Skip to content

Changelog

Release history for the Escher platform.

Release notes are generated automatically across all repos using v2-release-notes (Claude 4.5 Sonnet) and published to #escher-releases on each production deploy.


v4.x — Agent Development Kit (Current)

v4 introduces the Agent Development Kit (ADK), the Context Engine v2, and the formula-based execution model. Domain agents are registered as agent.yaml packages via adk register rather than deployed as standalone services. The v4 Context Engine is the authoritative store for all skills, tools, guardrails, templates, domain lenses, and expert graphs.

v4.1.0 — ADK stable release

Context Engine

  • Added /resolve/agent endpoint for Code Agent grounding at Phase 2C/2W miss
  • Agent Registry collection replaces deprecated Intent collection
  • Context Builder collection deprecated — context fields moved inline to Skill execution_plan
  • Cloud Knowledge collection merged into Domain Lens (content_type: "cloud_service")

ADK CLI

  • adk scaffold <name> --domain <domain> — generates agent.yaml template with all required fields
  • adk validate — validates schema, checks domain registry, verifies skill + tool refs
  • adk register — atomic registration to all CE collections in dependency order
  • adk status — shows registration state per collection
  • adk list agents / adk list skills — lists registered content

Asset Store

  • Go rewrite of Asset Store (replaces Python prototype)
  • Schema migration command: asset-store-migrate
  • SurrealDB backend (local or managed)

Integrations Agent

  • New v4-integrations-agent-go service — Slack/Jira/Linear message classification
  • claude-haiku-4-5 classifier, SSE passthrough to code-gen agent

v4.0.0 — Context Engine v2

Platform

  • Formula model formalised: Agent(x,d,t,e) = Skill(x) + ContextBuilder(x) + Estate(e) + Tool + Guardrail + Template + DomainLens + ExpertGraph + Artifact + Evidence
  • 13 CE collections defined (v4_arch_schema_json spec)
  • Domain Expert Graph added — compliance control → requirement → evidence → remediation chains
  • Tag Store added — request-scoped internal state, TTL auto-delete

Context Engine

  • Intent-based REST endpoints — currently implemented: /health, /tags/{write,read,delete}, /register/{skill,agent,tool,context_builder}, /resolve/{skill,context_builders,agent,tools,tool_class}, /list/{skills,tools}. Stubbed (501): /search/playbooks, /answer/knowledge, /resolve/domain_knowledge. Additional resolution endpoints (/resolve/template, /resolve/guardrail, /resolve/playbook, /answer/rag) are part of the v4 design and tracked for implementation.
  • Tenant → global collection scoping
  • Per-collection embedding strategies documented

v3.x — Agent Refactor

v3 migrated backend agents from Java to Go, introduced v3-escher-deployment as the central CI/CD pipeline, and added the v3 testing framework with full E2E suite coverage.

v3.2.0 — CI/CD centralisation

  • v3-escher-deployment pipeline owns all 63 repos
  • GitHub Actions OIDC — no long-lived AWS credentials
  • Parallel matrix builds across all component services
  • Tag-based promotion: v*-devv*-stgv*
  • Slack deployment notifications to #escher-deployments
  • E2E integration tests triggered automatically after deploy

v3.1.0 — Testing framework

  • v3-testing-framework introduced — Python-based E2E suite
  • Smoke, regression, and integration test tiers
  • Test results posted to #escher-test-results Slack channel
  • v2-release-notes tool added — Claude-powered multi-repo release note generation

v3.0.0 — Go agent migration

  • Analysis Agent rewritten in Go (was Java/Quarkus)
  • Playbook Agent rewritten in Go (was Java)
  • Java retained for Gateway and Autoresolver (Quarkus framework)
  • Context Engine Python prototype introduced (later replaced by v4 CE)
  • v3-architecture-docs published — full object model, noun model, UX model specs

v2.x — Stable Production

v2 is the current stable production release shipped to Design Partners. It includes the full Tauri desktop app, all Rust plugins, Java/Go backend agents, and the E2E test suite.

v2.15.0

Desktop App

  • Estate scan now shows incremental progress per resource type
  • Profile switcher redesigned — instant switch without re-auth
  • Finding detail panel added — click any finding to see full evidence chain

Analysis Agent

  • Cost anomaly detection improved — now uses 90-day rolling baseline
  • SSE stream stability fixes for long-running analyses
  • compute_stats tool added — pre-computes aggregates before LLM reasoning

Context Engine (v2)

  • Skill collection embedding updated — purpose + description + capability + context vectors
  • Domain Lens RAG chunking strategy changed to 512-token overlapping windows
  • Template resolution fallback chain: skill → agent → domain → global

v2.14.0

Desktop App

  • Bundle review screen introduced — approve/reject individual steps before execution
  • Evidence report export: PDF and JSON formats
  • Multi-cloud profile support — query across AWS + Azure simultaneously

Playbook Agent

  • Rollback support for all safety_class: supervised playbooks
  • Dry-run mode added (dry_run: true parameter)
  • Step-level evidence capture — before/after state recorded per write step

Gateway

  • Cognito OIDC integration — ALB listener rule replaces custom JWT validation
  • Rate limiting added per tenant: 100 req/min (configurable)
  • Request ID propagated through all downstream service calls

v2.13.0

Desktop App

  • Initial Tauri app GA release
  • AWS and Azure cloud connection setup
  • FinOps, SecOps, Compliance, IAM skill verticals
  • Plan generation and one-click playbook execution
  • v2-tauri-plugin-vault — OS Keychain credential storage
  • v2-tauri-plugin-scanner — local resource scan engine
  • v2-tauri-plugin-estate — local estate RAG (ONNX embeddings, SurrealDB)

Backend

  • ECS Fargate deployment on AWS us-west-1
  • 6 services: Gateway, Analysis Agent, Playbook Agent, Context Engine, Asset Store, Autoresolver
  • v2-deployment-ecs scripts for ECR push and ECS service update

Release note tooling

Release notes are generated using v2-release-notes:

bash
cd v2-release-notes
python generate_release_notes.py \
  --repos gateway,analysis-agent,context-engine,asset-store \
  --from-tag v2.14.0 \
  --to-tag v2.15.0 \
  --publish-slack \
  --channel "#escher-releases"

The tool aggregates commits across repos, summarises them with Claude 4.5 Sonnet, and formats the output as structured Slack Block Kit messages.

Escher — Agentic CloudOps by Tessell