Skip to main content
@zanetta/axion v1.4.0

Spec-driven development for your AI coding agent.

Axion installs a verified SDD workflow into your IDE: six slash-command stages, executable Python gates and persistent memory, so the agent proves its work instead of claiming it.

Domain skills
68+
Slash commands
13
Verification scripts
15
Native IDE adapters
5

Build your install command

Choose your IDE adapter and profile, then copy the command.

--tool

ax-* skills, rules, MCP config, AGENTS.md and GEMINI.md.

--profile

Default profile: core workflow, skills and verification scripts.

npx @zanetta/axion init --tool antigravity --profile standard
  • Cursor
  • Claude Code
  • OpenAI Codex
  • Windsurf
  • Google Antigravity
  • standard
  • fullstack
  • security
  • mobile
  • minimal

Core pillars

Guardrails that run, not guidelines that hope

Every pillar is backed by a file or script in your workspace, so behavior is inspectable and repeatable.

  • Socratic Gate

    No plan and no code until a SPEC exists or at least three real questions are answered.

    eval_routing.py --plan
  • Six-state SDD workflow

    discover, specify, plan, implement, verify and remember — each one a slash command with a concrete artifact.

    /ax-discover → /ax-remember
  • Executable hard gates

    Python scripts return real exit codes. A failed gate means the task is not done — no matter how the code looks.

    verify_gate.py · side_effect_guard.py · circuit_breaker.py
  • Persistent memory

    Conventions and decisions survive across sessions in .agents/memory/, scanned for secrets before every write.

    memory_secret_scan.py
  • End-to-end traceability

    Every SPEC item maps to a PLAN task and an ACCEPTANCE check with execution evidence.

    SPEC ↔ PLAN ↔ ACCEPTANCE
  • Write allowlist

    The plan declares which paths may change. Writes outside the list, and any .env or credentials path, are refused.

    ## Write allowlist

SDD workflow

Six stages from idea to remembered decision

Select a stage to see its command, the artifact it produces and the gate that guards it.

Stage 1 of 6

Discover

Explore the idea with Socratic questions and compare at least three options with pros, cons and effort.

Command
/ax-discover
Artifact
Exploration notes and a recommended option
Gate
No SPEC or code is written yet

Support commands

  • /ax-orchestrate

    Coordinate three or more specialist agents in parallel.

  • /ax-debug

    Evidence-based systematic debugging.

  • /ax-test

    Generate and run tests for changed code.

  • /ax-deploy

    Pre-flight checks and production deploy.

  • /ax-design

    Searchable UI/UX design guide and tokens.

  • /ax-status

    Agent and project progress report.

Agents & skills

Specialists and playbooks, loaded when needed

Personas route work to the right domain, and skills carry the patterns they apply.

16 entries in All

  • Agent

    frontend-specialist

    Web UI/UX with Next.js, React and Tailwind.

  • Skill

    nextjs-react-expert

    React and Next.js performance rules from Vercel Engineering.

  • Skill

    tailwind-patterns

    Tailwind CSS v4 tokens, container queries and patterns.

  • Agent

    backend-specialist

    APIs, server logic and authentication flows.

  • Agent

    database-architect

    Schema design, indexing and migrations.

  • Skill

    api-patterns

    REST, GraphQL and tRPC decisions, pagination and versioning.

  • Agent

    security-auditor

    Security reviews and audit reports.

  • Skill

    vulnerability-scanner

    OWASP-oriented analysis and supply-chain checks.

  • Agent

    test-engineer

    Unit, integration and end-to-end test strategy.

  • Skill

    verify-changes

    Prove code works by running it, not by inspecting it.

  • Agent

    debugger

    Root-cause analysis with evidence-based debugging.

  • Agent

    orchestrator

    Multi-domain coordination across specialists.

  • Agent

    project-planner

    Turns approved specs into phased, traceable plans.

  • Skill

    coordinator-mode

    Parallel workers, synthesis protocol and circuit breaker.

  • Agent

    mobile-developer

    iOS, Android, React Native and Flutter.

  • Agent

    devops-engineer

    Deployment, observability and server management.

Quick start

Install in one command, verify with three

Install the kit, then run the same checks your agent runs before it can call a task done.

Install

  1. 1. Install the kit

    Project the Axion pack into your IDE's native paths.

    npx @zanetta/axion init --tool antigravity
  2. 2. Stay current

    Merge the latest template without deleting installed files.

    npx @zanetta/axion update
  3. 3. Check the install

    Read the manifest and verify the adapter markers.

    npx @zanetta/axion status

Validation pipeline

  1. 1. Master checklist

    Security, lint, schema, tests, UX and SEO in priority order.

    python3 .axion/scripts/checklist.py .
  2. 2. Verify hard gate

    Fails when an acceptance check is open or lacks evidence.

    python3 .axion/scripts/verify_gate.py . --acceptance <slug>.acceptance.md
  3. 3. Traceability report

    Prints SPEC, PLAN and acceptance coverage for one slug.

    python3 .axion/scripts/trace_report.py . <slug>

FAQ

Frequently asked questions

Short answers about installation, adapters and gates.

What does Axion install into my project?

A workspace contract: rules, skills, slash commands, master scripts under .axion/scripts/, a kit manifest and, depending on the adapter, native files such as AGENTS.md, GEMINI.md or MCP configs. Existing files are merged, never silently deleted.

Which AI coding tools are supported?

Cursor (default), Claude Code, OpenAI Codex, Windsurf and Google Antigravity. Pick one with --tool and the pack is projected onto that tool's native paths.

What are profiles?

Profiles filter what gets copied: minimal, standard, security, fullstack and mobile. They never delete files that are already installed. The security profile also enables the offensive surface.

What makes the gates different from prompts?

Gates are executable Python scripts that return real exit codes. verify_gate.py, side_effect_guard.py and circuit_breaker.py give the agent a hard stop instead of a suggestion.

Should I commit the .agents folder?

Yes. Your assistant discovers rules, skills and memory by reading these files in the workspace, so keep them under version control.

Where is the documentation?

The full documentation is coming soon. Until then the Quick Start above and the repository README cover installation and the SDD workflow.