agent-lint

Read-only MCP server that lints agent skills, rules, and hooks/settings — with trust posture analysis and citation-backed rubrics.

Quick Start

npm install -g agent-lint
agent-lint --setup

That's it. Setup detects your AI clients (Cursor, Claude Code), registers the MCP server, and creates trigger rules so linting runs automatically when you edit agent config files.

Verify

agent-lint --doctor

What it checks

Tool	Purpose
agent_lint_scan	Discover artifacts (inventory only)
agent_lint_assess	Score artifacts — grade, issues, trust posture
agent_lint_suggest	Per-artifact improvement suggestions
agent_lint_explain_rubric	Rubric metadata (no filesystem reads)
agent_lint_trust_posture	Trust posture with five evidence axes

Manual setup

If you prefer to configure manually instead of using --setup:

Cursor — add to ~/.cursor/mcp.json:

{ "mcpServers": { "agent-lint": { "command": "agent-lint", "args": [] } } }

Claude Code — add to ~/.claude/settings.json:

{ "mcpServers": { "agent-lint": { "command": "agent-lint", "args": [] } } }

Disabling auto-trigger

Delete the rule file from your project:

• Cursor: .cursor/rules/agent-lint.mdc
• Claude Code: .claude/rules/agent-lint.md

MCP tools remain available for manual use — only automatic triggering stops.

Documentation

• Tools reference
• Trust posture
• Rubric & versioning
• Development

License

MIT — see LICENSE.