Nexus Brain
ReviewAudited by ClawScan on May 10, 2026.
Overview
Nexus Brain has a coherent SRE purpose, but it describes broad log-to-AI and service-restart workflows without enough scoping or guaranteed approval.
Install only if you are comfortable giving the skill access to operational logs and SRE tooling. Before use, require approvals for restarts, restrict which services it may touch, preview log snippets before they leave your machine, and verify the local opencode, docker, pm2, and psutil installations.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The agent could affect running services if allowed to act on its diagnosis, potentially causing downtime or changing the local runtime environment.
The skill contemplates service restarts through operational tooling, while approval is conditional and no service/container/process scope or rollback limits are specified.
`infra failures (restart service)` ... `Critical actions still require /approve if configured` ... `Ensure docker, pm2, and opencode are installed.`
Require explicit approval for every restart or recovery command, restrict the allowed docker/pm2 targets, and document rollback or dry-run behavior.
Private application, infrastructure, or user data in logs may be sent to an external AI service even after best-effort redaction.
Operational logs can contain sensitive data, and the artifacts do not clearly bound which logs are collected, how much is sent, provider retention, or required user approval before transmission.
`Automatically fetches and redacts relevant logs for the AI` and `sends redacted log snippets to your configured AI reasoning service`
Limit log sources, preview the exact log excerpts before sending, require user approval for outbound log sharing, and verify the AI provider's retention and privacy policy.
Users may overestimate how much sensitive information is removed from logs before they are sent to the AI service.
The redaction code covers only a small set of secret-like key names and email addresses, so the privacy protection is best-effort rather than comprehensive.
patterns = [(r'([Pp]assword|[Ss]ecret|[Tt]oken|[Aa]pi[Kk]ey)["\\s:=]+[^\\s,"]+', r'\\1: [REDACTED]'), ...]
Treat redaction as partial, add broader secret-detection rules, and manually review log excerpts before external analysis.
Security depends partly on the user's local installations and the provenance of those tools.
The skill depends on manually installed external packages and binaries, with no pinned versions in the artifact.
`pip install psutil` and `Ensure docker, pm2, and opencode are installed.`
Install dependencies from official sources, pin package versions where possible, and verify the opencode binary before use.