Is blucli safe?
https://clawhub.ai/steipete/blucli
blucli is a low-risk skill that provides CLI instructions for controlling Bluesound/NAD audio players. The skill files contain no prompt injection, data exfiltration, or malicious code patterns. The primary risk vector is the Go binary installation from a third-party GitHub repository (github.com/steipete/blucli), which is a standard pattern but requires trust in the upstream author. All monitoring signals (network, filesystem, canary files, process execution) are clean.
Category Scores
Findings (5)
MEDIUM Third-party Go binary installation -25 ▶
The skill's metadata specifies installation of a Go binary via 'go install github.com/steipete/blucli/cmd/blu@latest'. This compiles and installs code from an external GitHub repository. While this is a standard Go installation pattern, it means the user is trusting the upstream repository author (steipete) to not include malicious code in the binary. The binary is installed at install time, not at skill-load time, and the Go module system provides some supply chain protections (checksumdb, module proxy).
LOW External homepage URL reference -5 ▶
The skill references an external homepage URL (https://blucli.sh) in the metadata. This is informational only and does not instruct the agent to fetch or interact with the URL. No prompt injection risk, but noted for completeness.
INFO Device confirmation instruction -5 ▶
The skill instructs the agent to 'Confirm the target device before changing playback'. This is a reasonable safety measure to prevent unintended audio playback changes, not a manipulation attempt.
LOW Network device discovery capability -5 ▶
The 'blu devices' command performs local network discovery to find Bluesound players. While this is the intended functionality, it does mean the skill causes network scanning activity on the local subnet. This is expected behavior for the skill's stated purpose.
INFO Minimal skill surface area -5 ▶
The skill has an extremely small surface area — only a SKILL.md with CLI usage instructions, a _meta.json with version info, and origin tracking. No source code, no scripts, no hooks, no submodules. The risk profile is dominated by the external Go binary install rather than anything in the skill files themselves.