nvm-shell-integration
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PERSISTENCE]: The skill provides scripts intended to be added to shell profile files (
~/.bashrc,~/.config/fish/config.fish,~/.zshrc). - Evidence:
examples/auto-use-bash.md,examples/auto-use-fish.md, andexamples/auto-use-zsh.mdcontain functions and hooks that execute every time a shell session starts or a directory is changed. - [INDIRECT_PROMPT_INJECTION]: The provided scripts ingest data from
.nvmrcfiles found in the user's filesystem, which serves as an untrusted input surface. - Ingestion points: The scripts read
.nvmrccontent using$(<.../.nvmrc),cat, or$(cat ...)inexamples/auto-use-bash.md,examples/auto-use-fish.md, andexamples/auto-use-zsh.md. - Boundary markers: None present; the scripts directly use the file content as arguments for
nvmcommands. - Capability inventory: The skill uses
nvm installandnvm use, which can execute downloads and modify the environment. - Sanitization: The scripts use standard shell quoting for variables (e.g.,
"${nvm_version}") which mitigates basic word splitting but passes the full file content to thenvmbinary. - [COMMAND_EXECUTION]: The skill relies on shell functions and aliases to intercept the
cdcommand and executenvmlogic. - Evidence:
examples/auto-use-bash.mddefines acdnvmfunction and aliases it tocd.examples/auto-use-zsh.mdusesadd-zsh-hook chpwdto trigger logic on directory changes.
Audit Metadata