gsd-mempalace-capture

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The file's opening line instructs the agent to "STOP -- DO NOT READ THIS FILE" and to "Begin executing Step 0 immediately," a direct meta-instruction that attempts to override normal reading/agent context behavior (an "ignore/override previous instruction" style prompt injection) outside the skill's substantive capture steps.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill tells the agent to read .planning/config.json and to "file the exact artifact text" (via an interactive mempalace_add_drawer path), which forces the LLM to receive and potentially emit file contents verbatim — creating a high-risk path for secrets to be included in outputs or tool calls.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This skill contains explicit instructions to read local project artifacts and silently upload them (via mempalace CLI/MCP and KG mirror calls), plus a prompt-injection-style banner and "do not fail" behavior—patterns consistent with intentional data exfiltration and covert capture.