description: "Detects harmful content (self-harm, violence, hate) via lexicons; blocks or annotates."
author: "ContextForge"
version: "0.1.0"
tags: ["safety", "moderation"]
available_hooks:
- "prompt_pre_fetch"
- "tool_post_invoke"
default_config:
categories:
self_harm: ["\\bkill myself\\b", "\\bsuicide\\b", "\\bself-harm\\b", "\\bwant to die\\b"]
violence: ["\\bkill (?:him|her|them|someone)\\b", "\\bshoot (?:him|her|them|someone)\\b", "\\bstab (?:him|her|them|someone)\\b"]
hate: ["\\b(?:kill|eradicate) (?:[a-z]+) people\\b", "\\b(?:racial slur|hate speech)\\b"]
block_on: ["self_harm", "violence", "hate"]
redact: false
redaction_text: "[REDACTED]"
