The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides comprehensive instructions for the agent to execute shell commands using the glam-cli tool. These commands perform high-impact financial operations on the Solana blockchain, such as creating vaults, managing assets, and executing DeFi strategies.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external Node.js packages, specifically @glamsystems/glam-cli and @glamsystems/glam-sdk, alongside peer dependencies such as @solana/web3.js and @coral-xyz/anchor. These are verified as vendor-specific resources belonging to the skill author.
[DATA_EXFILTRATION]: The skill references and manages sensitive local file paths required for Solana blockchain interaction. It provides instructions for configuring ~/.config/glam/config.json and utilizes the standard Solana keypair path at ~/.config/solana/id.json for transaction signing. While these are standard practices for blockchain CLI tools, they involve access to sensitive cryptographic material.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to its interaction with untrusted external data:
Ingestion points: The skill fetches real-time data from Solana RPC nodes (e.g., api.mainnet-beta.solana.com) and token pricing/route data from the Jupiter API.
Boundary markers: Analysis of the skill instructions and templates reveals an absence of explicit boundary markers or instructions to disregard embedded commands in the processed data.
Capability inventory: The skill possesses significant capabilities, including fund transfers (glam-cli transfer), token swaps (glam-cli jupiter swap), and the ability to modify vault permissions and delegates.
Sanitization: No explicit sanitization or validation logic is defined for the data retrieved from external blockchain or API sources before it is incorporated into the agent's context.

glam