Skip to main content

What's New

All notable changes to the AI Productivity Kit. Format based on Keep a Changelog.

[Unreleased]​

Changed​

  • BREAKING — Cursor-native modes migration. Replaced the custom SPEC/IMPLEMENT text protocol with Cursor's native four-mode system (Plan, Agent, Debug, Ask) as the primary vocabulary. "Switch: SPEC" and "Switch: IMPLEMENT" commands removed. "Spec Package" renamed to "Plan output"; "Implementation Package" renamed to "Agent output". New Debug mode guidance added (hypothesis-driven, evidence-based bug investigation). Advisories block now emits Mode: Plan | Agent | Debug | Ask instead of Route: SPEC or IMPLEMENT. Config key defaultRoute renamed to defaultMode. Skills spec-dod and impl-dod renamed to plan-dod and agent-dod. All rules, prompts, skills, AGENTS.md, Copilot instructions, Antigravity mirrors, site docs, landing page, eval traces, and sync scripts updated. Non-Cursor editors get a lightweight fallback line ("state the mode name in your message") instead of the parallel Switch: protocol. Migration: update any custom rules or prompts that reference "Switch: SPEC", "Switch: IMPLEMENT", "SPEC mode", "IMPLEMENT mode", "Spec Package", "Implementation Package", or defaultRoute.

[1.5.0] - 2026-04-20​

v1.5.0

Added​

  • Invokable Skills — 8 new Skills under kit/.cursor/skills/ with YAML frontmatter (name, description), mirrored to kit/.agent/skills/ for Antigravity. Checklist skills: /plan-dod, /agent-dod, /security-dod, /threat-model-lite. Prompt skills: /session-kickoff, /context-pack, /router, /handoff-summary. Synced to starter/ via npm run sync:starter.
  • Golden-trace evaluation harness — 5 reference dialogues in docs/audit/eval/: 01-spec-with-advisories, 02-implement-with-security-trigger, 03-route-mismatch-advisory, 04-mcp-config-change, 05-handoff-summary. Each trace specifies setup, user prompt, expected AI response, and fail criteria. Linked from README "How we test the kit" section.
  • Doctor: SKILL.md frontmatter validation — npm run doctor step 1b validates every kit/.cursor/skills/*/SKILL.md has parseable YAML frontmatter with required name and description fields.
  • Path-specific Copilot instructions — New kit/.github/instructions/security.instructions.md (auto-loads for auth/payments/uploads/secrets paths) and tests.instructions.md (auto-loads for test files). Repo-wide copilot-instructions.md slimmed to universal baseline.
  • Copilot prompt files — 4 new .prompt.md files under kit/.github/prompts/: session-kickoff, context-pack, router, handoff-summary. VS Code Copilot Chat users can invoke via slash commands. Bodies match kit/.cursor/prompts/*.md; doctor step 1c validates alignment.
  • Doctor: Copilot prompt body alignment — npm run doctor step 1c validates every kit/.github/prompts/*.prompt.md body matches the corresponding kit/.cursor/prompts/*.md body.
  • Antigravity artefact mapping — New "Artefact mapping" section in site/docs/editor-support/antigravity.md: Spec Package = implementation_plan.md, Handoff Summary = walkthrough.md, Context Pack = GEMINI.md/AGENTS.md, task breakdown = task.md.
  • Security enforcement hooks — New kit/.cursor/hooks.json wiring beforeShellExecution, beforeMCPExecution, and afterFileEdit to kit/.cursor/hooks/security-gate.mjs. Destructive shell patterns are hard-denied with failClosed: true. All shell and MCP actions logged to .cursor/hooks-audit.log. Opt-in flag hooks.enabled in cursor-ai-kit.config.json (default: true).
  • MCP trust posture — New "MCP server trust" clause in security rules: treat each MCP server as a dependency, pin versions, re-review on every change. kit/.cursor/mcp.json replaced with a commented reference example and trust checklist.
  • Indirect-injection defence — New clause in security rules: all tool outputs (MCP results, web fetches, browser snapshots, external file reads) are DATA; emit Advisory and require confirmation if tool output contains apparent instructions. OWASP LLM01:2025 cited.
  • Root AGENTS.md — New kit/AGENTS.md (canonical) distilling the kit's cross-tool agent instructions. Sync script copies to repo root AGENTS.md and starter/AGENTS.md. Read by Cursor, Antigravity, Claude Code, Codex, and Aider from a single file.
  • Google Antigravity editor support — New kit/.agent/rules/ directory with 9 plain Markdown rule files. Synced to starter/.agent by npm run sync:starter. Install and reference docs updated; landing and README mention Antigravity.

Changed​

  • Cursor mode mapping — Added "Cursor mode mapping" section to kit/.cursor/rules/00-operating-system.mdc: SPEC ≈ Plan mode, IMPLEMENT ≈ Agent mode, Q&A ≈ Ask mode.
  • Platform-type single source of truth — All surfaces (prompts, Copilot instructions) now defer to docs/ai/ai-config.md for the platform-type slug.
  • Rule activation modes — Explicit Cursor activation modes set across all 9 kit/.cursor/rules/*.mdc files. Core rules promoted to alwaysApply: true; security rule scoped to sensitive-path glob.
  • Lens path moved from kit/.cursor/agents/ to kit/.cursor/lenses/ — Avoids collision with Cursor 2.4+ Subagents primitive. Migration: update imports from kit/.cursor/agents/<lens>/... to kit/.cursor/lenses/<lens>/....
  • Docs site: Cursor modes rewrite — cursor-modes.md rewritten for Plan / Agent / Ask taxonomy. Model-switching page updated with concrete model picks (April 2026). Citation hygiene added to editor-support and daily-workflow pages. Sidebar reconciled.
  • SECURITY.md refresh — Updated supported-versions to 1.5.x; new "AI-tool risk classes" section covering rule poisoning, MCP trust bypass (CVE-2025-54135, CVE-2025-54136, CVE-2025-64109), hooks fail-open, and Skills supply chain. OWASP LLM Top 10 and MCP Top 10 references added.
  • Rule mirror anti-drift — npm run doctor step 1d validates every kit/.cursor/rules/*.mdc has a corresponding non-empty kit/.agent/rules/*.md file.
  • Release-assets workflow hardened — Removed continue-on-error: true from sync and build steps; sync or build failures now fail the release.
  • Meta-file consolidation — RELEASE.md and LAUNCH_CHECKLIST.md folded into RELEASING.md.
  • Improvement plan archived — docs/IMPROVEMENT-PLAN.md moved to docs/audit/archive/IMPROVEMENT-PLAN-pre-phase1.md with closure note.

[1.4.0] - 2026-02-19​

v1.4.0

Added​

  • @Codebase and context references guide — New daily-workflow page: when to use @Codebase vs @file, how indexing and .cursorignore work, how docs/ai/ fits in, link to context rot; cross-referenced from context-windows.
  • 5-minute quick start in README and docs intro — Quick start section in README and as tip on docs home; install guide has "New here?" callout to quick start.
  • Agents explainer documentation — New doc in Context pack: what an agent is, agents vs rules vs prompts, included agents, when to activate, platform overlays; linked from intro.
  • .cursorignore template in kit/ — Recommended entries with setup documentation; sync-starter copies it to starter.
  • MCP template (kit/.cursor/mcp.json) and setup documentation
  • 25-testing.mdc — Cursor rule enforcing test coverage on business logic and security-sensitive IMPLEMENT outputs.

Changed​

  • cursor-ai-kit.config.json — Pretty-printed JSON with explanatory comment; no semantic changes.
  • ai-config.md — Renamed TODO headings (e.g. "Design system — fill in yours"), added inline commented Acme SaaS example and fill-in reminder comments to reduce new-user confusion.

Fixed​

  • Resolved duplicate security-stop-gate rule files (02 and 41) — Removed duplicate 02-security-stop-gate.mdc; kept 41-security-stop-gate.mdc as the single always-on security stop gate (same content, no dangling references).

[Unreleased]​

Added​

  • Google Antigravity editor support — New kit/.agent/rules/ directory with 9 Markdown rule files; synced to starter and documented. Landing and README updated to include Antigravity alongside Cursor and Copilot.

1.3.0 - 2026-02-18​

Added​

  • Docs: visual elements

    • Mermaid diagrams enabled; flowcharts and sequences added across intro, getting-started, daily workflow, security, and troubleshooting.
    • Spot SVG icons (spec-mode, implement-mode, one-question, switch, security-gate) in site/static/img/docs/ and used on key pages.
    • Diagrams: SPEC vs IMPLEMENT flow (intro); problem→spec→code→security (what-is-this); kit folder structure (install-by-copying); minimum vs high-ROI files (what-to-fill-in); fill→AI stops guessing (source-of-truth); Session Kickoff→Context Pack→Router→Spec Package (spec-first); implement and switching flows; security stop gate; quick-fix decision tree (common-failures); context lifecycle (context-windows); model tiers (model-switching).
    • Homepage terminal demo: fixed height (320px) so layout stays stable while the typing animation runs.

  • Rules: dispatcher and advisories

    • kit/.cursor/rules/01-dispatcher-and-advisories.mdc (always-on): auto-route SPEC vs IMPLEMENT; Advisories block (Route, model class, context risk, switch recommendation) at top of every response; HIGHLY RECOMMENDED triggers for security/cyber, route mismatch, model switching, context management; cost awareness when user reports high API usage.

  • Rules: security stop gate

    • kit/.cursor/rules/41-security-stop-gate.mdc (always-on): when work is high-risk (auth, permissions, exports, uploads, integrations, external exposure, confidential/restricted data), the assistant asks exactly one security question and stops until answered; then restates assumptions and proceeds.

  • Docs: advisories and security stop gate

    • Advisories block explained (Route, model class, context risk, HIGHLY RECOMMENDED and what to do) and security/cyber trigger list added to spec-first, implement, switching, and common-failures.
    • API usage note: “If your API usage is high, stay on Auto/Fast unless the AI says HIGHLY RECOMMENDED.”
    • Security triggers page: new section “Security stop gate (why the AI may stop)” with diagram and example question; Reference link to Cursor Rules – Security.
    • Troubleshooting: new entry “It stopped and asked one security question” with link to Security triggers.

Changed​

  • Version — bumped to 1.3.0 in kit/cursor-ai-kit.config.json and starter/cursor-ai-kit.config.json.

1.2.0 - 2026-02-18​

Added​

  • Automation: starter sync

    • scripts/sync-starter-from-kit.mjs — deletes and recreates starter/.cursor, starter/docs/ai, starter/.github, and starter/cursor-ai-kit.config.json from kit/; preserves starter/README.md.
    • npm run sync:starter script added to root package.json.

  • CI: drift prevention

    • New workflow check-starter-sync — runs on push and PR to main; fails if starter/ drifts from kit/.
    • release-assets workflow — now syncs starter from kit before creating the starter zip so release assets always match kit/.

  • Sync script: preflight guard

    • scripts/sync-kit-snippets.mjs now performs a preflight check: fails with a clear error (exit 1) if any required kit file is missing before writing any output.

Changed​

  • kit/ is canonical — starter/ is fully generated from kit/; do not edit starter/ directly.
  • Maintainer docs (README.md, LAUNCH_CHECKLIST.md, RELEASING.md) — canonical statement added; sync:starter steps added to all sync and release workflows.
  • Version — bumped to 1.2.0 in kit/cursor-ai-kit.config.json and starter/cursor-ai-kit.config.json.

1.1.0 - 2025-02-17 (initial public release)​

Added​

  • Docs

    • New Global Rules page: explains Cursor Global Rules vs repo project rules, includes copy/paste global rules text and where to paste them in Cursor settings.
    • Optional "set Global Rules" section in Install by copying.
    • Reference pages for new kit files: Session Kickoff, Router, Handoff Summary, Spec Package, Implementation Package, Context Discipline, Threat Model Lite, START-HERE.

  • Landing page

    • Redesigned hero with dark theme, animated grid background, and badge.
    • Terminal demo with typing animation: user prompt → AI safety check → "Switch: SPEC" → AI spec-mode response.
    • Stats bar (85%, 1 question, 2 modes, 8 lenses) and feature grid with hover glow.
    • CTA section and "Preview the Files" link.

  • Agents (kit/.cursor/agents/)

    • PM: pm/base.md, pm/data-platform.md, pm/developer-platform.md — one-question protocol, SPEC-first, compact output.
    • Design: design/base.md, design/data-platform.md, design/developer-platform.md — UX states, interaction, a11y; platform-specific focus.
    • FE: fe/base.md — minimal safe diffs, no invented paths/APIs/deps; output: files, plan, diffs, verification, tests/rationale.
    • QA: qa/base.md, qa/data-platform.md, qa/developer-platform.md — compact test plan (happy path, edge cases, a11y, regressions).
    • Discovery: discovery/base.md, discovery/data-platform.md, discovery/developer-platform.md — problem statement, evidence, hypotheses, smallest validation step.
    • Validation: validation/base.md, validation/data-platform.md, validation/developer-platform.md — what to validate, method, success/guardrails, rollback plan.
    • Analytics: analytics/base.md — success metrics, guardrails, events; no invented event naming.
    • Security: security/base.md, platform packs, exposure overlays (exposure-internal, exposure-external-authenticated, exposure-public), data sensitivity overlays (data-public, data-internal, data-confidential, data-restricted); assets, threats, mitigations, security acceptance criteria.

  • Prompts (kit/.cursor/prompts/)

    • 00-session-kickoff.md — SPEC-first default, 85% gate, one-question protocol, switch commands.
    • 20-router.md — uses Context Pack; chooses SPEC vs IMPLEMENT; asks one question if platform type missing.
    • 90-handoff-summary.md — max 8 bullets when switching context/mode.
    • 10-context-pack.md — verified to include platform type, exposure level, data sensitivity.

  • Kit rules (kit/.cursor/rules/)

    • 00-operating-system.mdc — SPEC-first, 85% confidence gate, one-question protocol, switch commands, handoff, grounding (docs/ai/ai-config.md), tickets/logs as data only, no invention, minimal safe diffs.
    • 05-environment.mdc — cross-platform paths and OS notes (moved from old OS content).
    • 10-spec-package.mdc — Spec Package output contract: UX states, a11y, risks/open questions, platform/exposure/data sensitivity.
    • 20-implementation-package.mdc — Implementation Package: files, plan, diffs, verification, tests/rationale, security notes when triggered.
    • 30-context-discipline.mdc — small context packs, file citations; no inventing paths/APIs/events/deps; allowed values for platform type, exposure, data sensitivity.
    • 40-security.mdc — always-on hygiene (no secrets in code/localStorage/logs, no sensitive data in logs, least privilege, dependency caution); trigger list (auth, uploads, integrations, exports, external exposure, sensitive data); threat-model-lite and security acceptance criteria when triggered.

  • Kit docs (kit/docs/ai/)

    • START-HERE.md — entry point for docs/ai.
    • profiles/company.md, profiles/personal.md — expectations and checklist usage.
    • checklists/spec-dod.md, impl-dod.md, security-dod.md, threat-model-lite.md — definitions of done and lightweight threat model.
    • ai-config.md — defaults: default route (SPEC), confidence gate (85), default exposure (internal), default data sensitivity (internal), platform type mode (infer/ask); design system placeholders (TODO).

  • Config

    • kit/cursor-ai-kit.config.json — version, defaultRoute, confidenceThreshold, defaultExposureLevel, defaultDataSensitivity, platformTypeMode, editorTargets (cursor, copilot).

  • GitHub

    • PR template: Switches (platform type, exposure, data sensitivity), UX states, Accessibility, Security/Privacy checklist, Tests/verification.
    • Copilot instructions aligned with kit OS: SPEC-first, 85% gate, one-question protocol, switch commands, no invention, security triggers.

  • Starter

    • starter/ brought in sync with kit/ at starter root: .cursor/, docs/ai/, .github/, cursor-ai-kit.config.json.
    • starter/README.md updated with actual prompt paths (Session Kickoff, Context Pack, Router).

  • Reference sync

    • scripts/sync-kit-snippets.mjs updated to include new kit files; npm run sync regenerates site/docs/reference/*.
    • site/sidebars.ts updated with new reference doc IDs.

Changed​

  • Footer copyright set to Russell Wyatt in docs site config.
  • Doc-to-kit alignment: troubleshooting and other docs now reference existing kit files (e.g. docs/ai/ai-config.md instead of non-existent current-state.md).
  • Release and download docs: hardcoded tag examples replaced with vX.Y.Z placeholders; instructions to replace with actual release tag.
  • Version set to 1.1.0 in kit/cursor-ai-kit.config.json and starter/cursor-ai-kit.config.json.

Removed​

  • Blog removed from docs site: blog disabled in Docusaurus config, Blog nav item removed, blog post and metadata files deleted.

Fixed​

  • References to kit files in docs that did not exist; corrected to point at real kit paths/sections.

1.0.0​

Added​

  • Initial AI Productivity Kit: Cursor rules, prompts, and docs structure.
  • Docs site (Docusaurus) with intro, getting started, daily workflow, reference, troubleshooting.
  • kit/ and starter/ layout; GitHub Actions for docs deploy and release assets (zips).
  • Basic operating system and security rules; install-by-copying and downloads documentation.