HarnessMatch
Menu

Pair programmer

Crush

Terminal-native, multi-model coding with LSP, MCP, and a read-only task agent.

Status
Active
License
FSL-1.1-MIT
Evidence
Documented + code-verifiable, 20 sources
Product record checked
2026-07-27
Compare harnessesInspect sources

At a glance

A source-available terminal coding agent for interactive or scripted work, with multi-provider and local-model support, a Hyper subscription option, mid-session model switching, project sessions, LSP, MCP, skills, hooks, per-tool approvals, and parallel read-only research delegation.

Good choice if

  • Terminal-first developers who want a polished interactive UI plus a non-interactive run command
  • People choosing between Hyper subscription access, API providers, and local models
  • Projects that benefit from LSP, MCP, skills, and a parallel read-only research agent

Check before choosing

  • The built-in task agent delegates parallel read-only search, not independent write-capable implementation
  • No first-party execution sandbox, browser automation, or file checkpoint mechanism was found
  • PreToolUse hooks cover top-level tool calls but do not intercept tools used inside the task agent
See 1 more considerations
  • Project configuration is trusted code and can execute shell expansion at startup; host approvals can also be bypassed with --yolo

Why it qualifies as a coding harness

This confirms category fit, not product quality. Every required criterion links back to first-party evidence.

Qualifies4 of 4 required criteria evidenced

  • Adaptive agent loop

    Documented

    The system repeatedly observes results and chooses the next action instead of following a fixed one-pass graph.

    EvidencePinned Crush overview

  • Repository tool execution

    Documented

    The system can use tools to inspect and change a repository or its execution environment.

    EvidencePinned agent tests

  • Task-aware context management

    Documented

    The runtime assembles, updates, compacts, retrieves, or persists task-relevant context while work proceeds.

    EvidencePinned session store

  • Model-independent runtime control

    Documented

    Permissions, budgets, interruption, policy, or stop controls operate outside the model's own text generation.

    EvidencePinned permission manager

Membership establishes category fit only. It does not score quality, safety, autonomy, model capability, or benchmark performance. · Read the membership rule.

How it works under the hood

Seven mechanisms mapped from first-party records. These labels describe what the harness provides, not how intelligent its model is.

7/7layers documented
  • Execution & isolationHost processDocumented mechanism, not a performance score.
  • Tooling & integrationsExtensible toolsDocumented mechanism, not a performance score.
  • Context & stateManaged contextDocumented mechanism, not a performance score.
  • Lifecycle & recoverySession resumeDocumented mechanism, not a performance score.
  • ObservabilityLogs/transcriptsDocumented mechanism, not a performance score.
  • VerificationTool-assistedDocumented mechanism, not a performance score.
  • Governance & permissionsApproval promptsDocumented mechanism, not a performance score.

Public code audit

3/5public artifacts present
Security policy
Not found
CI workflow
Present at inspected commit
Automated tests
Present at inspected commit
Evaluation assets
Not found
Contributor documentation
Present at inspected commit

The repository includes cross-platform CI, unit, integration, end-to-end, and VCR-backed agent tests; these are project-owned engineering tests, not an independent harness benchmark or a public performance result. No security policy or dedicated evaluation suite was found.

Inspect commit def12cc6d8e1, checked 2026-07-27

Measured configurations

No benchmark run passes the full metadata admission policy for this harness yet. Missing data is not scored as zero.

Benchmark policy and all runs

Capability support

Documented first-class product support, checked against the sources below.

External tools (MCP)
DocumentedProduct-supported MCP integrationSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Local models
DocumentedLocal or self-hosted model pathSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Agent parallelism
DocumentedDelegated or parallel agent workflowSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Runs without an open UI
DocumentedNon-interactive or automation surfaceSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Browser control
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
Isolated execution
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
Undo file changes
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.

Primary evidence

Each capability claim is tied to a first-party record and a verification date.

Product record checked 2026-07-27
Pinned Crush overviewTerminal workflow, providers, Hyper, local models, sessions, LSP, MCP, skills, hooks, permissions, configuration trust, logs, and storageOfficial repositoryCrush configuration schemaProvider endpoints, model configuration, permissions, tools, MCP, and session settingsOfficial docsCrush v0.87.0 releaseLatest reviewed release, MCP OAuth, Channels preview, LSP improvements, event reconnection, and signed artifactsOfficial repositoryHyper provider plansFirst-party free credit tier, paid subscription, hosted coding models, and zero-data-retention claimOfficial docsPinned CLI root commandInteractive terminal entry point, session continuation, yolo mode, project logging, and run command examplesOfficial repositoryPinned non-interactive run commandPrompt input from arguments or stdin, quiet and verbose output, model selection, session reuse, and event streamingOfficial repositoryPinned experimental server commandExperimental shared server and client mode for multiple terminal clientsOfficial repositoryPinned task-agent implementationDefault task-agent construction, child sessions, parallel delegation, and agent-tool invocationOfficial repository
View 12 additional sources
Additional first-party evidence12 sources
Pinned task-agent instructionsRead-only glob, grep, listing, and file-view tools exposed to the delegated research agentOfficial repositoryPinned agent and provider configurationCoder and task agents, read-only task permissions, provider setup, MCP configuration, tools, logging, and agent settingsOfficial repositoryPinned permission managerPer-tool prompts, allow once, session grants, denials, pre-approved tools, hook decisions, and automatic approval modeOfficial repositoryPinned hooks documentationPreToolUse allow, deny, and rewrite behavior plus the documented lack of interception inside subagentsOfficial repositoryPinned session storeSQLite-backed sessions, child task sessions, messages, usage, cost, and todo persistenceOfficial repositoryPinned log commandProject log inspection, follow mode, and runtime diagnostic outputOfficial repositoryPinned MCP lifecycleMCP server lifecycle, tools, prompts, resources, OAuth handling, and channel integrationOfficial repositoryPinned contributor guideProject architecture, coder and task agents, SQLite, LSP, MCP, test commands, and contribution workflowOfficial repositoryPinned licenseFSL-1.1-MIT terms, competitive-use restriction, and future MIT license conversionOfficial repositoryPinned cross-platform CI workflowUbuntu, macOS, and Windows builds plus race-enabled Go test executionOfficial repositoryPinned agent testsProject-owned VCR-backed tests for agent tool use, including reading, writing, and shell executionOfficial repositoryPinned server end-to-end testsProject-owned end-to-end tests for server events, agent turns, cancellation, and session behaviorOfficial repository

Ecosystem discovery

OpenRouter coding apps Discovery signal only; usage rank is not used as a quality or capability score. Observed 2026-07-27.