- Bound receipt canonicalization depth and reject non-finite signed numbers. Evaluator failures stop before invalid scores can enter signed evidence. Finite receipt bytes remain compatible with…
- Clear all positive proof classifications when evidence cannot be digested.
- Emit default-deny MCP policy for Claude Code and native Codex, Hermes, OpenClaw, Copilot, and OpenCode registrations. Preserve generated overlays and user policy edits during upgrades. Policy…
- Reject ambiguous command-like project names while preserving explicit new doctor and overwrite protection.
- Validate SDK names and duplicate registrations; omitted harness descriptions remain supported.
- Reject invalid SWE-bench cost limits in all five solver entrypoints.
Metaharness by rUv
🛠️ The meta-harness for AI agents — scaffold your own focused, branded agent harness with its own npx CLI, MCP server, memory, learning loop, and witness-signed releases. Works with Claude Code, Codex, pi.dev, Hermes, OpenClaw, and RVM (hardware-isolated sandbox).
Recently added on Oct 28, 2026. The description, category and pricing were filled in from public metadata and have not been checked by an editor yet.
Price
Free (OSS)
free and open source
Popularity
#36 in Frameworks
Gid Score 28 / 100 · #36 of 49 ranked frameworks and SDKs
Trend
Growing fast
usage up 35.3% over 30 days
Development
Actively updated
latest 0.4.19, 5d ago
Community
735 GitHub stars
59.6K package downloads a week
Usage data: latest readings Oct 29 to Nov 1, 2026, last checked Nov 1, 06:40 UTC. Date of each number
What Metaharness does
It is free and open source (MIT); you pay only for the models it calls.
Key facts
| Maker | rUv |
|---|---|
| First released | June 2026 |
| License | MIT |
| Main language | TypeScript |
| Open issues and PRs | 196 |
Pricing
Open source under the MIT licence; check the website for any paid hosted plans.
Checked Oct 28, 2026. Prices change often; confirm before you buy.
Quick answers
- Free to start?Yes
- Open source?Yes (MIT)
- Closest alternativeLangChain · all alternatives
What's new in Metaharness: October 2026
October brought refinements to Metaharness with two major API releases for contrastive testing and a maintenance update addressing validation, policy, and toolchain improvements.
- New opt-in contrastive API with invariant edits, measured composition, deduplicated probes, and admission budgets.
- Independent confirmation path for contrastive testing on unseen task pairs before adoption.
- Security improvements including default-deny MCP policy emission and validation of receipt canonicalization.
- Enhanced validation with stricter checks on command-like project names, SDK names, and cost limits.
3 releases this month · summarised from the official release notes · all months
Get started with Metaharness
| npm | npm install metaharness |
|---|
These commands use the packages tracked on this page. The vendor may recommend a different installer.
Add-ons for Metaharness: save tokens, add memory and tools
LiteLLM
One gateway for 100+ models with spend tracking, budgets, rate limits and fallbacks.
Langfuse
Traces every model and tool call with its cost and latency, and runs evaluations.
Firecrawl
Turns websites into clean Markdown or structured data the agent can read.
E2B
Runs agent-written code in isolated cloud sandboxes instead of on your machine.
Headroom
Compresses tool outputs, logs and files before they reach the model, so each run uses fewer tokens.
Composio
Connects the agent to 1,000+ apps and APIs with managed sign-in.
Listed because each tool documents support for Metaharness or for agent frameworks in general. Savings and results depend on your workload.
Metaharness alternatives
Best Metaharness alternatives compared · All agent frameworks and SDKs
| # | Agent | Gid Score | Downloads 7d | 7d | 30d | Stars | Latest release | Price | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 |
LangChainLangChain |
78 | 41.5M | up 4.3% | down 4.3% | 149K+68/day | 1.4.711d ago | Free (OSS) | |||
| 2 |
LangGraphLangChain |
72 | 18.8M | up 24.7% | up 18.5% | 44.2K+61/day | 1.2.192d ago | Free (OSS) | |||
| 3 |
AI SDKVercel |
67 | 41.9M | up 21.1% | up 30.9% | 27.7K+24/day | 7.0.1592d ago | Free (OSS) | |||
| 4 |
Claude Agent SDKAnthropic |
64 | 25.2M | up 20.2% | up 12.4% | 8,396+8/day | 0.2.176yesterday | Free (OSS) | |||
| 5 | 63 | 9M | up 0.4% | up 2.6% | 9,015+11/day | 0.3.32d ago | Free (OSS) | ||||
| 6 |
OpenAI Agents SDKOpenAI |
60 | 6.8M | up 10.0% | up 23.6% | 30.4K+25/day | 0.24.017d ago | Free (OSS) | |||
| No agents match that filter. | |||||||||||
Ratings from people who use Metaharness
No ratings yet. Used Metaharness? Rate it and say what it is good and bad at.
Metaharness news
-
9 agents added to AgentGid
PI Mcp Extension, Hermes Paperclip Adapter, Metaharness, Coding Agent Adapters, Smithers, Steward, Learnharness, Sdk, Plugin Moltbook
-
Metaharness 0.4.19 released
Bound receipt canonicalization depth and reject non-finite signed numbers. Evaluator failures stop before invalid scores can enter signed evidence. Finite receipt bytes remain compatible with…; Clear all positive proof classifications when evidence cannot be…
What changed: recent Metaharness releases
5 stable releases in the last 90 days · Monthly summary · Full changelog
- Adds independent confirmation to the opt-in contrastive API. A development improvement now has a separate path for testing on unseen task pairs before a caller considers adoption.
- 1. Frozen candidate, baseline, environment, execution envelope and decision policy. 2. Exact paired statistical comparison, minimum gain and regression gates, with every declared case required. 3.…
- Measured pilot: development improved 23/24 to 24/24, but baseline, contrastive candidate and control all scored 38/40 on confirmation. Both superiority tests rejected with p = 1. Candidate…
- Reported model cost was $0.007642223 across 253 requests. Reservations totaled $0.596 under a $1 client cap. No unresolved requests. The control selected the baseline, so shared confirmation…
- Validation: 830 passed tests, 19 existing skips, including 99 focused confirmation tests. Clean install, TypeScript build, offline adapters, fresh tarball consumer acceptance and production npm…
- The API does not grant deployment authority. Adapters must enforce real execution, cancellation, spend and data isolation. Task ID separation cannot prove semantic novelty. Repeated experiments…
Earlier releases
- Adds an opt-in API for invariant/domain edits, measured composition, deduplicated probes, admission budgets and scoped experimental memory. Existing evolve behavior is unchanged.
- Validation: 731 tests passed, 19 existing skips; TypeScript build; 47 adversarial tests; zero reported production dependency vulnerabilities; clean tarball install and public API execution.
- Synthetic benchmark: baseline 25/100, composed candidate 100/100, sampling mean 71.125/100 across 32 seeds. Equal logical operation budget, not matched model cost. Parser executions reduced 64 to…
- Receipts, source hashes and limitations are in docs/evidence/contrastive-0.11.0. Package entrypoint: @metaharness/darwin/contrastive. No new runtime dependencies or automatic promotion.
- Publication verified: npm latest is 0.11.0; registry integrity matches the tested tarball; public-registry installation and imports passed.
- Repository status: PR #407 remains open pending CI. Follow-up commit d3cf0bb fixes the root workspace lockfile for the existing CLI dependency on Darwin 0.10.x; clean npm ci and Darwin build pass.…
- Security: writeAtomic path traversal and host-adapter injection (#300), .env guard anchoring (#327), MCP detection in settings.json (#298), Grade A no longer masks mcpRisk High (#337),…
- Correctness: promotion-rule lockouts (#307, #320), sequential evidence wired into run/replay (#288), flywheel graph RangeError (#341), host adapters keep HarnessSpec.autonomous (#332),…
- Published: metaharness 0.4.17, @metaharness/darwin 0.10.3, flywheel 0.1.12, weight-eft 0.1.2, and patch releases of nine host adapters.
- Darwin Shield (#41): a defensive zero-day harness, Darwin Mode applied to security response itself (ADR-155…167).
- Self-learning WASM-memory cost-optimal escalator (#43): tiered compute escalation (E2/E3/E4 cores) with peer-reviewed mitigations (ADR-169).
- Non-stationary CFR/CFR+ poker solver (#49): Darwin evolving against a genuinely adversarial, shifting-strategy opponent via Chebyshev schedules — a harder generalization test than static benchmarks.
- GEPA Distilled Executor Genome (#64) + Agenticow branch-memory for GEPA lineage (#65): offline evolution of the cheap-executor policy, with full candidate lineage tracked like version-control…
- @metaharness/darwin 0.8.0 (#74): the GEPA engine shipped as a pluggable evaluator (darwin/gepa), with the cand-6 genome landing as the first real promoted candidate (#70).
- metaharness learn (#72): ADR-235 learning command, $0 dry-run by default, cand-6 seed packaged in.
- 20/20 templates ship .claude-plugin/plugin.json — minimal plus 19 verticals (devops, legal, research, support, trading, education, sales, gaming, crm, coding, ai, business, advertising, agentics,…
- Per-vertical categories + tags — vertical:devops lands categories: [agent-harness, metaharness-scaffold, vertical, devops/incident-response] so the Claude Code marketplace can filter by vertical.
- 42 new regression tests lock the propagation in — static (every manifest references the entry) + e2e (scaffold proof, substitution works, no leftover {{}}).
- CI guard — the iter-131 smoke-all-hosts job in published-smoke.yml now validates this across all 8 hosts (claude-code, codex, pi-dev, hermes, openclaw, rvm, copilot, opencode) from the live…
Metaharness popularity and usage data
Metaharness was downloaded 59.6K times from npm in the week ending Oct 29, 2026, up 5.0% on the week before and up 35.3% over the last 30 days compared with the 30 before. The GitHub repository has 735 stars, 57 of them added in the last 30 days. The latest stable release is 0.4.19, published Oct 27, 2026; there were 3 stable releases in the past 30 days. On Terminal-Bench 2.0 its best listed result is 76.4% with Claude Opus 4.6, ranking 10 of 43 agents on that board. It was mentioned in 1 Hacker News posts and comments over the last 30 days.
Download the raw daily series: metaharness.csv
Gid Score breakdown
Adoption40
Community0
Attention0
Momentum68
Each component is scored 0 to 100 from public signals; a dash means no data for it. Methodology.
How fresh these numbers are
| Number | Latest reading | Last checked |
|---|---|---|
| Package downloads | Oct 29, 2026 | Nov 1, 06:30 UTC |
| GitHub stars | Nov 1, 2026 | Nov 1, 06:34 UTC |
| Website rank (Tranco) | Oct 31, 2026 | Nov 1, 06:40 UTC |
| Hacker News mentions | Oct 31, 2026 | Nov 1, 06:35 UTC |
We fetch every source once a day. Registries publish downloads and page views a day or two late, so their latest reading trails the check. A reading more than 4 days old is marked out of date; the Gid Score still uses it until a newer one arrives.
Metaharness benchmark results
| Benchmark | Model | Effort | Tasks solved | Cost of run | Place |
|---|---|---|---|---|---|
| Terminal-Bench 2.0 | Claude Opus 4.6 · v1.1.0 | — | 76.4% ± 2.4 | — | #10 of 43 |
Up to four best runs per board. Scores are not comparable between Terminal-Bench versions. Source: Terminal-Bench (tbench.ai).
Metaharness FAQ
How much does Metaharness cost?
Open source under the MIT licence; check the website for any paid hosted plans.
Is Metaharness open source?
Yes. Metaharness is open source, released under the MIT license.
How popular is Metaharness?
Metaharness was downloaded 59.6K times from npm in the week ending Oct 29, 2026, up 5.0% on the week before and up 35.3% over the last 30 days compared with the 30 before. The GitHub repository has 735 stars, 57 of them added in the last 30 days. The latest stable release is 0.4.19, published Oct 27, 2026; there were 3 stable releases in the past 30 days.
What is the latest version of Metaharness?
The latest stable release we track is 0.4.19, published on Oct 27, 2026.
What are the alternatives to Metaharness?
The closest alternatives in the same category by Gid Score are LangChain, LangGraph, AI SDK, Claude Agent SDK.
For the makers of Metaharness badges, corrections
Add a live badge to your README or site:
[](https://agentgid.com/agents/metaharness/)[](https://agentgid.com/agents/metaharness/)[](https://agentgid.com/agents/metaharness/)Something wrong or outdated on this page? Send a correction with a link to the official source. More for makers.
Where these numbers come from
npm: metaharness. GitHub: ruvnet/metaharness. Site rank: cognitum.one on the Tranco list. See data sources for how each one is collected.
Product facts and pricing were checked against: Cognitum.One, github.com, npmjs.com.