The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes data from external sources, including Azure DevOps work item titles/tags and GitHub Dependabot alert metadata. This represents a potential surface where external content could influence the agent's actions.
Ingestion points: Data enters the agent context through the output of az boards query and gh api calls in Step 1.
Boundary markers: The skill does not implement specific boundary delimiters or safety instructions when interpolating this external data into subsequent steps.
Capability inventory: The skill utilizes several command-line tools including az, gh, npm, and git, as well as dynamic JavaScript execution via node -e.
Sanitization: There are no explicit validation or escaping mechanisms defined for the data used in the fix plan construction, relying on the agent's parsing logic.
Dynamic Script Execution: In Step 3, the skill uses node -e to programmatically modify package-lock.json. This pattern involves assembling executable JavaScript strings from variables derived from the fix plan (such as package names and tarball URLs). While effective for automation, this dynamic execution method is a security consideration as it creates a dependency on the integrity of the input variables.
Automated Dependency Management: The skill performs npm install to apply updates and verify fixes. Executing package manager commands that retrieve code from external registries (like the npm registry) is a core capability for this skill but inherently involves running code provided by external package maintainers.
Command Interpolation: Several steps involve interpolating variables (such as work item IDs or package names) directly into shell commands. This is a common pattern in CLI-based automation but requires that the agent properly handles the extracted identifiers to maintain command integrity.

security-alerts