seo-backlinks
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONOBFUSCATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read local markdown files from an Obsidian vault or knowledge base for grounding and to append session logs to a file within that vault. This ingestion of untrusted data from the user's environment, combined with file-writing capabilities, creates an attack surface where malicious instructions embedded in the notes could influence agent behavior.
- Ingestion points: Reads notes from local directories, specifically targeting Obsidian vaults and .md files (SKILL.md).
- Boundary markers: None provided in the instructions to distinguish between developer instructions and data read from the vault.
- Capability inventory: File reading (markdown notes) and file writing (appending to an SEO action log) (SKILL.md).
- Sanitization: No validation or sanitization is specified for the content read from or written to the vault.
- [OBFUSCATION]: The skill contains a 'non-negotiable' instruction to the agent to replace all em dashes (U+2014) and en dashes (U+2013) with other punctuation like commas or colons. The stated purpose of this instruction is to remove 'the single most recognizable tell of AI-written text' and avoid having generated outreach emails or guest posts deleted by editors. This is a deliberate technique to conceal the AI's origin and evade automated or human detection of machine-generated content (SKILL.md).
Audit Metadata