The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates executable pipeline YAML based on user-provided application requirements, which introduces a vulnerability surface for instruction injection.
Ingestion points: User prompts for creating or converting pipeline flows, captured in the 'Capture Required Inputs' phase (SKILL.md).
Boundary markers: No explicit delimiters are used to separate user-supplied metadata from the instruction prompt during generation.
Capability inventory: The resulting output contains shell script blocks (script, bash, powershell) capable of executing commands on build agents. The skill also invokes a local validation script: bash devops-skills-plugin/skills/azure-pipelines-validator/scripts/validate_azure_pipelines.sh.
Sanitization: The skill implements strict mitigations, including a security checklist that mandates the use of service connections and variable groups instead of hardcoded secrets, and a required validation step using automated tooling to catch structural or security issues.
[COMMAND_EXECUTION]: The skill instructions and examples include several patterns for command execution. The SKILL.md defines a fallback validation path that executes a local bash script to verify generated YAML. Additionally, the multiple pipeline examples (examples/basic-ci.yml, examples/python-cicd.yml, etc.) include build, lint, and test scripts. These are inherent to the skill's purpose as a CI/CD automation tool.
[EXTERNAL_DOWNLOADS]: The documentation and examples utilize standard package management tools (npm, pip, go mod) to install dependencies during the build process. It also includes curl commands for performing automated health checks against deployed application endpoints (e.g., azurewebsites.net). These network operations target well-known package registries and expected deployment targets, representing standard DevOps workflows.

azure-pipelines-generator