metaharness weekly downloads statistics

metaharness

Scaffold your own focused AI agent harness — like ruflo, uniquely yours.

Proven at the cost-Pareto frontier. Same "evolve the harness" thesis, measured: the Darwin harness resolves real SWE-bench Lite issues at 34.0% single-trajectory (~$0.005/inst) and 39.7% Best-of-3 (~$0.015/inst) — conformant, official harness. Live Cost–Performance leaderboard · @metaharness/darwin.

Published as metaharness (the metaharness and harness CLIs). Earlier versions were published as create-agent-harness.

Quick start

You'll be prompted for template, host, description. Out comes a complete npm package ready to npm publish.

Non-interactive

Add the forthcoming field-memory package only when the harness needs shared, history-dependent routing:

--field-memory is opt-in and currently depends on the upstream @metaharness/field-memory@^0.1.0 package PR. It adds src/field-memory.ts, the runtime dependency, and a machine-readable manifest.field_memory block. The generated bootstrap uses packed storage, requires three distinct verifier-derived principals, disables hysteresis, enables a bounded drift window, and fails closed until the deployment supplies an absolute storage path, storage adapter, and principal verifier backed by deployment authentication. It also requires a deployment-secret identityHashKey of at least 32 bytes, reused after state restore. No database path, identity, secret, credential, or principal proof is generated. For RuVector-backed storage, the package adapter requires a configuration-verified FlatIndex with in-place mutation and cosine distance; HNSW and unverified legacy stores are rejected.

Principal independence is only as strong as the verifier's identity and trust domain controls. Also inspect storage.writerScope: process requires one writer service per field, while multi-process or fleet-wide writers require a distributed adapter.

The package influence caps are centroid-local. Fleet-wide admission, revocation, and rate limits belong in the deployment-owned verifier. minimumSupport is routing quarantine, not confidentiality: authorized state exports and registry storage can include singleton aggregate embeddings and rewards, so protect both as sensitive deployment data.

The draft accepts only the exact runtime dependency range ^0.1.0. A missing or malformed generated package.json, another dependency range or section, or an existing src/field-memory.ts causes scaffolding to stop before any write. harness upgrade validates every manifest field against the exact current contract, then reapplies the same manifest-selected overlay so it cannot orphan the module or silently remove its dependency and documentation. The block carries contract_schema: 1; unsupported schemas fail closed and require an explicit migration instead of being reinterpreted.

Use --no-field-memory to force the default off state in shared scripts.

Templates

| Template | Best for |

| minimal | Custom starter — kernel only | | vertical:devops | Incident response, on-call workflows | | vertical:support | Customer support, KB-RAG, escalation | | vertical:trading | Quant trading (paper-default, circuit breakers) | | vertical:legal | Contract review with citation checking | | vertical:research | Multi-source dossier with evidence grading |

Hosts

--host selects which host adapter ships with your harness:

| Host | What you get |

| claude-code | .claude/settings.json with MCP + hooks | | codex | ~/.codex/config.toml with [mcp_servers.*] | | pi-dev | Pi extension (TypeScript, no MCP by design) | | hermes | cli-config.yaml + optional-mcps/*.yaml | | openclaw | ~/.openclaw/openclaw.json + workspace SKILL.md + install runbook | | rvm | RVM partition manifest + capability table + wasm-guest + install runbook (hardware-isolated) | | prime-agent | .prime/agent/skills/ Python-backed skill per tool (no MCP by design) + install-prime-agent.md runbook; deny-lists emit SANDBOX-REQUIRED.md (ADR-247) |

Multi-host: pass --host multiple times.

Also ships the harness CLI

harness score vs metaharness score — two different scorecards (#15)

The two CLIs both accept score but emit different, purpose-specific JSON — so check the schema discriminator before parsing:

| Command | Purpose | JSON discriminator |

| harness score --json | Runtime-readiness badges (score + mcpRisk / releaseReady / testsDetected / sbom / witnessSigned) | "schema": "harness-quickcheck-v1" (string) | | metaharness score --json | 5-dimension harness-fit scorecard (harnessFit / compileConfidence / …) | "schema": 1 (number) |

They are not interchangeable. A consumer wiring one into a pipeline expecting the other should branch on schema (typeof out.schema === 'string' ⇒ harness badges; === 1 ⇒ metaharness scorecard) and refuse the wrong shape rather than silently defaulting missing fields to 0.

Optional Cognitum Meta-Proxy sidecar

Meta-Proxy is an optional local Rust sidecar, not an npm dependency and not part of normal harness scaffolding. Install it only when you want a locally bound, Claude-compatible routing endpoint with Meta-Proxy's own Cognitum OAuth flow:

proxy run is the activation path: it supplies ANTHROPIC_BASE_URL and the local proxy bearer token only to the child process. It does not modify global Claude settings, project files, or persisted client configuration. With the Meta-Proxy automatic-failover release installed, ordinary requests use the user's Claude Passthrough identity; trusted rate-limit telemetry can then select consented Cloud or Sponsored capacity per request. Cognitum login is needed only for Cloud routing, not for installation or normal Passthrough.…

Description from the npm listing.