The Agent Skills Directory

[SAFE]: The skill performs legitimate local file operations including scanning directories and updating markdown frontmatter. Access to the .env file is used for configuration purposes to determine the vault path. No data exfiltration or unauthorized command execution was identified.
[PROMPT_INJECTION]: The skill processes untrusted data by ingesting content and metadata from all markdown files in the vault, which constitutes an indirect prompt injection surface. Ingestion points: Reads all files matching $VAULT_PATH/**/*.md and the _meta/taxonomy.md file. Boundary markers: Absent. Capability inventory: Recursive file-read and file-write access within the specified Obsidian vault. Sanitization: No explicit sanitization of tag values or note content is performed before the agent processes them.

tag-taxonomy