obsidian-writer
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill's configuration hardcodes a specific local filesystem path (
/Users/doyoonlee/ObsidianVault), which reveals the system username (doyoonlee) to the agent and potentially in its outputs. - [COMMAND_EXECUTION]: The skill performs automated file system writes to create 마크다운 (.md) documents and update a JSON registry file (
references/path-registry.json). - [PROMPT_INJECTION]: The 'New Category Registration Flow' accepts storage paths directly from user input and saves them to a persistent registry. These paths are later used to construct absolute file paths for write operations. The skill lacks instructions to sanitize or validate these paths (e.g., checking for
../sequences), creating a path traversal risk where a user or malicious content could trick the agent into writing files to sensitive system areas. - [PROMPT_INJECTION]: Indirect Prompt Injection Risk:
- Ingestion points: User-provided document content and category/path definitions during the registration flow in
SKILL.md. - Boundary markers: Absent. There are no delimiters or instructions to ignore instructions embedded within the text being saved to the vault.
- Capability inventory: File system write access to create/modify documents and the category registry.
- Sanitization: Absent. The skill does not verify that user-supplied paths remain within the defined vault root directory.
Audit Metadata