Multiple loops for infinite task generation and execution.
Most agent loops work like this:
Get task → Execute → Verify → Repeat → Stop (backlog empty)
They execute a predefined backlog until it's empty. Infinite Moves is different - it generates its own work:
┌───────────────────────────────────────────────────┐
│ │
│ Ideation ───→ Execution ───→ Debt Sweep │
│ ↑ │ │ │
│ └───────────────┴─────────────┘ │
│ │
│ The loops feed each other. Work never ends. │
└───────────────────────────────────────────────────┘
| Aspect | Typical Agent Loop | Infinite Moves |
|---|---|---|
| Work source | Predefined backlog | Self-generated via ideation |
| Loops | 1 (execution) | 3 (ideation, execution, debt) |
| Ends when | Backlog empty | Never - ideates more work |
| Maintenance | Manual | Self-maintaining via debt sweep |
Ideation scans docs, code, and your ideas to generate tasks. Execution works through them. Debt sweep keeps quality high. When the backlog runs low, ideation finds more. It's perpetual, self-sustaining development.
| Platform | Location | Status |
|---|---|---|
| Claude Code | skills/ + plugins/ | ✓ Full support |
| OpenCode | opencode/ | ✓ Full support |
See platform-specific READMEs for installation instructions.
Two components working together:
| Component | Purpose |
|---|---|
| Skill | Makes Claude aware of the plugin, enables autonomous invocation |
| Plugin | Provides the commands and agents |
User: "Help me with my backlog"
↓
Skill activates (matches "backlog")
↓
Claude knows to use /moves commands
↓
Invokes /moves status, /moves run automatically
↓
Spawns task-executor agents
| Command | Purpose |
|---|---|
/moves run | Execute tasks (single, parallel, continuous) |
/moves ideate | Generate tasks from topics or gap analysis |
/moves status | View pipeline, manage tasks |
/moves sweep | Scan for technical debt |
# Initialize the data directory
mkdir -p .infinite-moves/tasks .infinite-moves/designs
echo '{"tasks": []}' > .infinite-moves/manifest.json
# Let Claude work autonomously
"Generate tasks for user authentication and start working on them"
# Or use commands explicitly
/moves ideate --topic "user authentication"
/moves run --continuous --parallel 2
All artifacts are stored in .infinite-moves/ at the project root:
.infinite-moves/
├── manifest.json # Task manifest
├── tasks/ # Task spec files
│ └── G17-health-sync.md
├── designs/ # Design docs from ideation
│ └── push-notifications.md
├── debt-manifest.json # Technical debt tracking
└── debt-report.md # Human-readable debt report
infinite-moves/
│
├── skills/infinite-moves/ # Claude Code skill
│ └── SKILL.md
│
├── plugins/infinite-moves/ # Claude Code plugin
│ ├── plugin.json
│ ├── commands/
│ │ ├── run.md
│ │ ├── ideate.md
│ │ ├── status.md
│ │ └── sweep.md
│ └── agents/
│ ├── task-executor.md
│ └── verifier.md
│
└── opencode/ # OpenCode support
├── README.md
├── plugins/
│ └── infinite-moves.ts # TypeScript plugin
└── agents/
├── task-executor.md
└── verifier.md
Ideation Loop
Execution Loop
Debt Loop
MIT
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub mastra-ai/infinite-moves --plugin infinite-movesAI workflow and integration skills for Mastra framework, enabling automated agent orchestration and task execution.
Installer for TÂCHES' Get-Shit-Done workflow system - meta-prompting, context engineering and spec-driven development
Plan and autonomously build a software task end-to-end. Recons the codebase, applies preloaded memory, decomposes into the right number of phases, gets one confirmation, then prepares a single ready-to-paste /goal command — one paste between you and done — that drives execution to completion with built-in retry, fix-spec recovery, and per-phase memory writeback. Works on Claude Code and Codex.
Skills and agents for everyday AI-assisted development: structured git commits, safe repository refresh and branch cleanup, day-grouped changelog maintenance, project-local task and todo tracking as plain-markdown files (a base task skill plus create, check, auto-check, explain, select, implement, audit, finish, and fix siblings with auto_*_task helper agents, including task_auto_check's auto_drift_task intent check and task_fix's auto_shaper_task escalation), repo-root guardrail documents that keep autonomous agents anchored to human intent (a guardrail hub skill with per-type templates and rules for CHARTER, ARCHITECTURE, TESTING, and SECURITY docs), writing and formatting AI instructions (prompts, rules, skill files, agent definitions), harness portability for bundled scripts and plugin wiring, and language style guides for markdown, Python, and Rust.
Zero-dependency planning + execution with .flow/ task tracking and Ralph autonomous mode (multi-model review gates). Worker subagent per task for context isolation. Prime delivers an opinionated, structure-aware agent-readiness assessment: it classifies the project, judges substance over existence, and leads with an operability verdict + ranked next-actions. Includes 21 subagents, 24 commands, 28 skills.
Composable workflow skills, agents, and hooks for Claude Code
Ultra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.