The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions in ENVIRONMENT.md to fetch the Solana CLI installer from Anza's official release domain (release.anza.xyz) and platform-tools from the official Anza repository on GitHub (github.com/anza-xyz). These are well-known and trusted sources for Solana development tooling.- [REMOTE_CODE_EXECUTION]: Instructions in ENVIRONMENT.md include executing the Solana installation shell script after downloading it from the official Anza release site. This is a standard procedure for setting up the Solana development environment.- [INDIRECT_PROMPT_INJECTION]: The guidelines in SUMMARIZING-PROGRAMS.md instruct the agent on how to summarize and walk through Solana program code. This involves the agent processing untrusted data (source code).
Ingestion points: The agent reads user-provided project files (Rust source, TypeScript tests, and configuration files like Anchor.toml).
Boundary markers: The skill emphasizes verification through truthfulness rules and running actual tests, though it does not explicitly define escape-sequence boundaries for the analysis output.
Capability inventory: The agent has the capability to execute shell commands (e.g., sh, cargo, solana-keygen) and modify project files as part of its development assistance role.
Sanitization: No specific sanitization of untrusted code comments or identifiers is described; however, the skill mandates strict verification against the source code via grep and test execution, reducing the risk of being misled by malicious content.

solana