AI·2025 — Present
FlawCode (ForgeOS)
Autonomous Multi-Agent AI Software Engineering Operating System with Deterministic State Machine Orchestration
Multi-Agent Telemetry Capture PendingVerified Specification
Multi-Agent AI Engineering Architecture
Hermes orchestrator state transition logs and live Slack channel execution streams will be supplied in the next phase.
Agents
5 Roles
Sandbox
Docker AST
Tool Schema
Pydantic
The Problem
Autonomous AI coding agents frequently hallucinate syntax, overwrite critical codebases, and run destructive terminal commands when given direct system access. Without rigid state machine boundaries and containerized sandboxes, multi-turn AI coding produces broken repositories and irrecoverable bugs.
Engineering Highlights
- Hermes Orchestrator & OpenClaw DAG: Supervised development through deterministic state transitions, breaking user requests into atomic dependency graphs (DAG) dispatched to specialized sub-agents.
- Sandboxed Docker Execution with AST Checks: Ran all code generation, unit tests, and shell tools inside isolated throwaway Docker containers with strict file permission limits and AST validation prior to git commit.
- Pydantic Schema Tool Validation: Built a strict tool execution harness enforcing JSON schema validation across all agent calls, eliminating parameter hallucinations.
- Real-Time Slack Streaming & GitHub PRs: Streamed execution diffs and terminal output directly into assigned Slack threads, automatically raising GitHub pull requests once test assertion matrices passed.
Results
FlawCode delivered deterministic multi-turn software engineering workflows across active repositories, executing feature implementation, test verification, and automated PR generation safely within container boundaries.