linkedin-field-note
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill frequently generates and executes local Python scripts (e.g.,
build_<slug>_kit.py) to create Word documents. It also leverages PowerShell and Word COM objects for document pagination and manipulation, representing a high level of local system interaction. - [DATA_EXFILTRATION]: The skill is designed to read from sensitive local directories containing personal business content, brand tokens, and credentials (referenced as 'brand-tokens.md'). It then transmits content derived from these files to LinkedIn. This is the intended purpose, but it constitutes a bridge between private local data and a public network.
- [PROMPT_INJECTION]: The skill ingests 'Reactions and learning' from LinkedIn to update a ledger and influence future content proposals. This creates an indirect prompt injection surface where untrusted external comments could potentially manipulate the agent's logic or topic selection in later sessions.
- [COMMAND_EXECUTION]: The skill utilizes the
claude-in-chrometool to perform browser automation, including logging into LinkedIn and posting content, which involves complex interaction with the user's web session.
Audit Metadata