Skip to main content
EVOKORE// SKILLS / CHANGELOG
Sign in
Workspace
Agent-33markdownsha 2B2A0D93CC59

CHANGELOG

release history

Changelog

All notable changes to AGENT-33 are recorded in this file. The format follows Keep a Changelog, and versions follow Semantic Versioning.

Unreleased

No changes yet.

2.1.0 — 2026-05-21

Initial public release of AGENT-33 under the Apache License 2.0.

Added

  • Engine (engine/) — FastAPI service with lifespan-wired subsystems and an Alembic-managed PostgreSQL schema.
  • Frontend (frontend/) — React/TypeScript operator console.
  • CLI (agent33) — installable via pip install -e ".[dev]". Commands include status, sessions, bench, packs, and config introspection.
  • Docker Compose stack — single-command spin-up of Postgres, Redis, NATS, Ollama, and the engine.
  • Kubernetes manifests (deploy/k8s/) — base configuration and a production overlay with HPA wiring.
  • Multi-tenant security model — JWT and API-key authentication, tenant isolation enforced through tenant_id propagation across persistence and service layers.
  • Agent subsystem (engine/src/agent33/agents/) — agent registry, JSON definitions auto-loaded from engine/agent-definitions/, capability taxonomy, agent runtime with iterative and streaming invocation.
  • Workflow engine (engine/src/agent33/workflows/) — DAG executor with retries, timeouts, parallel groups, conditional branches, expression evaluator, and checkpoint persistence.
  • Skill registry (engine/src/agent33/skills/) — SKILL.md and YAML loaders, progressive disclosure (L0/L1/L2), runtime injection into the agent prompt context.
  • Pack registry (engine/src/agent33/packs/) — local registry, optional PackHub client, signature verification through integrity hashing.
  • Tool framework (engine/src/agent33/tools/) — JSON Schema validated tool execution, governance allowlists, builtin tools for shell, file operations, HTTP fetch, and headless browser.
  • Messaging (engine/src/agent33/messaging/) — adapters for Telegram, Discord, Slack, and WhatsApp; per-channel health checks; NATS event bus.
  • Code execution (engine/src/agent33/execution/) — sandbox contracts, input validation, CLI adapter via subprocess, progressive disclosure of results.
  • Observability (engine/src/agent33/observability/) — structured logs, metrics, lineage, replay, alerting, and a trace pipeline with failure taxonomy and retention policies.
  • Evaluation (engine/src/agent33/evaluation/) — golden tasks and cases, metrics calculator, regression gates, evaluation service and API.
  • Autonomy budgets (engine/src/agent33/autonomy/) — budget lifecycle state machine, preflight checks, runtime enforcement for file, command, and network scopes.
  • Release automation (engine/src/agent33/release/) — release lifecycle states (planned → frozen → rc → validating → released → rolled_back), pre-release checklist, sync engine, rollback manager.
  • Continuous improvement (engine/src/agent33/improvement/) — research intake lifecycle, lessons learned tracking, improvement checklists, roadmap refresh records.
  • Candidate ingestion (engine/src/agent33/ingestion/) — governed candidate → validated → published → revoked lifecycle for external assets with confidence/trust labels.
  • Knowledge ingestion (engine/src/agent33/knowledge/) — RSS, GitHub, web, and folder adapters with APScheduler cron support.
  • MCP integration (engine/src/agent33/mcp_server/) — hosted MCP server surface plus client integration with circuit-breaker proxy semantics.
  • Memory (engine/src/agent33/memory/) — short-term buffer, pgvector long-term store, hybrid RAG pipeline (BM25 + vector via RRF), token-aware chunking, session state.
  • LLM router (engine/src/agent33/llm/) — provider abstraction with Ollama and OpenAI-compatible adapters plus a model router and a provider catalog auto-registered from environment variables.
  • SkillsBench integration (engine/src/agent33/benchmarks/skillsbench/) — runner, adapter, task loader, CTRF reporting, smoke and full tiers.

Documentation

  • README, QUICKSTART, INSTALL, ARCHITECTURE at the repo root.
  • Operator manual, configuration reference, CLI reference, API reference, troubleshooting guide, upgrade guide, use cases, and walkthroughs in docs/.
  • Architecture deep-dive with mermaid diagrams in docs/architecture/.
  • Operator runbooks in docs/operators/ and docs/runbooks/.
  • Contribution standard, concepts, glossary, testing, releasing, and examples documents.

Known limitations

  • Single-instance default Kubernetes manifest. Multi-replica deployment is supported by the production overlay but requires operator review of state-boundary assumptions in engine/src/agent33/lifespan/.
  • Pack signing uses SHA-256 integrity. Asymmetric signing (Sigstore cosign) is on the near-term roadmap.
  • Some authenticated routes are not yet scoped per-tenant; see docs/architecture/api-surface.md for the current scope matrix.