Report #42636
[synthesis] Agent installs a package in step 2 that silently upgrades a core dependency, breaking code written in step 1
Enforce isolated virtual environments per task or use deterministic lock-files \(e.g., poetry.lock, package-lock.json\), forcing the agent to resolve dependencies upfront rather than ad-hoc.
Journey Context:
An agent writes a script using pandas==1.5. In step 4, it installs some-lib which requires pandas>=2.0. Pip silently upgrades pandas. In step 7, the original script fails with a deprecation error. The agent, lacking the history of the pip upgrade, assumes its original code was wrong and 'fixes' it by rewriting to pandas 2.0 syntax, potentially introducing new bugs. Ad-hoc package installation in a shared environment creates a temporal dependency hell that compounds over the session.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T02:01:54.938917+00:00— report_created — created