Invisible AI watchdog for every flow:
- Read-only access to context buffer
- Cannot emit messages to pipeline
- Agents have no way to detect or probe it
- Alerts via control plane (email, UI, auto-stop)
- Runs on cheap models (Mistral/Mixtral)
Watches for: endless loops, goal drift, prompt injection,
sandbox escape attempts, token budget exhaustion.
Added to Phase 2 (core safety feature).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Rename all nextra-* files to bloxserver-*
- Replace all "Nextra" references with "BloxServer"
- Update copyright year to 2026
- Add domain: OpenBlox.ai
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Monaco editor with AI assistance for writing AssemblyScript:
- Inline completion (like Copilot)
- Chat panel (like Claude Code)
- Frontend calls LLM API directly
- Human reviews before building
Ships with Phase 3 (Monaco integration).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The assistant is itself a Nextra flow (dogfooding):
- Builder agent with catalog, validator, examples tools
- Queries real available nodes dynamically
- Self-validates generated YAML before returning
- Uses marketplace flows as few-shot examples
- Same billing model (LLM tokens)
Added Phase 4.5 to implementation roadmap.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Simple two-state model: Stopped ↔ Running
- Edit only allowed when stopped
- No pause (simpler, matches Zapier/n8n/Make)
- No hot-edit (unsafe for mid-execution swarms)
- Future consideration: Graceful Stop for Pro users
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Monaco's built-in TS language service provides IDE features
- AS type definitions loaded at startup for autocomplete
- Real errors come from `asc` compiler at build time
- No separate LSP server (asls) needed = zero infra cost
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Defines shared API contract between frontend and backend:
- types.ts: TypeScript interfaces for Next.js frontend
- models.py: Matching Pydantic models for FastAPI backend
Covers: User, Flow, Trigger, Execution, WasmModule, Marketplace,
ProjectMemory, and pagination/error types.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Detailed prompt for generating the SaaS landing page with Vercel v0.
Includes: hero, features, pricing, testimonials, FAQ, and styling specs.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Move lsp-integration.md and secure-console-v3.md to docs/archive-obsolete/
(these features are now in the Nextra SaaS product)
- Update CLAUDE.md with current project state
- Simplify run_organism.py
- Fix test fixtures for shared backend compatibility
- Minor handler and llm_api cleanups
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
OSS restructuring for open-core model:
- Rename package from agentserver/ to xml_pipeline/
- Update all imports (44 Python files, 31 docs/configs)
- Update pyproject.toml for OSS distribution (v0.3.0)
- Move prompt_toolkit from core to optional [console] extra
- Remove auth/server/lsp from core optional deps (-> Nextra)
New console example in examples/console/:
- Self-contained demo with handlers and config
- Uses prompt_toolkit (optional, falls back to input())
- No password auth, no TUI, no LSP — just the basics
- Shows how to use xml-pipeline as a library
Import changes:
- from agentserver.* -> from xml_pipeline.*
- CLI entry points updated: xml_pipeline.cli:main
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Lua proposal covers:
- Sandboxed Lua handlers for SaaS custom logic
- XML <-> Lua table bridging with XSD validation
- Security sandbox (restricted stdlib, resource limits)
- Schema DSL alternative to raw XSD
- Hot reload mechanism
API spec updated with:
- Dedicated /ws/messages endpoint for message log streaming
- Filter by agent/thread/payload type
- Example for reconstructing agent buffers client-side
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Covers authentication (TOTP, WebAuthn, mTLS, OAuth), authorization
(RBAC, resource permissions), transport security, secrets management,
audit logging, prompt injection defense, and rate limiting.
Includes open questions and implementation checklist.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Defines the GUI-agnostic API for monitoring and controlling the organism:
- REST endpoints for queries and control
- WebSocket for real-time push events
- Data models for agents, messages, threads
- Example usage for 3D office GUI
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Documents the trust model for context and prompt management:
- Platform as trusted orchestration layer
- Context buffer: pure message history, keyed by UUID, auto-GC
- Prompt registry: immutable, invisible to agents
- LLM call assembly: platform controls, agents request
- Handler sandbox: what agents can/cannot do
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When detached, console changes prompt from '>' to '#' and rejects
@messages while still accepting /commands. Simpler than the ring
buffer approach - operator can still check status and re-attach.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Minimal, universal base for all capabilities
- Required listens_to class attribute
- Async handle() with convo_id preservation rule
- make_response() helper for thread continuity
- Explicit name, config, per-instance logger
- Clear documentation and NotImplementedError guard
This is the true substrate the entire organism will build on.
