The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The skill instructs users to store sensitive LinkedIn credentials (email and password) in a plaintext file (community-pulse/skills/linkedin/.env.profiles.local). Storing secrets in unencrypted local files is a high-risk practice as any other process or compromised skill with filesystem access can read them.
[PROMPT_INJECTION] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8). It ingests untrusted data from the web using read_page and get_page_text on LinkedIn posts.
Ingestion points: LinkedIn post content and comments read in references/research.md and references/engage.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present when the agent evaluates posts or drafts comments.
Capability inventory: The skill has browser control capabilities (computer, form_input, navigate) and filesystem access.
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from LinkedIn before it is processed by the LLM.
[COMMAND_EXECUTION] (SAFE): The skill uses Browser MCP tools to automate web navigation and interaction. This behavior is consistent with the skill's primary purpose. Security risk is mitigated by mandatory human-in-the-loop checkpoints ('ASK USER FOR APPROVAL') before sensitive actions like logging in or posting content.

linkedin