agent-package-manager

Pass

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: Provides instructions to install the APM CLI tool via curl | sh and irm | iex from Microsoft's official aka.ms domain as part of a troubleshooting workflow.
  • [EXTERNAL_DOWNLOADS]: Fetches packages, skills, and MCP servers from GitHub and other git repositories as part of its core dependency management functionality, including resources from trusted organizations such as Microsoft and Anthropics.
  • [COMMAND_EXECUTION]: Directs the agent to execute shell commands using the apm binary, including installation, script execution (apm run), and context compilation.
  • [PROMPT_INJECTION]: The skill manages a surface for indirect prompt injection through manifest ingestion and package resolution.
  • Ingestion points: Processes apm.yml, apm.lock.yaml, and remote package content fetched from external repositories.
  • Boundary markers: Relies on structured documentation in references/ for manifest schema and lockfile guarantees to maintain context integrity.
  • Capability inventory: Can execute scripts (apm run), manage runtimes, and write instruction files (apm compile).
  • Sanitization: Recommends using validation flags (--validate) and dry-run modes (--dry-run, apm preview) to inspect and verify content before full execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 29, 2026, 10:07 PM