Flow Agents: An Intelligent Agent System That Infuses Structured Workflow Capabilities into Local Development Environments

Flow Agents is a local development workflow agent system developed by Kontour. It supports mainstream AI coding tools such as Codex, Claude Code, and Kiro, and provides a complete workflow closed loop from idea to backlog, plan to execution, review to verification, release preparation to experience recording, helping development teams establish a reproducible and auditable agent collaboration model.

With the popularization of AI coding assistants, development teams face the problem of missing workflow consistency across tools/sessions. As a workflow-aware agent bundle, Flow Agents can be integrated into existing toolchains, providing a unified workflow path for runtimes such as Codex, Claude Code, and Kiro. Its core concept is to establish workflow discipline within familiar tools without giving up existing tools.

Flow Agents provides modular skills covering the entire software development lifecycle:

• idea-to-backlog: Convert ideas into structured backlogs
• pull-work: Recommend priority tasks
• plan-work: Generate execution plans
• execute-plan: Execute code according to plan
• review-work: Code review
• verify-work: Verify change effects
• evidence-gate: Quality control point (requires sufficient evidence)
• release-readiness: Evaluate release readiness
• learning-review: Record lessons learned

Installation methods:

• Basic installation: npx @kontourai/flow-agents init (guided)
• Headless mode: npx @kontourai/flow-agents init --dest /path/to/workspace --telemetry-sink local-files --yes
• Specific runtime: npx @kontourai/flow-agents init --runtime codex --dest /path/to/workspace --activate-kits --yes After installation, copy resources such as agent definitions and skills; telemetry is written to local files by default.

Applicable scenarios:

1. Idea implementation: Convert ideas to structured Issues via the idea-to-backlog skill
2. Task execution: Select tasks via pull-work → plan-work → execute-plan
3. End-to-end delivery: Complete planning/execution/verification using the deliver command
4. Bug fixing: Reproduce, diagnose, fix, and verify issues using the fix-bug skill
5. Long-term projects: Track agent work status, support session compression, etc.

Repository Structure: Distinguish between source code (agents/, skills/, etc.), generated files (dist/, etc.), runtime files (.flow-agents/, etc.) Verification Tools:

• Repository hooks: npm run setup:repo-hooks
• Source code verification: npm run validate:source
• CI baseline check: bash evals/ci/run-baseline.sh --lane source-and-static etc. Ensure the bundle is compliant and workflow contracts are followed.

Flow Agents represents the shift of AI-assisted development from single-tool enhancement to workflow coordination, solving problems in AI agent process management, quality, and traceability. It is a methodology that treats AI as a collaborative partner and defines clear workflow boundaries. In the future, it may become an industry standard, promoting the transformation of AI-assisted development from "tools" to "partners".