Five suites ยท one platform

The five agent suites in detail

Click any tab to deep-dive. Each suite has its own inputs, AI pipeline, tech stack and exit criteria.

Suite 01 ยท Planned

๐Ÿ“ Design Suite

Turns raw requirements into an approved design, INVEST story backlog and dependency-aware roadmap. The only suite that loops on Product-Owner comments.

Inputs accepted

  • PRD files โ€” PDF, DOCX, Markdown, TXT
  • Design assets โ€” Figma, PNG, draw.io, XML wireframes
  • Jira epic key โ€” pulls all child design stories
  • Free-form chat instructions from the PO

AI pipeline

  • Ingestion โ€” pypdf, python-docx, Figma API, OCR for diagrams
  • Comprehension โ€” chunk + embed โ†’ vector DB; extract a structured RequirementSpec
  • Design generation โ€” architecture, data model, API stub, UI wireframes
  • Story synthesis โ€” INVEST-compliant stories with ACs + LLM-estimated points
  • Roadmap โ€” dependency graph โ†’ sprint-by-sprint Gantt
  • PO review loop โ€” comments classified, fed back as re-design input

Tech stack

  • LangGraph / LlamaIndex multi-step chains
  • Pydantic schemas: RequirementSpec, DesignDocument, Story, Roadmap
  • Mermaid + Excalidraw for diagrams
  • Atlassian Python SDK for Jira
  • Qdrant / pgvector for retrieval

Exit criteria

PO-approved DesignDocument + StoryBacklog + Roadmap, persisted and version-tagged (design.v1, design.v2โ€ฆ). Hands off to the Development Suite.

Human gate

โœ… PO must approve before any code is written.

Suite 02 ยท Planned

๐Ÿ’ป Development Suite

Implements approved stories in priority order, commits code, opens PRs and stamps every story with full traceability.

Inputs

  • Approved roadmap (story order)
  • Repository handle (GitHub / GitLab / Bitbucket)
  • Coding standards: pyproject.toml, ESLint config, .editorconfig

Plan-and-Code loop (ReAct)

  • Story picker โ€” pulls next Ready story, marks In Progress
  • Context builder โ€” RAG over repo with tree-sitter AST-aware chunking
  • Planner LLM โ€” step-by-step plan
  • Coder LLM โ€” Claude / GPT-4 / Qwen-Coder / DeepSeek emit diffs
  • Self-review LLM โ€” re-reads diff against acceptance criteria
  • Local verify โ€” ruff, eslint, mypy, tsc, existing unit tests

Commit + PR workflow

  • One feature branch per story (feat/PROJ-123-โ€ฆ)
  • Conventional Commits โ€” reuses repo's existing workflow
  • PR description auto-generated, linked back to the story
  • Sandboxed execution in ephemeral Docker containers

Story closure payload

  • What was done โ€” LLM summary of the diff
  • DoD checklist โ€” ticked
  • Code pointers โ€” files + line ranges + permalinks
  • Test pointers โ€” tests added or modified
Suite 03 ยท Planned

๐Ÿงช Testing Suite

Independently verifies that merged code fulfils the design. Maintains a monotonically-growing regression registry and gates the Production Ready state.

Design โ†” Code conformance

  • LLM-as-judge scores every acceptance criterion
  • Mismatches file bugs immediately โ€” no test gen for failed ACs

Test generation

  • Positive (happy path) tests
  • Negative (invalid input, auth, boundary) tests
  • Edge cases (empty, max, unicode, concurrency)
  • Integration tests across services
  • UI tests via Playwright / Cypress
  • Stable ids: TC-<storyKey>-<n>

Execution engine

  • Runs full regression on every story completion
  • Parallelised via pytest-xdist, vitest, playwright workers
  • Captures screenshots, video, console + network HAR for UI tests
  • Security gates: bandit, semgrep, trivy, OWASP ZAP

Bug lifecycle & Prod-Ready gate

  • Any failure โ†’ auto-Jira bug with stack, screenshot, suspected file
  • Originating story reopened, re-queued to Dev
  • Story is Production Ready only when: ACs pass, all its tests pass, no open bugs, regression green
Suite 04 ยท Planned

๐Ÿš€ Production Suite

Promotes Production-Ready artifacts to live environments safely and verifies in-place.

Release composition

  • Bundles all Production Ready stories into a release candidate
  • Generates release notes from story summaries
  • Awaits PO / Release-Manager approval โœ…

Deploy strategies

  • Blue/green, canary, rolling โ€” selectable per environment
  • IaC: Terraform / Helm / Pulumi
  • Targets: Kubernetes, ECS, Lambda, Azure App Service
  • Reuses existing executors (Terraform, Kubernetes, Docker)

Post-deploy verification

  • Re-runs the Testing Suite against the production URL
  • Critical-path smoke now, full regression off-hours
  • Synthetic monitoring: uptime, Lighthouse, k6 baseline

Promotion / Rollback

  • Green โ†’ promote canary to 100%
  • Red โ†’ automatic rollback + bug + story re-queued
  • GitOps via ArgoCD / FluxCD
โ— Suite 05 ยท Live now

๐Ÿ›ฐ๏ธ DevOps Suite

The suite that already exists in this repository โ€” the other four are being built around it. Watches the live system, root-causes incidents, self-heals or files a bug story back to Dev.

DevOps deep dive โ†’

Capabilities live today

  • Pluggable agents: Jenkins, K8s, GitHub Actions, Ansible, Terraform, Docker, Git, generic API
  • Failure-context extraction
  • Decision engine with per-agent thresholds
  • Memory + audit store
  • Remediation orchestrator
  • Jira / PagerDuty / ServiceNow integrations
  • Grafana dashboard
  • Daily test runner + scheduler

Loop closure

When the DevOps agent cannot self-heal:

  • Builds a FailureContext (logs, metrics, traces, suspect commit)
  • Drafts a bug-story (title, repro, ACs for the fix)
  • Pushes to Jira and tags the originating release
  • Hands off to the Development Suite โ€” lifecycle restarts

What's coming next

  • Phase 2 โ€” multi-tenant SaaS mode
  • Phase 3 โ€” real-time RCA streaming dashboard
  • Phase 4 โ€” predictive failure detection
  • Phase 5 โ€” auto-PR generation for known patterns
See the full roadmap โ†’

Start with the live suite, grow into the full lifecycle

The DevOps Agent is production-ready today. The remaining four suites land in phases.