The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and uses it to drive agent behavior without explicit sanitization or boundary markers.
Ingestion points: Reads .liftli/voice.md and ~/.liftli/voice.md for stories and proof; gathers 'raw material' from chat history and user resumes; calls a strategist_briefing tool and follows its instructions.
Boundary markers: No delimiters or instructions are present to ensure the agent ignores malicious commands that might be embedded in the resumes or voice files.
Capability inventory: The skill has the ability to read local files, call external tools, and present instructions for command execution.
Sanitization: There is no logic for validation or escaping of the ingested material before processing.
[DATA_EXFILTRATION]: The skill accesses content from the user's home directory path ~/.liftli/voice.md. Although this is a vendor-specific subfolder, accessing files within the home directory is a form of sensitive data exposure.
[COMMAND_EXECUTION]: The skill instructs the user to execute commands such as npx skills add liftli-ai/linkedin-agent-skills --skill linkedin-voice, claude mcp add ..., and claude mcp login liftli. These instructions facilitate the installation and execution of external software and configurations.

linkedin-about-generator