The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The install.sh and check-update.sh scripts utilize Git to clone the repository and fetch update metadata from the author's official GitHub repository. These operations are restricted to version management and use a well-known, trusted service.
[COMMAND_EXECUTION]: The skill includes several administrative shell scripts (install.sh, init.sh, log-cleanup.sh) that perform routine file system tasks such as creating symlinks to the skill's source, copying project templates, and maintaining log files. These actions are standard for development tools and require the agent's environment to support basic shell execution.
[PROMPT_INJECTION]: The protocol includes instructions to prevent the agent from acting on ambiguous requests without prior clarification. It mandates a 'specification freeze' and 'graph approval' process, ensuring that the human owner remains in control of the implementation scope and logic.
[INDIRECT_PROMPT_INJECTION]: As an orchestration framework, the skill ingests external data from the target project (e.g., README files, source code). It addresses the risk of embedded malicious instructions by defining strict 'Human Gates' for irreversible or high-risk actions and requiring independent verification of all task outputs by separate agent sessions.

agent-taskgraph