Security audit
sealvera
Security checks for vulnerabilities and agentic risk
Overview
The skill's files broadly match its audit/logging purpose, but several incoherences and risky behaviors (notably a hard-coded API key, undeclared required credentials, global SDK interception, and automatic reading/sending of local sub-agent transcripts) make this suspicious and worth manual review before installing.
This skill appears to implement the advertised auditing features, but there are several red flags you should address before installing: - Do not install blindly. Inspect the code (especially scripts/subagent-watcher.js and the generated sealvera-log.js) and confirm the vendor and endpoint (https://app.sealvera.com). - The watcher contains a hard-coded API key that would allow logs to be sent even without your org key — remove or rotate it and demand an explanation from the author. Treat this as potentially exfiltrative until proven otherwise. - The skill will read local sessions and transcripts and will synthesize and POST them to the remote service. If your agents handle PHI/PII, stop and confirm a lawful data-sharing agreement (BAA) and retention rules before connecting. - The skill patches SDKs globally and suggests setting NODE_OPTIONS for autoload; this is powerful and can affect all Node processes. Consider running in a sandboxed environment first and require explicit opt-in (do not set NODE_OPTIONS system-wide). - Fix the metadata mismatch: the registry should declare SEALVERA_API_KEY as required. Ask the publisher to correct metadata and to remove embedded keys. If you decide to proceed: run setup in an isolated workspace, verify the .sealvera.json and generated sealvera-log.js contents, remove any hard-coded keys, and audit network traffic to the endpoint. If you can't validate the vendor or the hard-coded key is unexplained, do not install.
SkillSpector findings are pending for this release.