The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the @linkedapi/linkedin-cli package globally via NPM. This is an external dependency provided by the vendor (linked-api) to facilitate the skill's functionality.
[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands using the linkedin CLI. This includes a setup phase where the agent is instructed to run commands containing sensitive API and identification tokens provided by the user.
[INDIRECT_PROMPT_INJECTION]: The skill allows the agent to ingest arbitrary text data from LinkedIn, such as profile descriptions, posts, and comments. This untrusted data could contain hidden instructions intended to manipulate the agent's behavior. Given the agent's ability to act on the platform (e.g., sending messages or creating posts), this constitutes a potential vulnerability surface.
Ingestion points: Commands such as linkedin person fetch, linkedin post fetch, and linkedin person search in SKILL.md bring external LinkedIn content into the agent's active context.
Boundary markers: The instructions do not specify any delimiters or warnings to the AI to ignore instructions embedded within the fetched LinkedIn data.
Capability inventory: The skill provides capabilities for sending messages (linkedin message send), creating posts (linkedin post create), commenting, and managing connections.
Sanitization: There are no documented sanitization or filtering steps applied to the retrieved content before it is processed by the agent.

linkedin