The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to interact with the environment via the 'twg' CLI tool. It provides legitimate guidance for locating the tool's binary in standard user-local directories such as '$HOME/.local/bin/twg' on Unix systems or the equivalent 'LOCALAPPDATA' path on Windows, which is standard behavior for Atlassian developer tools.- [INDIRECT_PROMPT_INJECTION]: The skill processes external data from Jira issue summaries and descriptions, which are potential ingestion points for indirect injection. To mitigate this risk, the instructions explicitly require the agent to use output filters like '--agent-fields' to limit the information passed to the model, reducing the attack surface. Additionally, it defines strict 'Evaluation Gates' to ensure decisions are based on objective metadata rather than potentially manipulated text.- [SAFE]: The skill is authored by the vendor (Atlassian) and governs interactions with their own platform. The instructions enforce multiple layers of safety: resolving workitem keys and site URLs before mutation, reading the current state before applying updates, and obtaining user approval for any workflow transitions or required field population.

twg-jira