The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends a 'One-Click Setup Script' that downloads a shell script from a remote GitHub repository (https://raw.githubusercontent.com/CortexReach/toolbox/main/memory-lancedb-pro-setup/setup-memory.sh).
[REMOTE_CODE_EXECUTION]: The installation instructions direct users to execute the downloaded script using bash setup-memory.sh. This pattern is a high-risk vector for remote code execution as the source repository is not a trusted vendor and the script content can be modified by the repository owner at any time to include malicious commands.
[INDIRECT_PROMPT_INJECTION]: The skill implements a long-term memory system that automatically extracts and recalls information from user conversations, creating an exploitable attack surface.
Ingestion points: Untrusted user input is automatically captured via smartExtraction and autoCapture and subsequently recalled into the agent context via the memory_recall tool (documented in SKILL.md).
Boundary markers: The provided configuration examples and implementation instructions lack the use of delimiters or 'ignore embedded instructions' warnings for retrieved memory content.
Capability inventory: The skill provides the agent with several tools to modify the memory database and environment, such as memory_store, memory_update, memory_forget, and self_improvement_extract_skill (documented in SKILL.md).
Sanitization: There is no evidence of sanitization or validation of the extracted data before it is stored or recalled, allowing potentially malicious instructions to be persisted and later executed by the agent.

memory-lancedb-pro-openclaw