Cursor adapter: running the kit in Cursor

The kit is two layers. The method (the docs/claude/*.md playbooks) is plain markdown and tool-neutral, so it works in any assistant. The packaging (auto-loading, subagents, memory) was Claude Code-specific. This adapter ports the packaging to Cursor so the repo is dual-target.

What maps to what

Claude Code	Cursor	Notes
CLAUDE.md (auto-loaded context)	.cursor/rules/*.mdc with alwaysApply: true	Cursor doesn't read CLAUDE.md. The method ships as operating-method.mdc; project specifics as project-context.mdc.
The 5 method docs (docs/claude/)	same files, referenced from rules via @path	Tool-neutral. Kept as the single source of truth; rules stay thin and @-reference them.
.claude/agents/* (subagents w/ tools:)	Cursor custom modes or manual prompts	No direct equivalent. Degradation: Cursor can't structurally restrict a mode's tools, so "least-privilege tools" weakens from a wall to an instruction. Compensate by flagging risky ops for approval.
File-based memory + auto-recall	.cursor/rules/lessons.mdc (Agent-Requested) or Cursor Memories	The lessons port; the automatic recall mechanism differs.
BOOTSTRAP-PROMPT.md	BOOTSTRAP-CURSOR.md	Same flow, Cursor-worded; installs docs + .cursor/rules.

The rules shipped (.cursor/rules/)

• operating-method.mdc: alwaysApply: true. The persona + OOB + hard gates, inline, plus @-links to the full playbooks. The always-on core.
• project-context.mdc: alwaysApply: true. Template for this project's stack, conventions, locked decisions. The CLAUDE.md-equivalent; the bootstrap fills its {{placeholders}}.
• code-change.mdc: auto-attached (globs on source files). Points to the vigilance protocol + field notes when editing code. Tune the globs to your repo.
• ship-and-recover.mdc: Apply-Intelligently (description, no globs). Deploy discipline
  • the incident-response playbook; the Agent pulls it in for deploys/outages.

Rule-type frontmatter (Cursor)

.cursor/rules/*.mdc files use three fields: description, globs, alwaysApply.

• Always → alwaysApply: true
• Auto-attached (file-matched) → alwaysApply: false + globs
• Apply-intelligently (Agent decides) → alwaysApply: false + description, no globs
• Manual (@-mention only) → alwaysApply: false, no description/globs Plain .md in .cursor/rules/ is ignored unless it's .mdc (or named AGENTS.md).

Install

Use BOOTSTRAP-CURSOR.md (paste into the Cursor Agent, which self-clones and installs), or do it by hand: copy docs/claude/* and .cursor/rules/* into your repo, fill project-context.mdc, tune code-change.mdc globs, and fold memory-seeds/* into a lessons.mdc rule or Cursor Memories.

What you keep vs lose in Cursor

• Keep (≈90%): the whole method (persona, OOB, vigilance, field notes, incident response, the gates), applied automatically via always-on + auto-attached rules.
• Lose / degrade: structural tool-restriction on delegated agents (Cursor modes can't wall off tools the way Claude Code subagents can); automatic cross-session memory recall (becomes a rule). Both have the compensations noted above.

Cursor's rules format and Memories feature evolve quickly. If something here doesn't match your Cursor version, the frontmatter fields (description/globs/alwaysApply) are the part to trust; check the current Cursor docs for the rest.