[
  {
    "id": "k8s-sandbox",
    "name": "k8s Sandbox",
    "url": "https://k8s-sandbox.aisi.org.uk/",
    "desc": "Python package that provides a Kubernetes sandbox environment for Inspect.",
    "author": "UK AISI",
    "author_url": "https://github.com/UKGovernmentBEIS/inspect_k8s_sandbox",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "modal-sandbox",
    "name": "Modal Sandbox",
    "url": "https://meridianlabs-ai.github.io/inspect_sandboxes/modal.html",
    "desc": "Serverless container sandbox for Inspect using Modal's cloud infrastructure.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_sandboxes",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "daytona-sandbox",
    "name": "Daytona Sandbox",
    "url": "https://meridianlabs-ai.github.io/inspect_sandboxes/daytona.html",
    "desc": "Sandbox environment for Inspect using Daytona's cloud infrastructure.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_sandboxes",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "ec2-sandbox",
    "name": "EC2 Sandbox",
    "url": "https://github.com/UKGovernmentBEIS/inspect_ec2_sandbox",
    "desc": "Python package that provides a EC2 virtual machine sandbox environment for Inspect.",
    "author": "UK AISI",
    "author_url": "https://github.com/UKGovernmentBEIS/inspect_ec2_sandbox",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "proxmox-sandbox",
    "name": "Proxmox Sandbox",
    "url": "https://github.com/UKGovernmentBEIS/inspect_proxmox_sandbox",
    "desc": "Use virtual machines, running within a Proxmox instance, as Inspect sandboxes.",
    "author": "UK AISI",
    "author_url": "https://github.com/UKGovernmentBEIS/inspect_proxmox_sandbox",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "vagrant-sandbox",
    "name": "Vagrant Sandbox",
    "url": "https://github.com/jasongwartz/inspect_vagrant_sandbox",
    "desc": "Use any virtual machine hypervisor supported by Hashicorp Vagrant as Inspect sandboxes.",
    "author": "Jason Gwartz",
    "author_url": "https://github.com/jasongwartz",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "podman-sandbox",
    "name": "Podman Sandbox",
    "url": "https://github.com/VectorInstitute/inspect-podman",
    "desc": "Podman-backed sandbox environment for Inspect, enabling containerized tool calls without Docker.",
    "author": "Vector Institute",
    "author_url": "https://github.com/VectorInstitute/inspect-podman",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "openshell-sandbox",
    "name": "OpenShell Sandbox",
    "url": "https://github.com/32bitsret/inspect-openshell-sandbox",
    "desc": "NVIDIA OpenShell sandbox environment for Inspect, with kernel-enforced filesystem and network policies per task.",
    "author": "Retnan Daser",
    "author_url": "https://github.com/32bitsret",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "policy-sandbox",
    "name": "Policy Sandbox",
    "url": "https://github.com/Dedulus/inspect-policy-sandbox",
    "desc": "Sandbox wrapper that allows fine grained control over command execution and file I/O.",
    "author": "Arnab Mitra",
    "author_url": "https://github.com/Dedulus",
    "categories": [
      "Sandboxes"
    ]
  },
  {
    "id": "inspect-scout",
    "name": "Inspect Scout",
    "url": "https://meridianlabs-ai.github.io/inspect_scout/",
    "desc": "Transcript analysis for Inspect evaluations.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_scout",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-viz",
    "name": "Inspect Viz",
    "url": "https://meridianlabs-ai.github.io/inspect_viz/",
    "desc": "Interactive data visualization for Inspect evaluations.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_viz",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "docent",
    "name": "Docent",
    "url": "https://docs.transluce.org/",
    "desc": "Tools to summarize, cluster, and search over agent transcripts.",
    "author": "Transluce",
    "author_url": "https://transluce.org/introducing-docent",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "probity-native-result-reader",
    "name": "Probity native result reader",
    "url": "https://github.com/probityai/agent-evidence-observer/blob/71ac0b2126473316655184235000647a6dd0f5cf/docs/NATIVE-CONSUMER-CI.md",
    "desc": "Offline reader and publication gate for bounded native Inspect execution records, preserving errors and unstarted attempts.",
    "author": "Probity",
    "author_url": "https://github.com/probityai",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-wandb",
    "name": "Inspect WandB",
    "url": "https://github.com/DanielPolatajko/inspect_wandb",
    "desc": "Integration with Weights and Biases platform.",
    "author": "Arcadia",
    "author_url": "https://www.arcadiaimpact.org/",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-mlflow",
    "name": "Inspect MLflow",
    "url": "https://github.com/debu-sinha/inspect-mlflow",
    "desc": "Experiment tracking, execution tracing, LLM provider autolog, and artifact logging for Inspect AI evaluations.",
    "author": "Debu Sinha",
    "author_url": "https://github.com/debu-sinha",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-falsify",
    "name": "Inspect falsify",
    "url": "https://github.com/studio-11-co/falsify-inspect",
    "desc": "Pre-register an evaluation's success criteria (metric, threshold, dataset hash, seed) as a tamper-evident PRML manifest before the run, then verify eval logs against the locked claim via installed hooks.",
    "author": "C\u00fcneyt \u00d6zt\u00fcrk",
    "author_url": "https://falsify.dev",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "cje",
    "name": "CJE",
    "url": "https://github.com/cimo-labs/cje",
    "desc": "Calibrated judge evaluation \u2014 calibrate model-graded scorer accuracy using causal inference with optional oracle labels.",
    "author": "CIMO Labs",
    "author_url": "https://cimolabs.com",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-claim-support",
    "name": "Inspect Claim Support",
    "url": "https://github.com/avalyset/inspect-claim-support",
    "desc": "Faithfulness / claim-support scorer \u2014 checks whether a claimed answer is substantiated by the transcript. Absence of evidence is not treated as support.",
    "author": "Eirik Botten Nicolaysen",
    "author_url": "https://github.com/avalyset",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-judge-drift",
    "name": "Inspect Judge Drift",
    "url": "https://github.com/avalyset/inspect-judge-drift",
    "desc": "Judge-drift measurement \u2014 re-grades a stored eval log with two graders over the same samples, so the judge model is the only variable. Grade-parse failures are reported as unscored, never fabricated into verdicts.",
    "author": "Eirik Botten Nicolaysen",
    "author_url": "https://github.com/avalyset",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "grader-variance",
    "name": "Grader Variance",
    "url": "https://github.com/joanna-ciesielski/grader-variance",
    "desc": "Grader-variance decomposition \u2014 repeatedly re-scores hash-verified frozen completions with one judge, separating grader noise from question and model-sampling variance and reporting the repeats a target dependability needs.",
    "author": "Joanna Ciesielski",
    "author_url": "https://github.com/joanna-ciesielski",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-refusal-fidelity",
    "name": "Inspect Refusal Fidelity",
    "url": "https://github.com/IvanSFlowGit/inspect-refusal-fidelity",
    "desc": "Scores whether a system refuses when it cannot honestly answer, reporting over-refusal and false confidence as separate metrics.",
    "author": "Ivan Skachek",
    "author_url": "https://github.com/IvanSFlowGit",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "hermes-rubric",
    "name": "Hermes Rubric",
    "url": "https://github.com/hermes-labs-ai/hermes-rubric/blob/v1.2.1/docs/ADAPTERS.md#inspect-ai-scorer",
    "desc": "Rubric-based scorer that preserves assessment details in Inspect score metadata and can report assessment failures as unscored.",
    "author": "Hermes Labs",
    "author_url": "https://hermes-labs.ai",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "proofbundle",
    "name": "proofbundle",
    "url": "https://github.com/b7n0de/proofbundle",
    "desc": "Signed, offline-verifiable receipts for eval runs \u2014 Ed25519 over an RFC 6962 Merkle root. Binds what a run claimed and when; it does not attest that the evaluation itself was honest.",
    "author": "b7n0de",
    "author_url": "https://b7n0de.com",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "corrlog-for-inspect",
    "name": "Corrlog for Inspect",
    "url": "https://github.com/SaInT8888888888/corrlog",
    "desc": "Emits signed Agent Correction Records on INCORRECT eval scores via Inspect hooks \u2014 Ed25519, hash-chained, key-pinned receipts evidencing caught agent failures, verifiable offline.",
    "author": "SaInT8888888888",
    "author_url": "https://github.com/SaInT8888888888",
    "categories": [
      "Analysis"
    ]
  },
  {
    "id": "inspect-swe",
    "name": "Inspect SWE",
    "url": "https://meridianlabs-ai.github.io/inspect_swe/",
    "desc": "Software engineering agents (Claude Code and Codex CLI) for Inspect.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_swe",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "inspect-cyber",
    "name": "Inspect Cyber",
    "url": "https://ukgovernmentbeis.github.io/inspect_cyber/",
    "desc": "Python package that streamlines the process of creating agentic cyber evaluations in Inspect.",
    "author": "UK AISI",
    "author_url": "https://github.com/UKGovernmentBEIS/inspect_cyber",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "inspect-petri",
    "name": "Inspect Petri",
    "url": "https://meridianlabs-ai.github.io/inspect_petri/",
    "desc": "Framework for testing alignment hypotheses end\u2011to\u2011end, including automatic scenario generation.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_petri",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "petri-bloom",
    "name": "Petri Bloom",
    "url": "https://meridianlabs-ai.github.io/petri_bloom/",
    "desc": "Framework for generating multi-turn behavioral evaluations of frontier AI models.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/petri_bloom",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "control-arena",
    "name": "Control Arena",
    "url": "https://control-arena.aisi.org.uk",
    "desc": "Framework for running experiments on AI Control and Monitoring.",
    "author": "UK AISI",
    "author_url": "https://github.com/UKGovernmentBEIS/control-arena",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "linux-arena",
    "name": "Linux Arena",
    "url": "https://www.linuxarena.ai",
    "desc": "Framework for running experiments on AI Control and Monitoring.",
    "author": "Redwood Research",
    "author_url": "https://github.com/linuxarena/control-tower",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "inspect-boltons",
    "name": "Inspect boltons",
    "url": "https://github.com/tadamcz/inspect_boltons",
    "desc": "Reusable tools and utilities for Inspect, in the spirit of the Python boltons library.",
    "author": "Tom Adamczewski",
    "author_url": "https://github.com/tadamcz",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "evalport",
    "name": "EvalPort",
    "url": "https://github.com/adhabnr-ux/evalport",
    "desc": "Open standard for portable LLM evaluation datasets. Convert between Inspect AI and other eval frameworks (DeepEval, Promptfoo, LangSmith, etc.) using a portable JSON format.",
    "author": "Sahi Katukojula",
    "author_url": "https://github.com/adhabnr-ux",
    "categories": [
      "Frameworks"
    ]
  },
  {
    "id": "inspect-flow",
    "name": "Inspect Flow",
    "url": "https://meridianlabs-ai.github.io/inspect_flow/",
    "desc": "Workflow orchestration for reproducibly running evals at scale.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect_flow",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "inspect-hawk",
    "name": "Inspect Hawk",
    "url": "https://hawk.metr.org",
    "desc": "Platform for running Inspect AI evaluations on cloud infrastructure.",
    "author": "METR",
    "author_url": "https://github.com/METR/hawk",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "optstop",
    "name": "Optstop",
    "url": "https://github.com/UKGovernmentBEIS/optstop",
    "desc": "Adaptive optimal stopping algorithms to help efficiently determine when enough data has been collected to make reliable inferences.",
    "author": "UK AISI",
    "author_url": "https://github.com/UKGovernmentBEIS/optstop",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "evaljobs",
    "name": "Evaljobs",
    "url": "https://github.com/dvsrepo/evaljobs",
    "desc": "Run evals on Hugging Face GPUs and share results and code on the Hugging Face Hub.",
    "author": "Hugging Face",
    "author_url": "https://github.com/dvsrepo/evaljobs",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "inspect-vs-code",
    "name": "Inspect VS Code",
    "url": "https://marketplace.visualstudio.com/items?itemName=ukaisi.inspect-ai",
    "desc": "VS Code extension that assists with developing and debugging Inspect evaluations.",
    "author": "Meridian",
    "author_url": "https://github.com/meridianlabs-ai/inspect-vscode",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "inspect-evals-lint",
    "name": "Inspect Evals Lint",
    "url": "https://inspect-evals-lint.generality.org/",
    "desc": "Static checks for Inspect evaluations: file structure, test coverage, best practices and sandbox image pinning.",
    "author": "Generality Labs",
    "author_url": "https://github.com/Generality-Labs/inspect-evals-lint",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "inspect-costs-plugin",
    "name": "Inspect Costs Plugin",
    "url": "https://github.com/jasongwartz/inspect_costs_plugin/",
    "desc": "Automatically load pricing data for models under test.",
    "author": "Jason Gwartz",
    "author_url": "https://github.com/jasongwartz",
    "categories": [
      "Tooling"
    ]
  },
  {
    "id": "inspect-logittilt",
    "name": "Inspect LogitTilt",
    "url": "https://adrskapars.github.io/inspect_logittilt/",
    "desc": "Steers a model's sampling towards any target behaviour, surfacing on-policy examples in fewer samples, accessing only the model's logits.",
    "author": "Adrians Skapars",
    "author_url": "https://scholar.google.com/citations?user=9Vf0mWkAAAAJ&hl=en",
    "categories": [
      "Tooling"
    ]
  }
]