next-task
Orchestrate certainty-graded AI agent pipelines for code tasks: discover via GitHub issues or taskmaster-ai, plan changes, implement with TDD and auto-fixes, validate quality and remove AI slop in JS/TS/Python/Rust/Docker, detect doc-code drift, route to optimal models, review via HIGH/MED/LOW gates, and ship resumable GitHub PRs.
npx claudepluginhub elasticdotventures/_b00t_ --plugin next-taskComponent Overview
Component Details
Skills (6)
Apply HIGH/MEDIUM/LOW certainty grading to all agent findings and recommendations. Use to gate human review, auto-fix, or autonomous action.
Validate code quality using certainty-graded rules. Detect AI artifacts, anti-patterns, and b00t violations. Reports auto-fixable vs review-required findings.
Remove AI-generated artifacts from code. Three-phase certainty-graded cleanup. Use after any AI implementation session or before PR creation.
Detect documentation-code drift. Deterministic collection (grep/AST) feeds a single LLM semantic analysis call. Reports mismatches with certainty grades.
Select the right model for the task. Maps task cognitive tier to optimal model. Reads _b00t_ datums for available models. Prefer local/cheap for deterministic work; frontier for reasoning.
End-to-end task pipeline. Discovers the next task, plans, implements, reviews with certainty-graded quality gates, then ships. Resumable via git state.
README
🥾 b00t — Agentic Hive OS
"Tell me what I'm running on, what tools are available, what I'm allowed to do, what goals I should optimize for, and where the boundaries are."
b00t is a context-aware agentic operating layer: tool discovery, version management, tribal knowledge, multi-agent coordination, and MCP integration — batteries included.
⚡ Install
curl -fsSL https://raw.githubusercontent.com/elasticdotventures/_b00t_/main/install.sh | bash
Downloads the binary from GitHub Releases and verifies its SHA256 checksum when possible. Supports Linux x86_64/aarch64/armv7 and macOS Intel/Apple Silicon.
From crates.io:
cargo install b00t-cli
From source:
git clone https://github.com/elasticdotventures/_b00t_.git
cd _b00t_ && cargo install --path b00t-cli
🚀 Quick Start
b00t --version
b00t status # check tool versions vs desired
b00t cli check rust # is rustc installed?
b00t cli install uv # install/upgrade to desired version
b00t cli up # check all datums
b00t cli up --yes # update all tools to desired versions
b00t learn rust # load Rust dev context into agent session
b00t learn docker # container orchestration knowledge
b00t lfmf rust "PyO3: unset CONDA_PREFIX before cargo build to fix linker errors"
b00t advice rust "PyO3" # retrieve lessons for a tool+pattern
🤖 MCP Integration
Claude Code Marketplace (recommended)
# In Claude Code, add the b00t marketplace plugin
/plugin marketplace add elasticdotventures/_b00t_
/plugin install b00t@b00t-plugins
/plugin install skill-document-understanding@b00t-plugins
Provides /b00t skill, context-aware tool dispatch, and all available b00t skills.
Bundles publish deterministic MCP recipes at .claude-plugin/recipes/{skills,roles}/*.json
(for example: skill-document-understanding provides docling-mcp + fetch-url-as-markdown).
Direct MCP Server
b00t mcp install b00t claudecode # Claude Code
b00t mcp install b00t vscode # VS Code
b00t mcp list # list available MCP servers
50+ MCP tools exposed via b00t-mcp — b00t_up, b00t_status, b00t_learn, b00t_lfmf, b00t_advice, and more.
🐝 Hive Coordination
# Multi-agent mission coordination
b00t acp hive create mission-id 3 "Build and deploy microservice" leader
b00t acp hive join mission-id developer
b00t acp hive sync mission-id 1 # barrier: wait for all agents at step 1
b00t acp hive ready mission-id 2 # signal readiness for step 2
🔁 Ralph — OODA Loop Runner
b00t.sh implements the Ralph autonomous task loop (Observe→Orient→Decide→Act):
# Run with default claude tool, 10 iterations
bash b00t.sh
# Configure via env or flags
TOOL=mistralrs bash b00t.sh --max-iterations 5 --role executor
bash b00t.sh --tool codex --sleep 1
Tools: claude, codex, amp, opencode, mistralrs (local vLLM).
Loop exits on EXIT_SIGNAL=true in LLM output, or exit 75 (tempfail/restart) after max iterations.
📦 Datums
Datums in _b00t_/ declare tools with detect/install/desires/hint. b00t resolves DAG-ordered installs:
b00t cli detect fastmcp # run datum detect script
b00t cli desires rust # show target version
b00t cli install fastmcp # install: python → uv → fastmcp (DAG-aware)
🧠 Session & Budget Management
b00t session init --budget 25.00 --time-limit 120 --agent "code-reviewer"
b00t session status
b00t checkpoint "completed feature X"
🌐 Platform Support
| Platform | Architecture | Status |
|---|---|---|
| Linux | x86_64 | ✅ |
| Linux | aarch64 | ✅ |
| Linux | armv7 | ✅ |
| macOS | Intel | ✅ |
| macOS | Apple Silicon | ✅ |
🛠 Development
git clone https://github.com/elasticdotventures/_b00t_.git && cd _b00t_
cargo build
cargo test --workspace
just -l # available recipes
just release # dispatch GitHub release workflow
Release pipeline: release.yml (tag + GitHub Release) → build-release.yml (cross-platform binaries) → publish-crates.yml (crates.io).
📖 Docs
- b00t Gospel — philosophy and architecture
- Agent Guide — agent operation protocol
- CLAUDE.md — LLM alignment instructions
- Release Notes — changelog
Issues / hive recruitment: github.com/elasticdotventures/b00t/issues
Similar Plugins
b00t
b00t plugin for Claude Code - extreme programming agent framework with datum system, direnv pattern, and DRY philosophy
Help us improve
Share bugs, ideas, or general feedback.