nixos-coding-agents

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill guides the user to fetch and run software from external GitHub repositories, including numtide/llm-agents.nix, nothingnesses/agent-images, and 0xferrous/agent-box. It provides specific instructions for adding the Numtide binary cache to the system configuration.
  • [COMMAND_EXECUTION]: The instructions include shell commands for the user to invoke agents and manage containers, such as nix run, podman run, and ab spawn. The skill repeatedly advises the user to authorize or check these commands before execution and warns about the permissions granted to agents.
  • [INDIRECT_PROMPT_INJECTION]: By facilitating the use of AI agents for coding tasks, the skill creates a surface for indirect prompt injection where an agent might process malicious instructions embedded in a project's source code.
  • Ingestion points: Project directories mounted as workspaces or accessed via Git worktrees (referenced in SKILL.md and references/security.md).
  • Boundary markers: The skill promotes the use of containers and worktrees containing only tracked files to limit the agent's reach, but does not specify prompt-level delimiters for data processing.
  • Capability inventory: Depending on the configuration, agents may have access to the local shell, specific mounted directories, and the network.
  • Sanitization: No specific content sanitization is described; security relies on environment isolation and the agent's own safety protocols.
  • [DYNAMIC_EXECUTION]: The skill uses Nix to dynamically build and execute code from flakes and OCI images, and it describes how to create custom agent images at runtime using mkAgentImage.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 09:44 AM