List view

• V6 — Release 1.0

  API contract frozen. Semantic versioning. Documentation site. Stable Docker images published. Reference examples maintained.

  No due date

  •0/5 issues closed
  0% complete5 open0 closed

• V5 — MCP server adapter

  Expose the engine itself as an MCP server. Other clients (any MCP-compatible host) can use the engine as a tool provider for orchestration.

  No due date

  •0/3 issues closed
  0% complete3 open0 closed

• V4 — Multi-tenant

  Sandboxes are isolated per tenant_id. Quotas enforced per tenant. Auth supports multiple users.

  No due date

  •0/3 issues closed
  0% complete3 open0 closed

• V3 — Python SDK on PyPI

  Publish a Python SDK that wraps the WebSocket + REST API. Goal : a Python codebase can use the engine in a few lines.

  No due date

  •0/4 issues closed
  0% complete4 open0 closed

• V2 — Auth + state persistence

  Sessions persist across daemon restarts. Multiple parallel sessions are supported. Auth tokens are managed (rotation, revocation).

  No due date

  •0/3 issues closed
  0% complete3 open0 closed

• V1 — WebSocket API server

  Expose the conversational builder as a daemon. Clients connect via WebSocket, send prompts, receive a stream of events (assistant messages, tool uses, tool results, agent events, completion).

  No due date

  •0/4 issues closed
  0% complete4 open0 closed

• P9 — POC validated : Next.js + Laravel + QA demo

  End-to-end demo. The reference team scenario works : describe a Next.js + Laravel app, the builder designs the team, three agents collaborate in a sandbox, produce code in ./workspace/, run integration tests, write a QA report.

  No due date

  •0/6 issues closed
  0% complete6 open0 closed

• P8 — Pixel-art TUI dashboard

  The mockup screens become real : mission control with live agent states, focus drill-down, hangar gallery. Each agent shows up as a pixel-art sprite that animates with its state (idle/thinking/working/tool_use/done/error).

  No due date

  •0/7 issues closed
  0% complete7 open0 closed

• P7 — TEAM.md (multi-agent coordination)

  Multiple agents running in a single Docker container, communicating via the claude-presence MCP (broadcast + advisory locks). Working dirs are disjoint per agent.

  No due date

  •0/6 issues closed
  0% complete6 open0 closed

• P6 — Builder skills enriched

  The builder LLM gains its core skills : recognize-stack, configure-sandbox, generate-agent-md, design-agent-prompt, select-mcp-servers. Each skill is loaded as a SKILL.md from packages/core/src/builder/skills/.

  No due date

  •0/6 issues closed
  0% complete6 open0 closed

• P5 — Hardened sandbox + artifact extraction

  Production-grade Docker isolation : non-root user, read-only root FS, --no-new-privileges, allowlist network, resource limits. Files produced by the agent are extracted to ./workspace/ on host before container removal.

  No due date

  •0/6 issues closed
  0% complete6 open0 closed

• P4 — Native tools

  Implements the 6 native tools every agent can use : Bash, FileRead, FileEdit, FileWrite, Grep, Glob.

  No due date

  •0/8 issues closed
  0% complete8 open0 closed

• P3 — Builder launches the agent it designed

  Full loop : user describes an agent → builder generates AGENT.md → builder launches a container with that agent → result is shown back to the user.

  No due date

  •2/6 issues closed
  33% complete4 open2 closed

• P2 — Conversational CLI (minimal)

  REPL with Ink that hosts the builder LLM. The user types a prompt, the builder responds, the conversation history is maintained for the session.

  No due date

  •3/6 issues closed
  50% complete3 open3 closed

• P1 — Hello agent in Docker

  First runnable milestone. A single agent runs inside a Docker container, responds to a prompt via stdio, container is destroyed cleanly. Validates the full host↔container plumbing.

  No due date

  •6/6 issues closed
  100% complete0 open6 closed