The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides bootstrap commands that download and execute scripts directly from the internet using shell piping (curl | sh and irm | iex). These commands target official Microsoft URL shorteners (aka.ms).
[EXTERNAL_DOWNLOADS]: The instructions reference external resources, including installation binaries and documentation, hosted on Microsoft's official GitHub Pages (microsoft.github.io) and shortcut domains.
[DYNAMIC_EXECUTION]: The skill supports executing project-specific scripts defined in the apm.yml manifest through the apm run and apm preview commands. This allows for the execution of arbitrary commands stored within the project configuration.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external manifests (apm.yml) and dependency trees from remote repositories. This represents an indirect injection surface where untrusted data could influence agent behavior.
Ingestion points: The agent reads configuration from apm.yml, apm.lock.yaml, and remote package metadata during installation (SKILL.md, Step 1).
Boundary markers: The skill documentation notes that apm install performs security scans for hidden Unicode and content hash verification to mitigate risks (SKILL.md, Step 3.10).
Capability inventory: The APM CLI tools used by the skill can perform file system writes, network requests for package fetching, and shell command execution via defined scripts (references/command-workflows.md).
Sanitization: The skill describes built-in enforcement policies, such as blocking transitive MCP servers and verifying content hashes during deployment (references/manifest-and-lockfile.md).

agent-package-manager