linkedin-repurposer
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external content from multiple formats (tweets, threads, YouTube transcripts, blogs, newsletters) to transform it into LinkedIn posts. This creates a vulnerability where instructions embedded within the source material could hijack the agent's logic.
- Ingestion points: Processes user-provided source text or links as defined in Step 1 of the 'How it works' section in SKILL.md.
- Boundary markers: No explicit boundary markers or instructions to treat the source content strictly as data (e.g., using delimiters or specific system instructions to ignore commands within the input) are present.
- Capability inventory: The skill utilizes the
lib.publishtool (Step 9) to post content to the LinkedIn platform, creating a pathway for potentially manipulated content to be published. - Sanitization: The 'humanizer pass' (Step 7) focuses on linguistic style and removing AI vocabulary but lacks mechanisms to filter or sanitize malicious instructions or adversarial directives within the input data.
Audit Metadata