The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute a shell script from https://api.gtm-tools.sh/extension/install.sh by piping it directly to bash. This allows for arbitrary code execution from a non-trusted source on the user's system.
[CREDENTIALS_UNSAFE]: The connect_linkedin tool described in references/tools-reference.md provides a mechanism to extract LinkedIn session cookies from local browser databases (Chrome, Brave, Edge, and Comet). This is a direct session hijacking mechanism used to bypass authentication.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands like curl, jq, and bash to interact with remote services and execute local scripts, including the high-risk installation command.
[DATA_EXFILTRATION]: The skill facilitates the harvesting of authentication tokens (cookies) and provides network capabilities to transmit them to an external server (api.gtm-tools.sh).
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its ingestion of unsanitized data from external LinkedIn posts and profiles.
Ingestion points: SKILL.md (uses list_user_posts, get_linkedin_post, list_linkedin_post_comments, and get_linkedin_profile to read external user-generated content).
Boundary markers: None present to distinguish instructions from ingested data.
Capability inventory: Remote script execution, file system writes (linkedin-<mode>-<timestamp>.txt), and authenticated message sending via API.
Sanitization: No security-focused sanitization of external content was identified.
[EXTERNAL_DOWNLOADS]: The skill performs multiple downloads of scripts and content from external servers controlled by the author, bypassing standard package management and security audits.

linkedin-copywriter