Updated 2026-10-02

Agent Plugins

LyraShield prepares a portable plugin following the open Agent Plugins v1.0.0 format. One package bundles two things: the LyraShield MCP server, which gives your agent the scan, finding, fix-plan and verify tools and a security skill that tells the agent when to reach for them. Each client still activates the package through its own supported path.

Public Agent Plugin listings pending

The portable Agent Plugin package @lyrashield/agent-plugin@0.1.31 is published and passes offline Agent Plugins schema validation. Public marketplace listings and authenticated client acceptance remain pending. The marketplace default branch is preparation only. Until each client's listing and runtime are verified, follow its direct MCP guides with @lyrashield/mcp@0.2.12 and authenticate separately using npx -y lyrashield@0.2.14 login --oauth.

CLI config and skills setup

The pinned npm CLI is version 0.2.14. Config-writing commands are available only where the guide provides an exact installer command. Shared skills can be installed only for clients with one documented skill path. For other clients, use the linked guide's documented manual or client-native setup.

npx --yes lyrashield@0.2.14 install vscode --dry-run

Review the output, then run npx --yes lyrashield@0.2.14 install vscode to write the supported VS Code entry. Use the linked client guides for other commands and skill paths. A config or skill install does not prove the client discovered or authenticated the integration.

Current guided setup paths

The published portable source package contains client artifacts, but its presence does not establish a public listing or authenticated client acceptance. Kiro's native Power distribution is in preparation and is not publicly installable yet; its guide documents the MCP settings fallback. Other clients use the configuration, vendor CLI or guided-manual path documented in their own guide. Marketplace availability and runtime acceptance are tracked independently.

What the skill adds

The MCP server alone gives an agent tools; the bundled skill gives it judgement about depth. It maps a request like a pre-commit diff check to a quick bounded review and a deep security review to the intrusive agentic mode that only runs inside the authorised isolated sandbox for a repository target. That keeps a routine check from silently becoming a long expensive scan. Install shared skills with the pinned CLI only when the client guide documents a matching skill directory; skills are not proof of marketplace availability or client acceptance.

Findings are proposals, not automatic changes

Installing the plugin does not let an agent rewrite your code unattended. During remote OAuth, the user chooses one workspace and the requested workflows. Matching calls can then run without repeated LyraShield approval, while current role, target access, profile, budget, connection, scope, expiry and idempotency checks still apply. Mutating calls authenticated with an API key receive a connect_required response pointing at OAuth connect and out-of-grant delegated calls are denied. Historical exact-input approval records are separate from active hosted delegation. Fixes and pull requests remain proposals requiring review before they land; LyraShield never merges them automatically.

Uninstall

Marketplace and UI installations must be removed through that client's own plugin manager. For CLI-managed config, preview removal for a supported client, then remove only the LyraShield entry. Preserve all other configuration.

npx --yes lyrashield@0.2.14 uninstall vscode --dry-run
npx --yes lyrashield@0.2.14 uninstall vscode

Frequently asked questions

What is the LyraShield Agent Plugin?

It is a portable package that carries the LyraShield MCP descriptor and security guidance. Each client still controls activation through its own documented local, marketplace or MCP settings path.

Which clients have a native package path?

Portable plugin artifacts are prepared for Claude Code, Cursor, OpenAI Codex and GitHub Copilot, but their marketplace listings remain pending review and public readback. Use the direct MCP setup in each guide until those listings are verified. Kiro's native Power distribution is also in preparation; use its MCP settings fallback. CLI config installation and shared skills are separate and appear only for exact supported client contracts and documented skill paths.

Do I need to log in before installing the plugin?

For local stdio, run npx --yes lyrashield@0.2.14 login --oauth first so the MCP server can read the selected workspace credential. Hosted clients use their documented OAuth flow. The plugin listing remains pending; follow each client guide for its supported direct MCP fallback.

Does the plugin let the agent change my code on its own?

No. LyraShield findings and fixes are proposals requiring review before they land. A browser-confirmed OAuth delegation can run its recorded LyraShield workflows without repeated approval, but it remains limited by the user's current role, workspace, targets, profiles, budget, scope and connection expiry. LyraShield never auto-merges a pull request.

Why does VS Code use an MCP config file instead of the plugin?

No generated VS Code Agent Plugin shim exists. Its documented config path is .vscode/mcp.json, and CLI 0.2.14 includes a matching config installer. Preview it with npx --yes lyrashield@0.2.14 install vscode --dry-run, then run the install command after reviewing its changes. The VS Code guide documents the servers root key and per-entry stdio type. Runtime verification remains client-specific.