vercel-deploy
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION] (HIGH): The script
scripts/deploy.sharchives the project directory and uploads the resulting tarball to an external endpoint (https://codex-deploy-skills.vercel.sh/api/deploy). Evidence inscripts/deploy.sh:tar -czf "$TARBALL" -C "$PROJECT_PATH" --exclude='node_modules' --exclude='.git' .andcurl -s -X POST "$DEPLOY_ENDPOINT" -F "file=@$TARBALL". The exclusion list is insufficient as it does not prevent the inclusion of sensitive files like.env,.env.local, SSH keys, or cloud provider credentials often found in project roots. - [COMMAND_EXECUTION] (LOW): The skill relies on the execution of a local bash script to perform file system operations and network requests as part of its primary deployment function.
- [INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted local project data. Evidence chain: (1) Ingestion points:
scripts/deploy.shpackages local directory contents. (2) Boundary markers: Absent. (3) Capability inventory: Subprocess execution and network upload viacurl. (4) Sanitization: Absent. External instructions in processed files could influence agent behavior if displayed to the user.
Audit Metadata