What is harness-mods?
harness-mods is a Claude Code mod by ucsandman that harness Mod layer: heartbeat/canary, native usage, rewrite explanations, measured subagent accounting, supervisor routing, secret redaction. Modes per guard in. It is listed under Security and has 22 GitHub stars.
| Type | Claude code mod |
|---|---|
| Author | ucsandman |
| Repository | ucsandman/Agnostic-AI |
| Popularity | 22 GitHub stars |
| Category | Security |
| Security | Review before installing (64/100) |
| Rank | #301 of 5,272 Claude Code mods; #51 in Security |
| License | MIT |
| Last commit | 2026-10-02 |
| Hooks | 7 events |
| Sees | every prompt, every tool call, subagent spawns |
| Draws on | nothing |
Is harness-mods safe to install?
harness-mods scores 64/100 on the MCPMods security check: Review before installing. Things to review: writes files or config; reads environment variables; watches every prompt or event. It passed 5 of 8 checks.
Mod footprints come from claude plugin validate, which lists every event a mod hooks and every API call it makes without running it.
| Check | Result |
|---|---|
| No network access | Pass |
| Starts no processes | Pass |
| Writes files or config | Review |
| Reads environment variables | Review |
| Watches every prompt or event | Review |
| Doesn't drive Claude | Pass |
| Passes claude plugin validate | Pass |
| Has an open-source license | Pass |
API calls in its code
| Call | What it means |
|---|---|
$.command.register | |
$.env.get | reads env vars |
$.fs.exists | reads files |
$.fs.read | reads files |
$.fs.stat | reads files |
$.fs.write | writes files |
$.runtime.supports | |
$.session.id | |
$.session.model | |
$.session.usage | |
$.store.get | keeps its own state |
$.store.set | keeps its own state |
$.ui.log | draws interface |
$.ui.toast | draws interface |
How do I install harness-mods?
- Check your Claude Code version. Mods need Claude Code 2.1.287 or later and are on by default.
claude --version
- Read its footprint. This lists every event it hooks and every API call it makes, without running it.
git clone https://github.com/ucsandman/Agnostic-AI claude plugin validate ./Agnostic-AI/engine/mods/harness-mods
- Install it. Run this in your terminal, then /reload-plugins in an open session.
git clone https://github.com/ucsandman/Agnostic-AI && claude --plugin-dir ./Agnostic-AI/engine/mods/harness-mods
Or ask Claude to set it up
Paste this into Claude Code and it installs harness-mods for you:
Install the "harness-mods" Claude Code mod from https://github.com/ucsandman/Agnostic-AI/tree/master/engine/mods/harness-mods. Run claude plugin validate on it first and show me what it can access before you enable it.
More from ucsandman/Agnostic-AI
This repository has at least 9 Claude Code mods on MCPMods.
- claude-runtime: Shared Claude runtime adapter: normalises function-hook events into ClaudeRuntimeEvent, exposes $.runtime (emit bus, sup (High risk)
- bender: REALITY BENDER: intercepts a harmless tool result and alters what Claude receives; prints REAL vs RECEIVED (Low risk)
- blackbox: BLACK BOX RECORDER: structured per-session timeline of prompts, model steps, tool calls, subagents and denials; /blackbo (Review before installing)
- guardian: GUARDIAN: executable policy that protects lab/demos/protected/** from Write/Edit/Bash, whatever CLAUDE.md says (Low risk)
- matrix: MATRIX: freezes a harmless tool call before it runs, shows what Claude intended, asks you, then continues or denies. /ma (Low risk)
- tunnel-cost: MIDDLEWARE TUNNEL stage 2: Cost analyzer: estimates the call's token cost from the args size (Low risk)
- tunnel-observer: MIDDLEWARE TUNNEL stage 1: Observer: sees the call first, stamps t0, prints what enters (Low risk)
- tunnel-policy: MIDDLEWARE TUNNEL stage 3: Policy: allows everything harmless; denies a matching rule (Low risk)