Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Bash Loop operating mode for Claude Code - action over explanation, tight feedback loops, intelligent context preservation, and graceful subagent control
npx claudepluginhub theglitchking/aeon-flux --plugin aeon-fluxCreate or clear abort signal to stop all subagents. Use when you need to halt all running operations immediately.
Save current task state for later resumption. Creates a checkpoint in .claude/memory/checkpoint.md
Mark an item for attention preservation. Items marked with /focus survive context compaction.
Trigger self-assessment to evaluate progress and approach. Use when stuck or after extended work sessions.
Load the last checkpoint and resume work from saved state. Reads from .claude/memory/checkpoint.md
Pure bash loop execution agent. Use for tasks requiring direct action without planning or explanation. Operates in tight see-act-observe cycles.
Pattern extraction agent that analyzes errors and corrections to learn project-specific behaviors. Use after repeated errors or at session end.
Task decomposition and delegation agent. Use for complex multi-step tasks that benefit from breaking down into smaller units. Delegates to other agents.
Validation agent for testing changes after implementation. Runs tests, type checks, and verifies code quality. Use after significant code changes.
Activates tight action-feedback operating mode. Use for any task requiring shell commands, file operations, iterative problem-solving, debugging, or when the user wants efficient execution without verbose explanations.
Activates during long sessions or before context compaction. Manages attention markers and ensures critical information survives memory compression.
Activates when errors occur during execution. Provides self-correction patterns for immediate error fixing without verbose explanations or apologies.
Activates after code changes to suggest or run verification steps. Provides test-after-action patterns for different file types and frameworks.
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Interactive self-referential AI loops for iterative development. Claude works on the same task repeatedly, seeing its previous work, until completion.
連續自引用 AI 循環,用於互動式迭代開發。實現 Ralph Wiggum 技術,讓 Claude 在 while-true 循環中執行相同提示直到任務完成
Use when user wants to execute long-running tasks that require multiple sessions to complete. This skill manages task decomposition, progress tracking, and autonomous execution using Claude Code headless mode with auto-continuation. Trigger phrases: autonomous, long-running task, multi-session, 自主执行, 长时任务, autonomous skill.
This skill should be used when the model's ROLE_TYPE is orchestrator and needs to delegate tasks to specialist sub-agents. Provides scientific delegation framework ensuring world-building context (WHERE, WHAT, WHY) while preserving agent autonomy in implementation decisions (HOW). Use when planning task delegation, structuring sub-agent prompts, or coordinating multi-agent workflows.
Implementation of the babysitter technique - continuous orchestration loops for deterministic development. Run Claude in a loop with orchestration steps based on the babysitter-sdk and technique.
Open-source, local-first Claude Code plugin for token reduction, context compression, and cost optimization using hybrid RAG retrieval (BM25 + vector search), reranking, AST-aware chunking, and compact context packets.
Generate production-quality SVG technical diagrams (architecture, flowchart, UML, ER, sequence, etc.) with 7 built-in styles. Export as SVG+PNG via @resvg/resvg-js.
Autonomous loop engine with orchestrated subagents, DAG-based task execution, context persistence across iterations, and intelligent failure recovery with circuit breaker
Persistent markdown-based planning with task directories, progress tracking, and context engineering for multi-step Claude Code workflows
Eliminate context rot in Claude Code with intelligent memory hierarchy, auto-learning hooks, and specialized subagents. 8-layer architecture with SessionStart/End hooks and MCP knowledge base.
Polyglot progressive linting for monorepos — adapter-driven baselines, per-app drift detection, and idempotent skips. Supports JavaScript/TypeScript (ESLint, Biome), Python (Ruff), Go (golangci-lint), Rust (Clippy), Terraform/OpenTofu (tflint), and Ansible (ansible-lint).
Uses power tools
Uses power tools
Uses Bash, Write, or Edit tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
A Claude Code plugin that makes Claude faster, smarter, and more reliable by implementing the Bash Loop architecture.
Inspired by Geoffrey Huntley's "Ralphy Loop" - enhanced with solutions for its known limitations.
Aeon Flux changes how Claude Code behaves:
| Before (Default Claude) | After (Aeon Flux) |
|---|---|
| "Let me explain what I'm going to do..." | Just does it |
| "I apologize for the error..." | Fixes it silently |
| "Here's a summary of the output..." | Shows actual output |
| Forgets important context | Preserves critical info |
| Can't reliably stop subagents | /abort command works |
In short: Claude talks less and does more.
claude
Type this inside Claude Code (not in your regular terminal):
/plugin
Then select "Add Marketplace" and enter:
TheGlitchKing/aeon-flux
Run /plugin again, select "Install Plugin", and choose aeon-flux.
/setup
The plugin is now active. Bash Loop mode enabled.
If you want to modify the plugin or install from a local copy:
# Clone the repo
git clone https://github.com/TheGlitchKing/aeon-flux.git
# Go into the folder
cd aeon-flux
# Start Claude Code with the plugin directory
claude --plugin-dir ./
# Or add as a local marketplace inside Claude Code:
# /plugin → Add Marketplace → ./path/to/aeon-flux
The "Bash Loop" (aka "Ralphy Loop") is an agent architecture created by Geoffrey Huntley. The core idea:
See → Act → Observe → Repeat
"The more you can get the model to just shut up and do the thing, the better off you are."
The original Bash Loop has known limitations. Aeon Flux fixes them:
| Limitation | Problem | Aeon Flux Solution |
|---|---|---|
| Context Exhaustion | Loop runs out of memory on long tasks | Checkpointing system + attention preservation |
| No Hierarchical Planning | Flat loops can't handle complex tasks | Orchestrator agent for task decomposition |
| Memory Loss | Knowledge lost between sessions | Persistent memory files + PreCompact hooks |
| Single-Agent Bottleneck | One agent doing everything | 4 specialized agents (executor, verifier, learner, orchestrator) |
| Error Spirals | Repeats same mistakes | Error capture + pattern learning |
| Can't Stop Subagents | Ctrl+C/Escape don't work reliably | Signal file abort system (/abort command) |
| No Self-Assessment | No way to detect when stuck | /reflect command for self-evaluation |
<!-- ATTENTION --> tags - survives context compaction| Command | What It Does |
|---|---|
/checkpoint | Save your current task state |
/resume | Load the last checkpoint and continue |
/reflect | Make Claude assess if it's stuck or making progress |
/focus <item> | Mark something important to remember |
/abort | Stop all running subagents immediately |
/abort clear | Resume normal operation after abort |
/abort status | Check if abort is active |
Aeon Flux includes 4 specialized agents:
| Agent | Purpose | When to Use |
|---|---|---|
| executor | Pure action, no explanation | Any bash/file task |
| verifier | Runs tests, checks code | After making changes |
| learner | Extracts patterns from errors | After repeated mistakes |
| orchestrator | Breaks big tasks into steps | Complex multi-step work |
Skills activate automatically when relevant:
| Skill | Triggers When |
|---|---|
| bash-loop | Any task needing shell commands |
| error-recovery | An error occurs |
| context-preservation | Long session or before compaction |
| verification | After code edits |
Claude Code has a bug where Ctrl+C and Escape don't reliably stop subagents (Issue #3455).
Aeon Flux works around this with a signal file pattern: