Agent execution
Agent execution concepts mapped to actual Forge source packages.
Agent loops alternate between model generation and tool execution until the model finishes or a stop condition fires. Supply a concrete model, a tool registry and an explicit approval policy. Set finite budgets and handle failure before recording completion. Constructors, initialization and output fields differ between languages.
Implementation references
The references below are the Rust implementation. They include manifest versions, features, source hashes, and declarations.
Other languages
Use the language-specific package catalog for TypeScript, Python, Go, Swift and Kotlin. Similar responsibilities do not imply the same types, defaults, targets or deployed capabilities.
Validation
Read the exact package boundary and selected feature conditions. Run the relevant library tests and the applicable conformance fixtures before making a release claim. Source documentation and generated inventories do not replace runtime verification.