The Agent Skills Directory

EXTERNAL_DOWNLOADS (HIGH): The skill's GitHub Action templates in assets/workflows/ (build.yml, deploy.yml, and test.yml) use a piped remote execution pattern (curl | sh) to install the Dfinity SDK from internetcomputer.org. Since this domain is not included in the pre-approved trusted source list, this constitutes a high-risk external dependency.
REMOTE_CODE_EXECUTION (HIGH): Piped remote execution is a critical vulnerability surface. Furthermore, an automated scanner (URLite) detected a blacklisted URL within the main.rs structure that this skill manages and deploys.
COMMAND_EXECUTION (MEDIUM): Multiple Python utility scripts (e.g., deploy.py, test.py) execute system commands using the subprocess module. These commands use project-specific parameters, which could lead to shell injection if canister or project names are maliciously crafted.
DYNAMIC_EXECUTION (HIGH): The skill generates Motoko, Rust, and TypeScript code by interpolating user-provided strings into templates without proper sanitization. This allows for arbitrary code injection into the generated canisters during the scaffolding process.
INDIRECT_PROMPT_INJECTION (LOW): \n
Ingestion points: User-controlled project names, canister names, and symbols passed to add-canister.py and add-token.py.\n
Boundary markers: None are present in the generated code templates to prevent the LLM or build system from executing instructions embedded in these strings.\n
Capability inventory: The skill possesses filesystem write access and the ability to execute shell commands via the dfx tool.\n
Sanitization: Input strings undergo minimal transformation (such as .title()), which is insufficient to sanitize against malicious code or script injection into the generated project files.

icp-appgen