The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from a remote sandbox using tools like read_file, grep, and list_directory. It possesses capabilities to write files and execute shell commands (run_command), which could be exploited if malicious instructions are embedded in the processed code. The instructions do not include boundary markers or sanitization requirements for these inputs.
Ingestion points: read_file, grep, list_directory (found in SKILL.md)
Boundary markers: Absent
Capability inventory: write_file, run_command, and network operations via OAuth connectors (found in SKILL.md)
Sanitization: Absent
[DYNAMIC_EXECUTION]: The skill involves generating and writing executable TypeScript and JavaScript files to a remote sandbox for automated deployment. It also instructs the agent to perform verification by running these scripts through the run_command tool.
[COMMAND_EXECUTION]: The skill utilizes the run_command tool to execute shell-based tasks such as npm run build and npx tsc --noEmit within the sandbox environment to verify code changes.

base44-sandbox