Skip to content

Troubleshooting by Symptom

Use this page when you have an error and want the fastest fix path.

Quick Symptom Table

Symptom Likely Cause Fix
Pipeline not found Agent not compiled for that framework Run super agent compile <agent> --framework <framework>
OPENAI_API_KEY is required while using Google Runtime/provider mismatch Pass --cloud --provider google-genai --model gemini-2.5-flash on both compile and run
DefaultCredentialsError from Vertex Model/provider path resolved to Vertex instead of Google GenAI API-key flow Ensure provider/model pair is google-genai + Gemini model and GOOGLE_API_KEY is set
name 'false' is not defined JSON boolean leaked into generated Python Recompile with latest templates, then rerun
Exceeded maximum retries for output validation Structured output too strict for current model response Increase retries, reduce strictness, or switch to stronger model
StackOne tool returns 400 Bad Request Tool args or account/provider mismatch Verify STACKONE_ACCOUNT_IDS, action filters, and tool input args
DSPy program timed out Tool/model latency exceeded timeout Increase timeout env setting or simplify tools/model path
No such file ... playbook Missing generated sidecar/spec artifacts Re-run super agent compile ... and keep generated pipeline+sidecar together

Key Checks

# Verify keys in current shell
echo $GOOGLE_API_KEY
echo $OPENAI_API_KEY
echo $ANTHROPIC_API_KEY
echo $STACKONE_API_KEY
echo $STACKONE_ACCOUNT_IDS
# Recompile cleanly for target framework
super agent compile <agent_id> --framework <framework> --cloud --provider google-genai --model gemini-2.5-flash
# Run with matching provider/model flags
super agent run <agent_id> --framework <framework> --cloud --provider google-genai --model gemini-2.5-flash --goal "..."

StackOne-specific Checks

  • Confirm connector account is active for the target app (for example Calendly).
  • Start with identity calls first (*_get_current_user) before list/query calls.
  • For event listing calls, include explicit date/time window and timezone where required.

RLM Note

RLM is currently experimental. Unified sandbox support is coming soon. If an RLM path is unstable for your use case, keep a non-RLM fallback module in the same agent.