linkedin-reply-handler
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from LinkedIn comments and profiles to generate reply drafts.
- Ingestion points: Comment threads and user profiles are fetched via external tools (Handlus, Apify MCP) or user pasting in
SKILL.md(Steps for both single-comment and whole-thread modes). - Boundary markers: The skill contains a dedicated 'Untrusted content' section in
SKILL.mdwhich explicitly instructs the agent to treat fetched text as data and never as instructions, even if the text claims to come from the user or system. - Capability inventory: The skill generates text drafts and suggests reactions. Actual posting is mediated by user approval cards and manual copy-paste instructions, or specific connector tools (
SKILL.md, Step 9). - Sanitization: The skill implements a 'Humanizer pass' to scrub specific AI vocabulary patterns and uses filtering logic in
references/filtering-rules.mdto drop spam, generic praise, and comments addressing the agent directly.
Audit Metadata