← DIRECTORY
AI TOOL

Fusengine Harness

Open-source governance and policy layer for AI coding-agent harnesses with evidence gates, verification receipts, hooks and compatibility adapters.

8.7OUT OF 10 · LOW CONF.

Product facts

Category
—
Pricing
Free open-source software under the MIT License; no separate paid Fusengine Harness software tier verified.
Free access
YES
Platforms
Windows, Linux, macOS
API
Not confirmed
Open source
YES
Integrations
Claude Code, Codex, Cursor, Cline, Gemini CLI, Hermes, Aider, Windsurf, OpenHands
Last updated
2026-09-09
PRODUCT / PRICING SOURCE ↗

Fusengine Harness review

Updated 2026-09-09

Product & capabilities

Fusengine Harness is an MIT-licensed governance and policy layer for AI coding agents. It packages a reusable policy core with evidence-backed gates, verification receipts, lesson memory, caching, lifecycle tracking and adapter logic rather than providing a model or coding assistant itself. The current repository documents hook adapters for Claude Code, Codex, Cursor, Cline, Gemini CLI and Hermes, plus CLI-mode use for Aider, Windsurf and OpenHands. 1

The package can initialize harness-specific hook wiring, evaluate file and command policies, record agent/reference activity, gate staged changes, coordinate PRD task ownership, and expose reusable subpath modules for policy, runtime, tracking, refs, cache and memory. These are documented project capabilities, not AiToolMap hands-on findings. 1

Pricing & access

Fusengine Harness is free open-source software under the MIT License. No separate paid software tier was verified. The CLI installs from npm and the library can be consumed through Bun; users still need the coding-agent harnesses and any model/API access those products require. 1

The repository includes platform-specific notification handling for macOS, Linux and Windows and documents a single package intended to bridge multiple agent environments. 1

Evidence & trust

The strongest trust signal is not a third-party certification but the repository's unusually specific disclosure of enforcement ceilings. It states that cross-harness behavior is not identical: Cursor file-edit hooks are advisory after the edit, Codex can fail to enforce a correct upstream deny, some richer lifecycle events require extra wiring, Hermes lifecycle coverage is only partly verified, and the CONFIRM mechanism is explicitly not a security boundary. 1

The repository reports 662 tests across 120 files and CI that runs tests plus typechecking. Those are useful engineering signals but remain first-party evidence. The canonical 50-source panel found no exact independent review, benchmark or eligible numeric rating corpus. No AiToolMap hands-on testing was performed. 1

Who it's for

Fusengine Harness best fits developers and platform teams already using coding agents who want reusable governance across projects and agent vendors: policy guards, evidence freshness, receipts, memory, lifecycle hooks and a common operating model. 1

It is less suitable for users expecting a fully sandboxed security boundary, identical enforcement across every supported harness, or independently validated performance and reliability claims.

Strengths & weaknesses

Strengths

MIT license; broad adapter surface; explicit policy and lifecycle architecture; fail-closed custom guard chain; evidence/receipt tooling; documented pinning and stale-runtime protections; extensive first-party test surface; unusually candid compatibility and security caveats. 1

Weaknesses

host-harness limitations directly cap enforcement; some lifecycle coverage requires manual or marketplace wiring; Cursor is advisory for edits; Codex enforcement includes an upstream caveat; Hermes coverage remains partly unverified; no exact independent panel evidence was found. 1

SOURCES

Sources & references

1 source
  1. Official sourceFusengine Harness — GitHub
    OFFICIAL

Related AI tools

Products matched by shared tasks, audience, workflow, product type and capabilities.