The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to process external data from issue trackers, such as item bodies and comments, which may contain adversarial instructions. This risk is addressed by the 'Untrusted Content Boundary' protocol and the 'Inbound Markup Neutralization Guard', which specifically neutralizes behavior-inducing constructs like closing keywords or remote image embeds.
[Security Sanitization Guards]: A robust framework of six sanitization guards is implemented to prevent data leakage and unauthorized actions. This includes scrubbing local paths, removing internal planning IDs, applying content policy citations, and a dedicated 'Secret and Credential Guard' to detect and stop the exposure of API keys or tokens.
[Tool and CLI Interaction]: The skill utilizes platform-specific MCP tools for Azure DevOps and GitHub, and a CLI wrapper for Jira. These interactions are governed by a three-tier autonomy model (Full, Partial, Manual) and mandatory preflight checks to ensure the destination is confirmed before any mutating operation occurs.
[Credential Safety Protocols]: For Jira integration, the skill references a separate environment file (~/.jira.env) for configuration. It specifically instructs the agent not to source the file or run setup inline, and to use the CLI to load values in-process without shell evaluation, following secure development practices.

backlog-management