Agent Execution
Common agent execution problems and how to resolve them.
"Skill not available" error
Symptom: A step fails with a message like "skill 'X' is not available."
Cause: The agent references a capability that is not registered in the current version of ThinkForge. This can happen if the agent was created with a newer version or if something changed in the skill registry.
Solutions:
- Update ThinkForge Desktop to the latest version
- If the skill was renamed, recreate the agent (the creation process will select the correct current skill)
"Agent has no execution steps"
Symptom: The agent fails immediately without running any steps.
Cause: The agent definition is empty or corrupted — it has no steps in its execution pipeline.
Solution: Delete the agent and recreate it through Tool Forge. A valid agent always has at least one execution step.
AI steps fail with API errors
Symptom: Steps that use AI processing fail with messages about prompts, keys, or provider errors.
Causes:
- No API key configured for the provider
- API key is invalid or expired
- Rate limits exceeded (multiple agent steps calling AI in quick succession)
- Provider is experiencing an outage
Solution: See AI Provider Keys for detailed troubleshooting of API key issues. Make sure the profile used for agent execution has a valid key.
File write errors
Symptom: The save step fails with permission or path errors.
Causes:
- No output folder configured on the deployment
- The output folder does not exist or is not writable
- The file path contains unsafe characters or patterns
Solutions:
- Set an output folder in the deployment inspector (drag a folder or use Browse)
- Make sure the selected folder exists and your user account can write to it
- If file saving is not needed, disable it in agent settings (output still displays on screen)
External API integration errors
Symptom: Steps that connect to external services (GitHub, Slack, ClickUp, etc.) fail with authentication or configuration errors.
Cause: The agent uses an API integration that has not been configured with credentials or target parameters.
Solution:
- The first time you run an agent with API integration, it prompts you for credentials and target parameters (repository name, list ID, channel, etc.).
- Enter the required information — it is saved for future runs.
- If credentials were revoked or changed, update them in agent settings.
"No active project"
Symptom: Agent fails with a "no active project" error.
Cause: The agent was triggered (from the extension, command menu, or a trigger) without a project being open in ThinkForge.
Solution: Open a project in ThinkForge Desktop before running agents. Agents that use project memory or file operations need an active project context.
User cancelled / user rejected step output
Symptom: Agent stops mid-execution with a cancellation message.
Cause: In guided mode, you rejected a step's output, which aborts the remaining pipeline. This is expected behavior — not an error.
Solution: If you want the agent to run without pausing for approval, switch to autonomous mode in Agent Settings → Execution Mode. If you do want approval but made a mistake, simply rerun the agent.
Agent runs from the browser extension fail
Symptom: Running an agent from the Chrome extension produces an error.
Causes:
- Desktop is not connected
- No project is open
- The agent referenced does not exist (stale name or ID)
Solutions:
- Verify the extension shows "Connected" in TF Desktop settings
- Open a project in ThinkForge Desktop
- Refresh the agent list in the extension; ensure the agent name matches exactly
General debugging steps
When an agent fails:
- Read the error message on the failed step card — it usually identifies the specific problem
- Check that a project is open and the correct one is active
- Verify API keys are configured if the agent uses AI steps
- Confirm input sources are assigned to the deployment
- Try running a simpler agent to isolate whether the issue is with a specific skill or with general configuration
Related
- Run Agents — How to configure and run agents
- Permissions — What agents can and cannot do
- AI Provider Keys — API key troubleshooting
- File Ownership — How file operations work for agents