The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around the blueprint CLI tool, which is used to build, test, and run scripts on the local machine.
[REMOTE_CODE_EXECUTION]: The blueprint run command executes arbitrary TypeScript scripts located in the scripts/ directory, which have access to a powerful NetworkProvider API.
[CREDENTIALS_UNSAFE]: The documentation guides users to store highly sensitive WALLET_MNEMONIC and WALLET_VERSION environment variables in a .env file for non-interactive deployments.
[EXTERNAL_DOWNLOADS]: The contract verification feature fetches configuration data from the ton-community organization's public repository on GitHub.
[DATA_EXFILTRATION]: The verify command transmits contract source code and metadata to the verifier.ton.org service as part of its intended verification workflow.
[PROMPT_INJECTION]: The toolchain exhibits an attack surface for indirect prompt injection as it processes external contract source code and blockchain API data using high-capability tools.
Ingestion points: Smart contract source files (.tolk, .fc, .tact), external API responses via NetworkProvider, and configuration files.
Boundary markers: None identified in the provided documentation snippets to prevent the model from obeying instructions embedded in source code or data.
Capability inventory: The skill facilitates full shell command execution, file system modification (e.g., blueprint pack updating project files), and network access for contract interactions.
Sanitization: No specific sanitization or validation protocols for processed external data are documented.

ton-blueprint