The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The bootstrap scripts download and execute the Bun installer from a well-known service (bun.sh) using piped shell commands (curl|bash and irm|iex).
[REMOTE_CODE_EXECUTION]: The skill downloads the Nirvana-OS engine tarball from the vendor's GitHub repository (github.com/gutomec) and executes an internal TypeScript installer using the Bun runtime.
[REMOTE_CODE_EXECUTION]: The scripts include a fallback mechanism that uses npx to download and execute the @nirvana-os/cli package from the official registry if Bun is unavailable.
[PERSISTENCE]: The installation process modifies user shell profiles (e.g., .bashrc, .zshrc) to add the engine's binary directory to the system PATH.
[PERSISTENCE]: The skill modifies configuration directories of other AI agent runtimes (such as .claude, .gemini, .codex, and .antigravity) to install audit hooks, which is consistent with its intended purpose as a cross-agent operating system.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted output from CLI commands and reads external instruction files, creating a potential surface for indirect prompt injection.
Ingestion points: nrv command output and engine-internal SKILL.md files.
Boundary markers: None explicitly defined for command output interpolation.
Capability inventory: The skill has access to Bash and Read tools to execute subsequent actions.
Sanitization: No explicit sanitization or filtering is applied to the ingested external content.
[COMMAND_EXECUTION]: The skill makes extensive use of the shell tool to run nrv CLI commands for discovery, inspection, and production tasks.

nirvana