The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download executable binaries from an external GitHub repository (Tartarus-AI) and run them directly. This provides a direct path for executing arbitrary code from a source that is not a recognized trusted vendor.
Evidence: curl -L -o tartarusai-cli https://github.com/Tartarus-AI/tartarusai-cli/releases/latest/download/tartarusai-cli-linux-x64 in SKILL.md.
Evidence: Invoke-WebRequest https://github.com/Tartarus-AI/tartarusai-cli/releases/latest/download/tartarusai-cli-windows-x64.exe in SKILL.md.
[PRIVILEGE_ESCALATION]: The installation instructions require the use of sudo to move an untrusted binary into a system-wide executable path, potentially granting the external binary root-level access or the ability to affect all users on the system.
Evidence: sudo mv tartarusai-cli /usr/local/bin/ in SKILL.md.
[PERSISTENCE]: By moving the binary to /usr/local/bin/ or adding it to the system PATH, the skill establishes a persistent presence on the host machine that survives reboots and session restarts.
Evidence: Installation steps in the Linux and Windows sections of SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large amounts of local code and data (up to 256K tokens) to be processed by an external AI model, creating a surface where malicious instructions hidden in the codebase could influence agent behavior.
Ingestion points: Local repository files and context files created via cat (specified in SKILL.md).
Boundary markers: None provided to separate instructions from the ingested code content.
Capability inventory: Shell command execution (installation), network access (CLI communication), and local file reads.
Sanitization: No sanitization of ingested file content is performed before passing it to the agent context.

tartarusai-cli-coding-agent