Claude pluginClaude Code
- Manifest
.claude-plugin/plugin.json- Workflows
commands/*.md and skills/*/SKILL.md- Distribution
- Claude plugin marketplace
Package Line: 1.0.0 · Public Release
Bring governed AI engineering into the agent workspace.
One portable EAINE assessment workflow, projected for Claude Code, Cursor, and Visual Studio Code. The packages help teams inspect architecture and the complete AI SDLC without turning AI output into approval, certification, or release authority.
eaine-agent-plugins · v1.0.0Validation Active
$eaine audit --target workspace --rules default-deny
✔Framework:EAINE v1.0
ℹProjections:Claude Code · Cursor · Visual Studio Code
✔Portable Skills:7 skill definitions · 11 workflow commands
✔Validation:Structural checks complete
⚡Runtime status:Behavioral validation pending
🔒Capability:Advisory local assessment (Non-enforcing)
Three host projections
Native packaging, one governed meaning.
Each projection carries the same seven Skills, eleven workflow commands, knowledge references, assessment templates, evidence boundaries, and EAINE v1.0 limitations. Host-specific packaging cannot widen the portable EAINE authority boundary.
Agent Plugins 1.0Cursor
- Manifest
plugin.json- Workflows
rules/*.mdc, commands/*.md, and skills/*/SKILL.md- Distribution
- Cursor Marketplace
Agent Plugins 1.0Visual Studio Code
- Manifest
plugin.json- Workflows
com.github.copilot/* and skills/*/SKILL.md- Distribution
- Agent plugin marketplace or Git source
Direct IDE Protocol · Streamable HTTP & SSE
Explore MCP Server →Connect via Model Context Protocol (MCP)
Prefer direct tool execution over plugins? Connect Cursor, Claude Desktop, Google Antigravity, and VS Code directly to our live remote MCP server (https://mcp.eaine.org/mcp) or run the zero-network local stdio bundle.
AI SDLC coverage
Every lifecycle stage stays visible.
The lifecycle is iterative. Experiment, emergency, rollback, improvement, and retirement paths do not waive evidence, security, human authority, or release decisions.
- 01Discovery
- 02Requirements Engineering
- 03Architecture and Design
- 04Development
- 05Quality Engineering
- 06Security Validation
- 07Deployment and Release
- 08Operations
- 09Continuous Improvement
Assessment workflows
Architecture, risk, evidence, and decisions remain connected.
- Scope and applicability
- Inherent risk and tailoring
- Nine-stage lifecycle planning and review
- Architecture and authority review
- Requirement-level conformance proposals
- Separate maturity observations
- Evidence-gap review and bounded export
Architecture review covers components, trust zones, identities, interfaces, data and knowledge flows, authority contracts, control planes, failure containment, rollback, recovery, operations, and retirement. It proposes findings; it does not enforce controls.
Validation boundary
A public release, not an unverified claim.
The packages are available for governed validation preparation. Installation, behavioral compatibility, decision equivalence, and project outcomes still require target-runtime and independent human evidence.
What is already checked
- Package structure and manifest presence
- Seven portable Skill definitions
- Eleven workflow command projections
- Portable knowledge, templates, and schemas
- Per-file SHA-256 inventories
- Default-deny authority invariants
What remains open
- Behavioral testing in each supported host version
- Skill selection, command routing, and instruction precedence
- Relative-resource resolution and refusal behavior
- Decision-equivalence testing across projections
- Independent assessment and field evidence
- Public distribution and marketplace authorization
Marketplace path
Validate privately, then publish through native stores.
Use a version-pinned private Git marketplace for named validation teams. Submit the same frozen release to public stores only after runtime tests, provenance, support, and distribution approval are complete.
Claude Code
- Validate package structure and tool definitions in Claude Code runtime.
- Publish from a version-pinned Git repository or private enterprise registry.
- Submit reviewed listing through Anthropic's plugin verification channel.
- Link the verified manifest and checksum digest back to EAINE.
Cursor
- Test .cursor/rules (.mdc) and command routing in the Cursor IDE.
- Distribute via team-pinned Git repository or private extension catalog.
- Submit reviewed package through Cursor's marketplace submission portal.
- Publish release notes with pinned commit hash and policy baseline.
Visual Studio Code
- Validate Copilot Agent extension and skill manifests in VS Code runtime.
- Publish from an internal Git endpoint or private extension gallery.
- Submit reviewed extension through the Visual Studio Marketplace review channel.
- Verify SHA-256 integrity and lock evidence bindings to the release.
EAINE does not claim permanent feature parity between providers. Host requirements and review processes can change; confirm the linked official guidance before each submission.