The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the web via web_extract and processes it into the wiki structure. This creates a surface where malicious instructions embedded in source materials could influence future agent actions when the wiki is queried.
Ingestion points: raw/ directory files populated via web_extract and external URL downloads.
Boundary markers: The skill uses YAML frontmatter to track sources and integrity (sha256), but lacks explicit 'ignore' delimiters for the ingested body content during summarization.
Capability inventory: File read/write, search, and Python code execution via execute_code.
Sanitization: No explicit sanitization or filtering of ingested text is defined in the instructions.
[DYNAMIC_EXECUTION]: The skill provides instructions for the agent to use execute_code to generate and run Python scripts for linting tasks, such as detecting orphan pages and broken wikilinks.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the installation of the obsidian-headless NPM package and the llm-wiki-compiler tool from an external GitHub repository.
[PERSISTENCE]: Instructions are provided to create a systemd user service (obsidian-wiki-sync.service) to maintain a background process for continuous synchronization of the knowledge base.
[COMMAND_EXECUTION]: The skill utilizes various shell commands for environment configuration (WIKI_PATH), file system management, and synchronization tool interactions (ob sync, ob login).

llm-wiki