MindContext Core is session persistence infrastructure for coding agents, built on composable hooks and a unified /next workflow that keeps you in context across sessions.
Every time you start a Claude Code session, Claude has no idea what you were working on yesterday. Memory compaction wipes out your decisions. Context switching means re-explaining everything.
MindContext Core fixes this.
When your session starts, hooks automatically inject your current work context—the issue you're implementing, where you left off, which files are hot. You don't have to remember or re-explain anything.
When you're ready to work, you type /next. It finds your in-progress issue, or shows you what's ready to pick up. No hunting through task lists or trying to remember what was blocked.
When you finish something or need to switch context, /next shift saves your progress notes and file context to beads—a git-backed issue tracker that survives memory compaction. Your notes become comments on the issue, not ephemeral session state that disappears.
When you come back tomorrow, it all gets re-injected automatically. You pick up exactly where you left off.
The system is designed to be minimal and compatible. It doesn't enforce a methodology—it provides the foundation that any workflow can build on. Use it with OpenSpec for spec-driven development, with superpowers for TDD enforcement, or on its own for just session persistence.
# Add marketplace
/plugin marketplace add tmsjngx0/mindcontext-core
# Install
/plugin install mindcontext-core@tmsjngx0Verify with /help—you should see /next, /commit, /project-init.
-
SessionStart - Activates when session begins. Injects your in-progress beads issue, HOT files to read, progress notes from last session. You start with context, not a blank slate.
-
/next- Activates when you need work. Shows in-progress issues, discovers ready tasks, handles context shifts. One command replaces hunting through task lists. -
Implementation - You write code. Hooks stay out of the way. PreToolUse validates CHANGELOG/README updates before commits. Stop hook updates timestamps.
-
/next complete- Activates when task is done. Auto-generates summary, closes the beads issue, shows what's available next. -
/next continue- Activates before ending session. Saves progress notes, classifies files into HOT/WARM/COLD tiers for efficient resumption. Sets P0 priority so tomorrow's session knows to pick this up first. -
SessionEnd - Activates when session ends. Syncs beads to git remote, cleans up. Your state is persisted, ready for next time.
The hooks run automatically. Session persistence without friction.
| Hook | When | What |
|---|---|---|
| SessionStart | Session begins | Inject beads context, P0 continuations, HOT files, detect installed plugins |
| UserPromptSubmit | Before processing | Workflow reminders, TDD mode if active |
| PreToolUse | Before Bash (git) | CHANGELOG/README validation |
| Stop | Between responses | Update last_active timestamp |
| PreCompact | Before compaction | Preserve critical context |
| SessionEnd | Session ends | Sync beads, cleanup |
| Skill | Purpose |
|---|---|
/next |
Unified work discovery. Find in-progress work, pick ready tasks, shift context, complete with auto-summary. /next proceed launches feature-dev. Classifies files into HOT/WARM/COLD tiers (~60-70% token savings on resume). |
/bd-sync |
OpenSpec ↔ beads synchronization. audit (gap reports), import (create epics), update (bidirectional sync), specgap (find untracked work). |
/commit |
Conventional Commits with auto-detection. Detects change type, breaking changes, creates granular commits. |
/tdd |
TDD workflow reference. Separate RED/GREEN/REFACTOR commits that prove the methodology was followed. |
| Command | Purpose |
|---|---|
/project-init |
Initialize project with CLAUDE.md, beads, workflow selection |
/changelog |
Update CHANGELOG.md following Keep a Changelog |
/audit-claude-md |
Restructure CLAUDE.md for clarity |
If you used the old commands, here's what changed:
| Old Command | New Equivalent |
|---|---|
/prime-context |
/next (auto-shows status on first run) |
/update-context |
/next shift (saves notes to beads) |
/focus |
bd list --status=in_progress |
focus.json |
Beads comments (via /next continue) |
The old commands still exist but just redirect to /next.
/next continue classifies files for efficient session resumption:
| Tier | What | Behavior |
|---|---|---|
| HOT | Modified files, their tests, type definitions | Auto-read first next session |
| WARM | Related patterns, context files | Reference on demand |
| COLD | Already analyzed (understanding in notes) | Skip entirely |
Result: ~60-70% token reduction when resuming sessions.
Core auto-detects installed plugins and adapts:
| Plugin | Detection | Behavior |
|---|---|---|
| beads | .beads/ directory |
Show in-progress issues, sync on session end |
| OpenSpec | openspec/ directory |
Include OpenSpec tasks in /next discovery |
| feature-dev | installed_plugins.json |
Inject TDD enforcement context |
| superpowers | installed_plugins.json |
Compatible, no conflicts |
- Minimal - Just session management, no opinions on methodology
- Compatible - Works with any workflow plugin without conflicts
- Persistent - Uses beads comments for state that survives memory compaction
- Efficient - File tiers and self-contained descriptions save tokens
- Foundation - The layer that everything else builds on
MindContext is evolving toward a CLI-first, agent-agnostic platform. The future isn't "a Claude Code plugin"—it's an orchestration layer that works with any coding agent.
- Session hooks with memory compaction preservation
- Unified
/nextskill for work discovery /bd-syncfor OpenSpec ↔ beads synchronization with bite-sized task decomposition- Tiered context classification (~70% token savings)
- TDD enforcement with separate RED/GREEN/REFACTOR commits
- Multiple in-progress issue detection with workflow guidance
- Remove deprecated commands - Clean up 400+ lines of dead code
- Unify /next - Single context-aware command replacing prime/update/focus
- Workflow discipline - PreToolUse hooks to enforce beads issue claiming before implementation
- CLI-first architecture -
mctxCLI that bridges openspec ↔ beads ↔ feature-dev - Agent portability - Works with Claude Code, Cursor, Windsurf, Aider, any CLI tool
- Implementation planning - Full planning workflow with detailed guides for hand-coding
- Parent-submodule awareness - Unified context across monorepos and worktrees
- MCP bridges - Multi-LLM routing via standardized adapters
- Unified audit - Single
/auditdispatcher for project, code, and CLAUDE.md checks - Release automation - Streamlined version bumping for plugin workflows
Track progress at GitHub Issues.
MindContext Core is designed as the foundation layer. Combine with:
| Plugin | What it adds |
|---|---|
| beads | Git-backed issue tracking with dependencies |
| OpenSpec | Spec-driven development with change proposals |
| superpowers | TDD enforcement, micro-task planning |
| feature-dev | Codebase exploration, architecture guidance |
Recommended stack: Core + beads + OpenSpec + feature-dev
Changes go through OpenSpec proposals:
- Fork the repository
- Create
openspec/changes/your-feature/with spec files - Follow the OpenSpec workflow
- Submit a PR
See openspec/AGENTS.md for the complete guide.
/plugin update mindcontext-coreMIT License - see LICENSE file for details
- GitHub
- beads - Git-backed issue tracking
- OpenSpec - Spec-driven development
- feature-dev - Codebase exploration (Anthropic)
- superpowers - TDD enforcement & micro-tasks