The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external Solidity source code and blockchain data from RPC providers, which creates an inherent surface for indirect prompt injection should malicious instructions be embedded in contract comments or metadata.
Ingestion points: Solidity source files and RPC data (via SKILL.md and references/foundry-cheatsheet.md).
Boundary markers: None explicitly mentioned for separating untrusted code from instructions.
Capability inventory: Execution of forge, cast, and anvil commands, and various MCP tools for contract compilation, testing, and deployment.
Sanitization: Promotes the use of static analysis tools like Slither and Aderyn to detect vulnerabilities before deployment.
[EXTERNAL_DOWNLOADS]: Mentions the use of forge install for managing project dependencies from external GitHub repositories.
Evidence: forge install <author>/<repo> referenced in SKILL.md as a standard dependency management step.
[CREDENTIALS_UNSAFE]: Demonstrates best practices for secret management by using environment variables rather than hardcoded credentials.
Evidence: Use of vm.envUint("PRIVATE_KEY") and vm.envString("API_KEY") in references/foundry-cheatsheet.md to securely access environment-provided secrets.

solidity-foundry-development