[RFC] ROADMAP: 2026 5/6 Development Plan

[MoveCloudROY]

ROADMAP: 2026 5/6 Development Plan

This issue tracks the major feature development plan for ecs-agent. Items are organized by target release version and priority.

---

v0.1.1 — Built-in State Machine for Agent Gatekeeping

Status: Planned | Priority: High

Introduce a declarative state machine system for controlling agent lifecycle and access.

• State Machine Core

  • StateMachineComponent with transition tables
  • Event-driven state transitions (tool calls, user input, errors)
  • Hierarchical states (nested state machines)

• Agent Gatekeeping

  • Approval workflows for sensitive tools (human-in-the-loop)
  • Quota management (rate limits, budget caps)
  • Permission escalation paths

• Observability Integration

  • State transition logging
  • Audit trails for security-critical transitions
  • Metrics export (Prometheus-compatible)

Motivation: Production deployments need explicit control over agent behavior, especially for autonomous or long-running agents.

---

v0.1.2 — Agent Script Syntax Sugar

Status: Planned | Priority: High

Simplify agent definition and orchestration with declarative syntax sugar.

• Agent DSL Enhancements

  • Compact YAML/JSON syntax for agent configuration
  • Inline tool registration with decorators
  • Conversation flow templates (branching, loops)

• Script-Based Agent Definitions

  • @agent decorator for Python functions
  • Automatic component injection from type hints
  • Async generator support for streaming agents

• Composition Macros

  • Pre-built agent templates (ReAct, Plan-and-Execute, Multi-Agent)
  • Skill composition operators (+, |, >>)
  • Hot-reload for development workflows

Motivation: Lower the barrier to entry for new users while maintaining the full power of ECS composition.

---

v0.1.3 — Agent Capability Benchmark

Status: Planned | Priority: Medium

Standardized evaluation framework for measuring agent capabilities.

• Benchmark Suite

  • Tool use accuracy (function calling, parameter extraction)
  • Multi-turn reasoning (planning, error recovery)
  • Context management (compaction, memory windowing)
  • Subagent orchestration (delegation, result aggregation)

• Benchmark Infrastructure

  • Dataset format specification
  • Automated runner with result persistence
  • Regression detection across model versions

• Leaderboard & Reporting

  • HTML report generation
  • Model comparison matrix
  • Cost-performance tradeoff analysis

Motivation: Enable data-driven model selection and track framework improvements over time.

---

v0.1.4 — Code CLI Support

Status: Research | Priority: Medium

Integration with code-native CLI tools for agent-driven development.

• CLI Adapter Layer

  • Abstract interface for code CLI providers
  • Session management (context preservation across commands)

• Kimi Code Integration (Tentative)

  • Adapter for kimi code CLI
  • File editing via structured diff protocol
  • Symbol search and navigation

• Generic Code Tooling

  • LSP client integration for semantic operations
  • AST-based code manipulation
  • Git operation wrappers (branch, commit, diff)

Motivation: Enable agents to perform real software engineering tasks with the same tools developers use.

---

Release Cadence

Version	Target Date	Focus
v0.1.1	May 2026	Production readiness
v0.1.2	May 2026	Developer experience
v0.2.0	June 2026	Evaluation & trust
v0.2.1	June 2026	Code-native agents

Contributing

Interested in a specific feature? Comment below with:

1. Use case description
2. Proposed API/design
3. Willingness to contribute implementation

We prioritize items with community demand and clear use cases.