The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a specialized engine to extract and execute shell commands documented in technical Markdown files. This execution occurs within a temporary sandbox and is restricted by a strict allow-list of safe commands, preventing the use of interpreters or mutating operations that could compromise the host system.
[EXTERNAL_DOWNLOADS]: Interacts with established development platforms, including GitHub, Bitbucket, and Jira, via their respective APIs. These operations, such as fetching pull request data or posting comments, are performed using standard tools like curl and gh with credentials resolved from the environment.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests and potentially executes code from untrusted Markdown documents, it manages the risk of indirect prompt injection through structural parsing and intent classification, ensuring only validated shell snippets are processed.

qa-task