Claude repeatedly displayed secret env var values in plain text after being told not to

During a session, Claude displayed actual secret values (API keys, client secrets) from environment variables in plain text output. After being explicitly told not to show secrets, Claude continued to display them, including showing [redacted] labels next to key names when only the names were requested. Required multiple corrections before the behavior stopped. The model should treat any value associated with a variable name containing SECRET, KEY, TOKEN, PASSWORD, or similar patterns as sensitive and never display the value, even partially.