Free open-source tool

Your AI harness has a score.

agentstack-init audit checks your Claude Code or Hermes config for broken MCP wiring, missing CLAUDE.md, and unreachable servers — then gives you a score out of 100.

Discuss your workflow View on GitHub

illustrative output · not a security assessment

$ pip install agentstack-init
$ agentstack-init audit

✗  CLAUDE.md not found at project root
   Fix: Run agentstack-init init --harness claude_code
✗  No MCP config at ~/.claude/claude_mcp_settings.json
   Fix: Run agentstack-init init --harness claude_code

Score: 60/100
Full report: .fusional/audit-report.json

→ Discuss next steps with FusionAL

What the audit checks.

The CLI checks the listed configuration conditions and reports what it can inspect locally. Its score is a configuration aid, not a security assessment or a guarantee that a harness will work.

CLAUDE.md present

Your harness needs a project-level CLAUDE.md to give the model context.

MCP config wired

Are your MCP servers registered in the right config file for your harness?

Servers reachable

A config that points at a dead port is worse than no config.

Ask about your harness or workflow.

Run the audit yourself and tell us what you need help with. We confirm fit before proposing paid work. Configuration changes depend on access and a separately agreed scope.

Send only non-sensitive context. Do not send credentials or confidential records. A request is not a booked engagement.

Tool first, service only if it fits.

The CLI is open source. Earlier paid harness tiers are historical, not current fixed-price enrollment. Individually agreed legacy terms remain unchanged. New assistance needs a named problem, agreed scope, prerequisites, acceptance checks, handoff, and a separate quote.

Discuss your workflow