@dzhechkov/adapter-cursor

The Cursor platform adapter for DZ Harness Hub — transforms canonical skills into per-skill Cursor project rules.

dz init --target cursor --select design-thinking

What it emits

Cursor (cursor.com/docs/context/rules) reads project rules from .cursor/rules/ — one .mdc file per rule:

.cursor/rules/<skill-id>.mdc

The extension MUST be .mdc. A plain .md file in .cursor/rules/ is ignored by Cursor.

Each file is YAML frontmatter with Cursor's three keys, then the skill body as Markdown:

---
description: "Human-centered product development orchestrator…"
alwaysApply: false
---

# Design Thinking
…skill body…

| Key | Meaning | |-----|---------| | description | the skill's description — Cursor uses it to decide relevance | | globs | optional file-glob scope; omitted here → a general (unscoped) rule (emitted only if the source skill declares a globs string) | | alwaysApply | false → the rule is agent-requested (pulled in on demand), not always-on |

Intentionally transforming — excluded from byte-identical equivalence

Like copilot, this is not a lossless skill-tree adapter. The canonical SKILL.md frontmatter (~26 project-local keys such as name, trust_tier, version) is replaced by Cursor's own 3-key frontmatter, so the emit is not byte-identical to the source. The cursor target is therefore excluded from the cross-adapter byte-identical equivalence suite (which covers only the five lossless per-skill tree adapters). scripts/ and other assets are not carried.

Per the @dzhechkov/core Adapter contract the transform loss surfaces as a warning; under strict the adapter throws. The canonical skill stays the lossless source of truth — recompiling to claude is still full fidelity.

Exports

• renderCursorMdc(skill) (from @dzhechkov/core) — renders one .mdc file's content.
• CURSOR_RULES_ROOT — .cursor/rules.
• ADAPTER_CURSOR_VERSION — 0.1.1.
• cursorAdapter — the Adapter implementation.