{"version":"2.1.290","anchor":"prompt-string-ablation-via-env-var-or-client-data","canonical_anchor":"prompt-string-ablation-via-env-var-or-client-data","heading":"Strings can be removed from prompts with an environment variable","tier":"use","area":"Agents","scope":"both","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.290\/e\/prompt-string-ablation-via-env-var-or-client-data","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.290","markdown":"### Strings can be removed from prompts with an environment variable\n\nSetting CLAUDE_CODE_REMOVE_PROMPT_STRINGS to a JSON list removes those strings from the prompts Claude Code sends to the model\n\n**Unclear.** Which prompts this removal applies to is not stated.\n\n**What**\n\nClaude Code can now remove chosen pieces of text from its prompts before they are sent to the model. A prompt here means the instructions and context Claude Code hands to the model on each turn. The list of text to remove comes from one of two places:\n\n- `CLAUDE_CODE_REMOVE_PROMPT_STRINGS`, an environment variable (a setting you put in your shell before starting Claude Code) holding a JSON array of strings\n\n- a `remove_prompt_strings` value supplied by the program that launched Claude Code\n\nWhen removal happens, Claude Code records how many strings were configured and how many were actually found.\n\n**Why**\n\nThis changes what the model actually receives, so it can be used to test how the model behaves without certain instructions. Anyone setting it should expect the model's behaviour to change accordingly.\n\n- Area: Agents\n- Tier: Use it now\n- Useful: 5\/5\n- Signal: 4\/5\n- Scope: both\n- Heads-up: no"}