Architecture Guard is an SDD governance & orchestration engine that unifies your backlog, specifications, multi-agent implementation, and PR queue across any code host — keeping AI development disciplined, verified, and aligned.
Open source • Works with Claude Code, Antigravity, OpenCode & Codex • MIT License
From raw backlog intent to autonomous pull request management, every phase is grounded by your architecture constitution and verified against drift.
Brainstorm with Flash-Mem memory context and Jira/Beads story scope.
/ag-governed-discover
Schema delta specs verified against base-branch code with Premise Grounding.
/ag-governed-spec
Technical design and task breakdowns enforcing Ponytail minimalism.
/ag-governed-plan
Parallel implementation via Claude Teammates, OpenCode, or Antigravity.
/ag-governed-implement
Consumer inventory, hygiene runner, Sonar rules, and change checks.
/ag-verify base=main
Autonomous PR triage across GitHub, GitLab, Bitbucket, and Azure DevOps.
/ag-pr-queue
Brainstorm features with Flash-Mem historical context and Jira/Beads story scope. Architecture Guard warns on architectural drift early and decomposes multi-item initiatives into clean backlog items.
Architecture Guard avoids vendor lock-in. It coordinates native teammates and subagents on your preferred AI platform under one unified safety floor.
Generates agteam-*.md specialist roles (Author, Reviewer, Implementers, Quality Gate) with task ownership and peer messaging.
Generates skills and workflows (agx-*.md) using invoke_subagent for parallel reviews and isolated task work.
Lead coordinates native local subagents (agoc-*) with strict permission envelopes ("*": deny) and sequential fallback.
Installs skills in .codex/skills/ with sequential lead execution and isolated child session delegation for targeted reviews.
Explore real terminal commands and orchestrated phases for single-feature delivery, batch backlogs, and PR queues.
Single-feature end-to-end flow: automatically chains Spec, Plan, and Task generation with auto-fixing review loops before implementation.
# 1. Explore idea with architecture context & story scope /ag-governed-discover "Real-time analytics dashboard" # 2. Automatically generate Spec, Plan, and Tasks with artifact repair loops /ag-governed-delivery # 3. Implement tasks within strict architecture boundaries /ag-governed-implement # 4. Verify against destination branch (same checks as PR review) /ag-verify base=main # 5. Open PR and manage feedback queue /ag-pr-queue
Architecture Guard bridges your high-level architectural rules with agentic execution across every phase.
Supports OpenSpec, SpecKit, and custom adapters with schema-aware delta specs and automatic change scoping.
Premise Grounding against real base-branch code, Consumer Inventory, Artifact Consistency, and Hygiene scans.
Deliver Jira JQL, Beads, or YAML backlogs across Full, Fix, Light, and Research lanes with 4 batched gates.
One command scans PRs, triages feedback in isolated worktrees, and reviews team code across all major forges.
Coordinate Claude Code Teammates, Antigravity Subagents, OpenCode, and Codex with strict file boundaries.
Architecture Guard connects SDD frameworks, code forges, trackers, and AI agent hosts.
Architecture Guard coordinates AI-assisted development with enterprise rigor — protecting architectural boundaries so you can ship faster without technical debt.