ruvnet/metaharness

Compare Harnesses diffs two scaffolded harnesses and reports manifest, host, and file fingerprint changes. Use it to spot drift between a fork and a baseline, or to emit ADR-031 bundle JSON for CI.

Create Harness scaffolds a focused AI agent harness with its own CLI, host config, memory namespace, and package metadata. Use it to build a custom assistant for Claude Code, Codex, pi.dev, or Hermes.

Darwin Mode evolves a frozen-model harness by mutating one surface file per generation, sandboxing each child, and keeping only variants that measurably improve. Use it to run controlled self-improvement experiments on agent harness code.

Diag Harness checks whether a scaffolded harness matches the installed `@metaharness/kernel` version and reports the skew verdict. Use it before other harness commands or when compatibility errors appear.

Example Harness scaffolds a ready-made AI agent harness from published @metaharness example packages. Use it to create host-specific or vertical workflow setups in one command.

Harness Secrets checks GCP Secret Manager setup, fetches secret values, and validates an NPM token with npm whoami. Use it to rotate publish-time secrets without long-lived CI keys.

List Templates shows the harness templates that create-agent-harness can scaffold and what each includes. Use it when you want to compare available verticals or see what a scaffold will generate.

OIA Manifest emits and validates .harness/oia-manifest.json for an OIA v0.1 harness. Use it to record layer alignment, horizontal spans, and adjacent standards, or to check drift with --check.

Plan Change turns a feature request into a minimal implementation plan before code is written. Use it to restate the goal, identify files, define the smallest interface, and spot wider ripples.

Publish Harness runs build, test, signing, verification, and npm publish for a generated harness. Use it when releasing a tagged harness with provenance and witness manifest checks.

Repo Genome checks a local repository for harness readiness and reports repo type, agent topology, MCP risk, test confidence, release readiness, and a recommended harness plan. Use it before scaffolding, in CI, or when you need a deterministic readiness snapshot.

Score Harness assigns scaffolded harnesses a 0-100 grade across repo understanding, agent usefulness, MCP safety, test coverage, and publish readiness. It also emits a 6-field badge block and exit codes for A/B, C, and F results.

Threat Model renders MCP scan findings as a PR-ready security artifact with allowed and denied tools, permission counts, secrets reachability, and network or shell exposure. Use it when reviewing harness changes or permissions before release.

Upgrade Harness re-renders a scaffolded harness against the latest template, compares added, removed, and changed files, and applies safe updates. Use it to catch template drift and resolve conflicts with markers or .rej files.

Validate Harness runs release-readiness checks for a scaffolded harness in one command. Use it to verify file shape, signatures, path safety, MCP config, and GCP secrets before release.

Verify Witness checks a scaffolded harness witness manifest by validating its Ed25519 signature and reporting valid or invalid. Use it for fast CI checks or federation handshakes before a full harness validation.