Agent keeps "forgetting" it has skills - anyone else seeing this?
Hey everyone,
I've been running into a frustrating pattern with my agent (Chad) and wondering if anyone has ideas or similar experiences.
The problem: My agent has several installed skills but consistently tries to do things manually instead of using them, even though they're listed in <available_skills> in the system prompt.
Examples:
• Has searxng skill for web search → tries to use web_search tool (which needs Brave API key I don't have) and tells me "search doesn't work"
• Has skill-creator skill → when I ask to create a skill, writes everything from scratch instead of following the skill's structure
• Has self-improver skill for logging errors/lessons → makes a mistake, I correct it, but doesn't log it to .learnings/
The skills are there. The instructions say to scan <available_skills> before acting. But somehow the "instinct" to go straight to the solution overrides the checkpoint step.
What I've tried:
• Adding reminders to MEMORY.md
• Logging lessons about "use your skills!"
• The agent even diagnosed itself and says it's a habit/checkpoint problem
But it keeps happening. Anyone found a reliable way to make skills "stick" in agent behavior? Some kind of forcing function?
Maybe the skill matching logic could be more aggressive? Or skills could somehow hook into tool calls?
Curious what others think
I've been running into a frustrating pattern with my agent (Chad) and wondering if anyone has ideas or similar experiences.
The problem: My agent has several installed skills but consistently tries to do things manually instead of using them, even though they're listed in <available_skills> in the system prompt.
Examples:
• Has searxng skill for web search → tries to use web_search tool (which needs Brave API key I don't have) and tells me "search doesn't work"
• Has skill-creator skill → when I ask to create a skill, writes everything from scratch instead of following the skill's structure
• Has self-improver skill for logging errors/lessons → makes a mistake, I correct it, but doesn't log it to .learnings/
The skills are there. The instructions say to scan <available_skills> before acting. But somehow the "instinct" to go straight to the solution overrides the checkpoint step.
What I've tried:
• Adding reminders to MEMORY.md
• Logging lessons about "use your skills!"
• The agent even diagnosed itself and says it's a habit/checkpoint problem
But it keeps happening. Anyone found a reliable way to make skills "stick" in agent behavior? Some kind of forcing function?
Maybe the skill matching logic could be more aggressive? Or skills could somehow hook into tool calls?
Curious what others think
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Similar Threads
Was this page helpful?
