The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill documentation suggests a setup command (ob login --email <email> --password '<password>') that leads to credentials being entered in cleartext on the command line, potentially exposing them in shell history files.\n- [EXTERNAL_DOWNLOADS]: The skill recommends installing the obsidian-headless package from npm and references a third-party repository (atomicmemory/llm-wiki-compiler), which are not part of the standard trusted resources.\n- [PRIVILEGE_ESCALATION]: Setup instructions include the use of sudo loginctl enable-linger, which modifies system-level user session behavior to allow background processes to continue after logout.\n- [PERSISTENCE]: The skill provides a systemd unit configuration to ensure knowledge base synchronization remains active across reboots and user sessions.\n- [DYNAMIC_EXECUTION]: The wiki's 'lint' functionality uses the execute_code tool to run Python scripts that scan the filesystem for structure and link integrity.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from web sources and external files for summarization, creating a surface for potential injection of malicious instructions.\n
Ingestion points: Data is processed through web_extract and stored in the raw/ folder.\n
Boundary markers: No specific boundary markers or 'ignore' instructions are used during content ingestion.\n
Capability inventory: The skill uses read_file, write_file, search_files, and execute_code.\n
Sanitization: No sanitization or filtering logic is provided for external content processing.

llm-wiki