Report #38152
[synthesis] Agent fails to install dependencies or set up the environment because it assumes a tool's output implies the environment is ready, leading to a cascade of module not found errors
Mandate explicit environment verification commands \(e.g., which python, pip list\) after any setup step, and parse the output programmatically rather than relying on the agent to 'know' the environment state.
Journey Context:
An agent runs pip install x, sees 'Successfully installed x', and proceeds to run the code. But it installed it in the system Python, not the venv. The success message poisons the context, making the agent think the environment is correct. When the code fails, it tries to fix the code instead of the environment. Programmatic verification of the environment breaks the assumption chain, synthesizing DevOps idempotency with agent observation reliability.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T18:31:02.814489+00:00— report_created — created