tracist@0.0.57
Upload and continuously stream local AI coding transcripts to Tracist
AI Security Review
scanned 3mo ago · by lpm-firewall-aiThe source establishes a real remote AI-agent control capability with no per-prompt local confirmation, but it is activated through an explicit user command rather than an npm lifecycle hook. This warrants a warning rather than a publication block under the stated policy.
Decision evidence
public snapshot- `scripts/capture-resume-multiturn.js` accepts server queue messages over an authenticated WebSocket.
- Queued messages are forwarded as prompts to local agent harnesses without a local approval step.
- `WarmCodexHarness` starts/resumes Codex with `sandbox: "danger-full-access"` and `approvalPolicy: "never"`.
- `bin/tracist.js install` explicitly creates a persistent user daemon that runs this streamer.
- Remote filesystem list/search/read requests are serviced for known local transcript sessions.
- No preinstall/install/postinstall hook; `prepare` only invokes esbuild.
- The CLI and README disclose transcript uploading and an explicit `tracist install` daemon command.
- Uploads use the package-aligned `tracist.dev` origin and bearer authentication.
- `tools/secret-redaction.js` redacts common credential patterns before transcript processing.
Source & flagged code
7 flagged · loading sourceA manifest entrypoint or package-local install chain reaches persistence behavior.
bin/tracist.jsView on unpkg · L7Findings
Affected versions and remediation
This report applies to tracist@0.0.57.
See all versions of tracist for other recorded verdicts.
- Review the evidence and your use of tracist@0.0.57 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.