@chozzz/vargos@3.2.20
Moved: please install @vargos-labs/vargos (repo now at github.com/vargos-labs/vargos)
Self-hosted agent OS with persistent memory, multi-channel presence, tools, scheduling, and sub-agent orchestration
AI Security Review
scanned 14d ago · by lpm-firewall-aiThe bundled skill deliberately extracts and prints a Claude bearer token, creating a concrete credential exposure capability when explicitly invoked. Because inspection found no automatic invocation or remote exfiltration, this warrants a warning rather than a publish block.
Decision evidence
public snapshot- The package manifest defines a postinstall hook that rebuilds better-sqlite3; no evidence inspected ties it to the token-capture behavior.
- The token-capture skill is bundled and copied into the Vargos data directory when missing during explicit chat startup.
- When explicitly run, the script captures the Authorization header from Claude /v1/messages traffic and prints the bearer value.
- The agent source loads workspace prompt files; the inspected location does not show token forwarding or agent configuration mutation.
- The capture is initiated by running a specifically named skill script; inspection found no package install hook that invokes it.
- The script routes the Claude CLI request through a local proxy and outputs the captured value locally; inspected code showed no remote token recipient.
Source & flagged code
10 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgPackage source references dynamic require/import behavior.
dist/core/edges.jsView on unpkg · L37Findings
Affected versions and remediation
This report applies to @chozzz/vargos@3.2.20.
See all versions of @chozzz/vargos for other recorded verdicts.
- Review the evidence and your use of @chozzz/vargos@3.2.20 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.