Audit Agent Skills for trust, token budget, duplicates, and tool-surface risk — without leaving Cursor. Run the Skill Development Pipeline, Trust & Safety findings, and manage skill versions directly from your editor.
Built and maintained by ModelBound, the unified knowledge index and MCP tool proxy for AI agents.
All slash commands shell out to the ModelBound CLI via node .modelbound/mb.mjs (installed by this plugin). The launcher uses a global modelbound binary if present, otherwise npx modelbound from npm. Same semantics as the Cursor extension and MCP. You never need skill UUIDs: pass a file path or slug and the CLI syncs + resolves internally.
| Component | Purpose |
|---|---|
Skill · skill-health-lens |
Invoke with /skill-health-lens to run the four core checks |
Rule · skill-authoring |
Inline authoring standards on all watched skill paths |
Trust & Optimize · /mb-findings, /mb-suggest, /mb-compare, /mb-benchmark |
Test & Optimize phase (extension v1.9.16 parity) |
Pipeline · /mb-pipeline, /mb-test, /mb-sync, /mb-context |
Skill Development Pipeline with auto sync |
Versions · /mb-versions, /mb-restore, /mb-diff |
Checkpoint history and restore |
Auth · /mb-login, /mb-logout, /mb-whoami, /mb-health |
Device login + MCP connectivity |
Hook · beforeFileEdit + afterFileEdit on skill files |
Pre-edit backup to .mb-backup/ and token-budget hint on save |
MCP server · modelbound |
Hosted at https://mcp.modelbound.co (requires MODELBOUND_API_KEY) |
/mb-context set --repo org/repo
/mb-sync .modelbound/prompt-pr-contributor.md
/mb-findings .modelbound/prompt-pr-contributor.md
/mb-findings ignore ... --key "escalation:critical:..."
/mb-pipeline .modelbound/prompt-pr-contributor.md --stage test_optimize.modelbound/**/*.md|.json.kiro/skills/**/*.md.cursor/rules/**/*.md|.mdc.claude/**/*.md.agents/skills/**/SKILL.md
npx modelbound-cursor-plugin@latest installRequires Node ≥ 20. Auth via MODELBOUND_API_KEY or run /mb-login once.
| Command | CLI equivalent |
|---|---|
/mb-findings <skill> |
findings list --skill … |
/mb-suggest <skill> |
suggest --skill … |
/mb-compare <skill> |
compare --skill … --from latest --to current |
/mb-benchmark <skill> |
benchmark --skill … |
/mb-pipeline <skill> |
pipeline run --skill … --watch |
/mb-test <skill> |
test --skill … |
/mb-sync <file> |
sync --file … |
/mb-context set |
context set [--repo org/repo] |
/mb-optimize <file|slug> |
optimize … |
/mb-versions <skill> |
versions --skill … |
/mb-restore <skill> <ver> |
version restore --skill … --version … |
/mb-diff <skill> <from> [to] |
version diff --skill … |
/mb-health |
health |
/mb-login / /mb-logout / /mb-whoami |
auth login / auth logout / auth status |
Pipeline, findings ignore, benchmark, compare, and suggest may hit hosted-backend bugs until Lovable deploys fixes. The CLI surfaces these errors explicitly (Pipeline failed:, [MCP_ERROR], etc.). See docs/BACKEND-BLOCKERS.md in modelbound-cli.
scripts/pre-skill-write.mjs snapshots staged skill files (same paths as above) to .modelbound/backups/ before commit and blocks empty files or dropped YAML frontmatter. Opt out with MODELBOUND_SKIP_HOOK=1 or --no-verify.
Edit hooks in hooks/hooks.json back up to .mb-backup/ and print a token-budget hint on save for the same skill paths.
| Project | Description |
|---|---|
| ModelBound CLI · npm | Terminal + CI for token optimization, skill pipeline, and version management |
| ModelBound MCP Server · npm | Local-first MCP server for skill lint, convert, and cloud sync |
| Cursor Extension · Marketplace | VS Code/Cursor extension for rules sync and MCP bridge |
| Claude Code Plugin | Claude Code plugin for pipeline, hooks, and skill sync |
| Dev Packs | Open-source curated AI context packs for engineering teams |
Install hub: modelbound.co/connect
MIT