The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references an external utility named defuddle for parsing HTML into Markdown and points to the author's GitHub repository (github.com/drvoss/llm-wiki-skill) for optional helper scripts. These resources are provided as optional tooling or for user-initiated ingestion workflows.
[COMMAND_EXECUTION]: The instructions direct the agent to use shell commands (PowerShell/Bash) for local file management, including reading schema files and searching the wiki directory. These actions are scoped to the local wiki path defined by the user.
[PROMPT_INJECTION]: The skill processes untrusted external data including URLs, transcripts, and pasted text for ingestion into the wiki. This creates a surface for indirect prompt injection where malicious instructions within the source data could attempt to influence agent behavior during synthesis or querying.
Ingestion points: External URLs, user-provided files, and transcripts (SKILL.md).
Boundary markers: None explicitly defined for isolating untrusted data during the ingestion phase.
Capability inventory: Local file reading/writing and execution of the defuddle parsing tool (SKILL.md).
Sanitization: Uses defuddle to normalize HTML to Markdown, which provides structural normalization but is not a dedicated security sanitizer.

llm-wiki