Overview
ForgeFlow Agent Platform is an open-source engineering project exploring how AI agents can participate in software development through structured roles, bounded tools and verifiable evidence.
Rather than relying on a single general-purpose agent to plan, implement, test and review a change, ForgeFlow separates software engineering responsibilities into specialised roles covering areas such as architecture, implementation, testing, security and final verification. The long-term objective is to coordinate these roles within a traceable workflow in which findings can be reviewed, challenged and validated before they affect a repository.
The platform combines an agent orchestration layer with Model Context Protocol tools and a deterministic repository analysis subsystem. Its current development milestone includes a command-line repository analyser that performs bounded, read-only inspection without executing or modifying the target repository.
The analyser discovers programming languages, dependency manifests, test directories, continuous-integration definitions, container files and Kubernetes resources. It produces normalised security and engineering findings with rule identifiers, severity levels, confidence scores, evidence and recommended remediation.
Current deterministic checks cover risks such as hard-coded credentials, unsafe shell execution, missing test or CI structures, mutable container image tags and containers running without a non-root user. Results are exported as human-readable reviews and machine-readable JSON reports, allowing them to be integrated into automated engineering workflows.
ForgeFlow also includes foundations for MCP-based tool access, policy enforcement, shared memory, telemetry, Java analysis services, containerised deployment and Kubernetes infrastructure. Specialist-agent analysis is being introduced incrementally after validating the deterministic and safety-oriented foundation.