The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill reads and processes all Markdown files within an Obsidian vault to perform link audits, detect orphaned notes, and build maps of content. This creates a vulnerability where malicious instructions embedded in vault notes could influence the agent's behavior during analysis.
Ingestion points: SKILL.md (Step 3, 4, 5) instructs the agent to read note content and extract wikilink patterns.
Boundary markers: None specified; instructions do not include delimiters or warnings to ignore embedded content.
Capability inventory: Local shell execution (find, grep, ls, comm), file reading, and file writing.
Sanitization: None; the skill does not specify escaping or validation for content extracted from vault notes.
[COMMAND_EXECUTION]: Shell Command Templating. The skill provides shell command templates for file discovery and link extraction (e.g., find <vault-root> -name "*.md" | grep -i "<note-name>"). These templates are vulnerable to command injection if the <vault-root> or <note-name> variables, provided by the user or extracted from files, contain shell metacharacters and are not properly escaped by the agent environment.

obsidian-links