The Agent Skills Directory

[DATA_EXFILTRATION]: The deployment script packages local directory contents for upload to an external service without sufficient filtering.
Evidence: The scripts/deploy.sh script uses tar to create a package: tar -czf "$TARBALL" -C "$PROJECT_PATH" --exclude='node_modules' --exclude='.git' ..
Risk: This exclusion list fails to omit .env files, SSH keys, or other sensitive configuration files typically stored in project roots, potentially exposing credentials to the remote endpoint.
[PROMPT_INJECTION]: The skill processes untrusted local data, creating a surface for indirect prompt injection.
Ingestion points: The scripts/deploy.sh script reads the project directory and package.json content in detect_framework.
Boundary markers: No markers are present to distinguish file content from agent instructions.
Capability inventory: The skill possesses file read (tar) and network upload (curl) capabilities.
Sanitization: No filtering or sanitization is applied to the project files prior to ingestion and packaging.
[COMMAND_EXECUTION]: The skill uses a shell script to perform its primary functions.
Evidence: The SKILL.md frontmatter allows the Bash tool to run scripts/deploy.sh, which executes shell commands like tar, find, and curl to manage files and network requests.

deploy-on-vercel