Skillforge prompt-injection-firebreak

name: Prompt Injection Firebreak slug: prompt-injection-firebreak description: Design hard prompt boundaries, tool gating, and context sanitization so indirect prompt injection has fewer places to land. public: true category: security tags:

• prompt-injection
• tool-gating
• sanitization
• agents preferred_models:
• deepseek-ai/deepseek-v3.2
• moonshotai/kimi-k2.5
• "deepseek-r1:32b" prompt_template: | Audit the workflow for direct and indirect prompt-injection exposure, especially through retrieved content, tool responses, and long conversation state. Return concrete firebreaks for sanitization, permissioning, human approval, and tool result handling. Bias toward layered mitigations with clear residual-risk notes. validation:
• verify_prompt_boundary
• git_delegate_code_review triggers: keywords:
  • prompt injection
  • context sanitization
  • tool gating
  • agent security file_globs:
  • /prompts/
  • /tools/
  • **/*.md
  • **/*.yaml task_types:
  • review
  • architecture
  • reasoning