gsd-mempalace-recall

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The file's opening line impersonates the "command system" and instructs the agent to ignore/stop reading the document and immediately begin executing Step 0—a deceptive, override-style instruction that attempts to influence agent behavior outside normal, transparent skill directions.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). SKILL.md Step 2–4 requires calling MemPalace search/knowledge-graph tools (mempalace_search/mempalace_*), whose returned drawer/fact text is then distilled into MEMORY-RECALL.md and subsequently consumed by the planner/agent LLM context, making it runtime ingestion of outsider-authored free text from a non-vetted external MemPalace server.