proffesor-for-testing__agentic-qe/docs/releases/v3.7.1.md

v3.7.1 Release Notes

Release Date: 2026-02-25

Highlights

Full OpenCode integration makes AQE's complete toolkit — 60 agents and 78 QE skills — accessible from any OpenCode-compatible agent client. Instead of configuring agents and skills manually, run aqe init --auto and OpenCode clients automatically discover all AQE capabilities through standard YAML configs.

What is OpenCode Integration?

OpenCode is an open standard for AI agent clients. With this release, AQE generates standardized configuration files that any OpenCode-compatible client can consume:

• Agent configs (.opencode/agents/*.yaml) — Each of AQE's 59 agents gets a rich YAML profile including its identity, capabilities, memory integration instructions, learning protocol, and available MCP tools. OpenCode clients can browse and invoke any agent without custom setup.
• Skill configs (.opencode/skills/*.yaml) — All 86 skills (78 QE + 8 general dev) are exposed with step-by-step workflows, required tools, and domain tags. Clients can discover and execute skills by category.
• Tool wrappers (.opencode/tools/*.yaml) — 5 core MCP tool wrappers for fleet management, test generation, coverage analysis, quality assessment, and memory operations.
• Transport routing — SSE, WebSocket, and HTTP transports with automatic selection based on client capabilities.
• Graceful degradation — Skills are tagged with minimum model tiers (tier1-any, tier2-good, tier3-best). When running on less capable models, the system warns or suggests simpler alternatives.
• Output compaction — Large agent outputs are automatically compressed to fit within OpenCode's context window limits.

Quick Start

# New project — auto-detects OpenCode and provisions everything
npx agentic-qe init --auto

# Existing project — add OpenCode assets
npx agentic-qe init --auto --with-opencode

This creates .opencode/agents/, .opencode/skills/, .opencode/tools/, and opencode.json in your project.

Added

• 59 OpenCode agent configs — Rich YAML profiles generated from the full agent definitions in .claude/agents/v3/. Each config includes identity/mission, core capabilities, operating principles, memory integration (namespaces and query/store instructions), learning protocol, output format, and coordination notes. Covers all QE domains: test generation, coverage analysis, security compliance, defect intelligence, chaos engineering, and more.

• 86 OpenCode skill configs — All QE and general dev skills exposed as step-by-step workflows. Each config includes the skill description, domain tags, required MCP tools, and actionable steps extracted from the skill definitions.

• 5 MCP tool wrappers — Fleet management, test generation, coverage analysis, quality assessment, and memory operations available as OpenCode tool definitions.

• SSE/WebSocket/HTTP transport routing — Automatic transport selection with fallback chain for maximum client compatibility.

• Output compaction middleware — Compresses large agent outputs to stay within context window limits while preserving essential information.

• Provider capability matrix — Documents which skills work with which model providers (Claude-only, Claude/GPT, any provider, local-safe).

• Graceful degradation — 3-tier model routing for skills ensures appropriate warnings when running on less capable models.

• aqe init --with-opencode flag — Provisions all OpenCode assets into user projects during initialization. Auto-detected in --auto mode when opencode.json exists.

• RVF production wiring — All QEReasoningBank creation sites now inject the shared RvfDualWriter singleton for best-effort native vector replication with AQE_RVF_MODE=sqlite-only operator override.

Fixed

• YAML string escaper now properly escapes backslashes (CodeQL finding)
• O(n^2) similarity edge explosion in dream engine spreading activation
• Non-semantic metrics JSON skipped from embedding pipeline
• Persistence reliability improvements in learning pipeline
• RVF shutdown ordering and reset race condition
• Stale skill/agent counts corrected across all documentation (75 → 78 QE skills, consistent 60 agent count)

Changed

• Agent generator (generate-opencode-agents.ts) rewritten to read from .claude/agents/v3/*.md with full XML section extraction
• Skills generator (generate-opencode-skills.ts) updated with CRLF normalization, collision-safe naming, and richer step extraction
• All @ruvector/rvf-node platform variants externalized in build scripts

OpenCode Asset Counts

Asset Type	Count	Source
Agent configs	59	.claude/agents/v3/*.md (53 QE + 6 cross-domain)
Skill configs	86	.claude/skills/ (78 QE + 8 general dev)
Tool wrappers	5	Core MCP operations
Transport modes	3	SSE, WebSocket, HTTP

Note: The project has 60 total agents (59 with standalone definitions + 7 TDD subagents that share parent configs, minus overlaps). OpenCode generates 59 individual agent configs from the standalone .md files.