The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing core developer tools via shell-piped remote scripts from vendor-associated domains.
Instructions for installing the Solana CLI using sh -c "$(curl -sSfL https://release.anza.xyz/stable/install)" in references/common-errors.md.
Instructions for installing Surfpool via curl -sL https://run.surfpool.run/ | bash in references/surfpool/overview.md and references/testing.md.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface due to its interaction with untrusted on-chain data and program logs.
Ingestion points: Ingests external account data and program logs via RPC lookups in references/rpc-quick-lookups.md and @solana/kit account fetching in references/kit/accounts.md.
Boundary markers: The skill contains a dedicated 'Untrusted data handling' section in SKILL.md which instructs the agent to use delimiters and ignore embedded directives.
Capability inventory: Access to curl, and various CLI tools (anchor, solana, surfpool) which can be used to execute commands or write files.
Sanitization: The skill provides explicit instructions to validate RPC responses and program logs before interpolation.
[COMMAND_EXECUTION]: The skill uses curl for one-shot RPC lookups and drives various CLI tools (anchor, surfpool, solana) to perform development tasks.
[DYNAMIC_EXECUTION]: The skill documents the use of unsafe Rust blocks for zero-copy memory access when building high-performance programs in references/programs/pinocchio.md.

solana-dev