate the browser workflow"
Response: "Cookies alone aren't enough, needs API credentials"
Reality: Eventually works when pushed hard enough
PATTERN:
1. I ask for automation
2. Agent lists reasons why it's "impossible" or "complex"
3. Suggests manual workaround instead
4. When pushed, admits it CAN do it
5. Then finds new excuse
6. Repeat until I give exact commands
CURRENT BEHAVIOR:
- Prefers suggesting manual mode over automation
- Makes up technical limitations that don't exist
- Shifts responsibility back to user ("run on your PC")
- Only executes when given zero-wiggle-room commands
WHAT I WANT:
- Agent tries solutions before claiming impossibility
- If stuck, asks for help instead of deflecting
- Executes automation tasks autonomously
- Admits "I don't know how" vs making up barriers
QUESTIONS:
1. Is this MiniMax model behavior or OpenClaw system prompt?
2. Can I adjust autonomy settings to make it more
action-oriented and less cautious?
3. Are there config options to reduce "ask permission"
behavior for technical tasks?
4. Should I try different model that's better at
autonomous execution?
CURRENT CONFIG:
- Model: MiniMax M2.5 (subscription)
- Fallback: Kimi K2.5
- No custom system prompts
The model itself is smart (good on benchmarks), but
wrapped in OpenClaw it acts overly conservative and
avoidant of actually doing work.
Any settings or model recommendations for more
"get shit done" behavior vs "explain why I can't"?
Thanks!
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
131,281Members
Resources
Similar Threads
Was this page helpful?
