KTCrisis

Flux7 Supervisor — AI skill for Claude Code

AI community

Standalone L1 supervisor agent for flux7-mesh — rules engine + pluggable LLM (Ollama, Anthropic, Claude Code MCP callback).

How to install Flux7 Supervisor

This entry records only its repository, not the path inside it, so there is no exact command to give. Open KTCrisis/flux7-supervisor and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Flux7 Supervisor does

Standalone L1 supervisor agent for flux7-mesh — rules engine + pluggable LLM (Ollama, Anthropic, Claude Code MCP callback). Auto-resolves approval gates.

Alternatives in AI

AIHawk — Open-source AI browser agent: describe any task in plain language and it autonomously browses, clicks, types a 30.3k ★
Architecture Diagram Generator — Generate beautiful dark-themed system architecture diagrams as standalone HTML/SVG files 7k ★
Open Multi Agent — TypeScript multi-agent orchestration engine — one runTeam() call from goal to result 5.8k ★

README

flux7-supervisor

Standalone L1 evaluation agent for [flux7-mesh](https://github.com/KTCrisis/flux7-mesh). Sits between the policy engine (L0) and human operators (L2) — polls pending approvals, evaluates them with rules and an LLM, and resolves or escalates.

flux7-mesh (L0: policy engine)
     │
     │  pending approval
     ▼
flux7-supervisor (L1: rules + LLM)
     │
     ├── rule match → auto-approve/deny
     ├── LLM evaluation → approve/deny/escalate
     └── unknown → escalate to human (L2)

Install

pip install flux7-supervisor

# With Anthropic provider
pip install flux7-supervisor[anthropic]

Quick start

# Check connectivity
sup7 -c sup7.yaml status

# Start the supervisor loop
sup7 -c sup7.yaml start

Requires a running `mesh7 serve` instance. Optional: `mem7 serve` for decision persistence.

Run as a service

A systemd unit is provided in [`contrib/systemd/sup7.service`](contrib/systemd/sup7.service) (`After=`/`Wants=mesh7.service`, restart on failure):

sudo cp contrib/systemd/sup7.service /etc/systemd/system/
sudo systemctl daemon-reload && sudo systemctl enable --now sup7

Adapt `User=` and paths first. Pair it with `approval.channel: queue` in the mesh config — a service has no TTY to prompt on.

Configuration

mesh:
  url: http://localhost:9090
  agent_id: supervisor

memory:
  url: http://localhost:9070
  enabled: true
  store_decisions: true

evaluator:
  provider: ollama           # ollama | anthropic | claude-code | jev
  model: qwen3:14b
  url: http://localhost:11434
  timeout: 30
  confidence_threshold: 0.8

poll:
  interval: 2s

rules:
  - name: safe-reads
    condition: "tool contains read"
    action: approve
    confidence: 0.95
  - name: project-writes
    condition: "params.path starts_with project_dir"
    action: approve
    confidence: 0.9
  - name: injection-risk
    condition: "injection_risk == true"
    action: escalate
    confidence: 1.0

project_dirs:
  - /home/