OpenClaw-inspired enhancements for Claude Code -- better multi-agent coordination, session continuity, task management, and memory architecture.
OpenClaw is an open-source personal AI assistant with sophisticated multi-agent systems, persistent memory, and proactive behavior. Claude-Claw distills its best patterns into a Claude Code plugin that makes Claude behave more like a reliable personal manager.
| Skill | What it does |
|---|---|
/handoff |
Session continuity -- saves structured task state (objective, progress, files modified, pending work, blockers) so the next conversation picks up exactly where you left off. |
/manager |
Personal task manager -- persistent task board with status, add, plan, review, and context modes. Tracks work across sessions with daily logs and cross-project awareness. |
| Principle Set | What it covers |
|---|---|
| Multi-Agent Safety | 6 rules for safe concurrent agent work: scoped file ownership, no stash/branch switching, commit scope discipline, conflict avoidance. |
| Memory Architecture | 6 principles codifying a three-tier memory system (global / project-local / daily logs) with cross-project search and pre-compaction flush. |
Always-on rules that activate when relevant, invisible otherwise:
- Multi-agent git safety -- prevents agents from stepping on each other's work
- Context preservation -- proactively saves critical state before context compression
- Pre-compaction memory flush -- writes structured summaries before information is lost
- Daily log pattern -- maintains timestamped daily logs for day-to-day continuity
- Cross-project memory search -- reads across project memories when questions span projects
claude plugin install --url https://github.com/andrehuang/claude-claw/handoff
Writes a structured handoff file with your current objective, progress, files modified, pending work, key decisions, and resume instructions. Next session, say "pick up where I left off" and Claude reads the handoff.
/manager add Refactor auth middleware to use JWT
/manager status
/manager plan Migrate database from PostgreSQL to SQLite
/manager review
/manager context
- status -- Show current task board by priority
- add -- Add a task with auto-categorization
- plan -- Break a goal into concrete steps with dependencies
- review -- Summarize accomplishments, update statuses, suggest next priorities
- context -- "Where am I?" briefing at the start of a session
During complex work, Claude appends notable events to memory/daily/YYYY-MM-DD.md:
# Daily Log -- 2026-03-28
- 10:30 -- Started refactoring auth middleware; key files: src/auth/middleware.ts
- 11:15 -- Decision: use JWT instead of session cookies (rationale: stateless scaling)
- 14:00 -- PR #42 merged. Handoff saved for remaining auth-v2 work.At session start, Claude reads today's and yesterday's logs for continuity.
Claude-Claw codifies a three-tier memory system:
| Tier | Scope | What goes here |
|---|---|---|
| Global | All sessions | User identity, routines, cross-project knowledge |
| Project-local | One project | Task state, handoffs, evaluations, project decisions |
| Daily log | One day, one project | Ephemeral events: decisions, files touched, problems encountered |
- Research Companion -- Strategic research thinking (idea evaluation, project triage, brainstorming)
- Academic Writing Agents -- 12 agents for reviewing, auditing, drafting, and polishing academic papers
This plugin distills patterns from OpenClaw, an open-source personal AI assistant with:
- Subagent registry with push-based completion, crash recovery, and depth/concurrency limits
- Session persistence surviving process restarts with atomic writes and file locks
- Multi-part compaction preserving task state, identifiers, and TODOs across context compression
- Heartbeat system for proactive check-ins with active-hours awareness
- Cron service with isolated sessions per job and thundering-herd prevention
Claude-Claw adapts these ideas for Claude Code's architecture -- no gateway needed, just better behavior through skills, principles, and rules.
MIT