HarnessMatch
Menu

Coding agent

Claude Code

Polished Claude-first agent across terminal, IDE, desktop, and web.

Status
Active
License
Proprietary
Evidence
Documented + independently measured configuration, 56 sources
Product record checked
2026-07-27
Compare harnessesInspect sources

At a glance

A Claude-centric coding agent with interactive and non-interactive workflows, explicit permissions, native sandboxing, MCP, hooks, plugins, delegated agents, background sessions, and subscription or API authentication.

Good choice if

  • Developers already paying for Claude Pro or Max
  • Interactive and delegated work on large repositories
  • Teams that want hooks, MCP, subagents, and explicit permission modes

Check before choosing

  • The local CLI is host-first: OS sandboxing is disabled by default, may fall back to unsandboxed execution when unavailable unless failIfUnavailable is enabled, and covers Bash subprocesses rather than every tool
  • Worktrees isolate file edits rather than processes; browser control runs in the user's real Chrome profile and can access authenticated sites
  • Checkpoint rewind omits Bash-created file changes and remote side effects; agent teams and agent view are research previews, and parallel background sessions consume subscription quota independently
See 8 more considerations
  • Server-managed settings are not fail-closed on an uncached first launch unless forceRemoteSettingsRefresh is explicitly enabled
  • Hosted model access is Claude-only, local open-weight models are unsupported, and Chrome integration requires a direct Anthropic plan
  • Managed Code Review is a separate Team and Enterprise GitHub surface whose neutral check does not block merges unless users add their own CI gate
  • Plugin marketplaces can install agents, hooks, MCP and LSP servers; organizations must configure marketplace allowlists when arbitrary sources are not acceptable
  • The optional unattended retry watchdog can retry capacity errors indefinitely and other transient failures for roughly three hours, so CI owners still need explicit time and spend boundaries
  • Auto mode is classifier-backed rather than a sandbox: Anthropic's published evaluation reports a 17% false-negative rate on 52 real overeager actions and says it is not a replacement for careful review in high-stakes workflows
  • Zero Data Retention is a separately enabled option for qualifying Enterprise API organizations and excludes web and Cloud sessions, Remote Control, feedback, third-party tools, and MCP integrations
  • The public GitHub repository exposes releases, plugins, examples, and issue automation rather than the proprietary core implementation or product test suite

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.

    EvidenceHow Claude Code works

  • Repository tool execution

    Documented

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

    EvidenceTools reference

  • Task-aware context management

    Documented

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

    EvidenceHow Claude Code works

  • Model-independent runtime control

    Documented

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

    EvidencePermissions

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 & isolationSandbox availableDocumented mechanism, not a performance score.
  • Tooling & integrationsExtensible + browserDocumented mechanism, not a performance score.
  • Context & statePersistent stateDocumented mechanism, not a performance score.
  • Lifecycle & recoveryCheckpoint/rewindDocumented mechanism, not a performance score.
  • ObservabilityStructured tracesDocumented mechanism, not a performance score.
  • VerificationTool-assistedDocumented mechanism, not a performance score.
  • Governance & permissionsPolicy controlsDocumented mechanism, not a performance score.

Public code audit

Unrankedsupport-only repository
Security policy
Present at inspected commit
CI workflow
Not found
Automated tests
Not found
Evaluation assets
Not found
Contributor documentation
Present at inspected commit

The v2.1.220 public tree contains release material, plugins, examples, a security policy, and 12 issue/support automation workflows, but not the proprietary core implementation, engineering tests, or product evaluation suite.

Inspect commit 7ef6eec9d9ba84ea6f233f26c45f1df5c5991843, checked 2026-07-27

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
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
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
DocumentedBuilt-in or product-supported browser controlSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
Isolated execution
OptionalOptional OS sandbox locally; managed isolation on cloud surfacesSource · checked 2026-07-27Local sandbox coverage depends on configuration and does not contain every process or side effect.
Undo file changes
DocumentedProduct-supported file or session rollbackSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.

Primary evidence

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

Product record checked 2026-07-27
Claude Code overviewSupported surfaces, CI integrations, MCP, cloud sessions, and remote workflowsOfficial docsHow Claude Code worksAgentic loop, built-in tools, context management, and host interaction modelOfficial docsPlatforms and integrationsCLI, Desktop, IDE, web, mobile, Slack, and CI surface comparisonOfficial docsProduct pageOfficial product surfaces, install paths, Pro and Max plan inclusion, and feature announcementsOfficial announcementCLI referenceTerminal, print mode, permissions, MCP, and automation flagsOfficial docsRun programmaticallyNon-interactive print mode, Agent SDK entrypoints, and headless automation constraintsOfficial docsTools referenceBuilt-in tool inventory, per-tool permission requirements, and execution behaviorOfficial docsSecurityRead-only defaults, sandbox Bash, working-directory boundary, prompt-injection and cloud safeguardsOfficial docs
View 48 additional sources
Product and interfaces6 sources
Chrome integrationBeta visible-browser automation, site permissions, authenticated state, and plan restrictionsOfficial docsComputer useCLI computer-use for native apps, screen control, and host GUI automation boundariesOfficial docsClaude Code on the webDurable cloud sessions, isolated VMs, network policy, credential proxying, handoff, and limitsOfficial docsDesktop applicationDesktop sessions, parallel worktrees, previews, computer use, and enterprise configurationOfficial docsVS Code extensionIDE integration, inline diffs, permission modes, and extension security and privacyOfficial docsJetBrains IDEsJetBrains plugin surfaces, selection context, and terminal-backed Claude Code sessionsOfficial docs
Execution and control8 sources
SandboxingOpt-in OS sandbox, fail-open default, scope, escape hatch, platforms, and limitationsOfficial docsSandbox environmentsBash sandbox vs runtime vs devcontainer vs VM vs web isolation boundaries and enforcementOfficial docsDevelopment containersContainer isolation, egress firewall example, managed policy, and unattended skip-permissions useOfficial docsPermissionsTool rules, approval modes, managed restrictions, sandbox interaction, and bypass controlsOfficial docsPermission modesManual, acceptEdits, plan, auto, dontAsk, and bypassPermissions tradeoffs and protected pathsOfficial docsConfigure auto modeClassifier trusted infrastructure, hard deny rules, environment slots, and org overridesOfficial docsAuto mode announcementClassifier-backed autonomy as a safer alternative to dangerously-skip-permissionsOfficial announcementAuto mode engineering evaluationClassifier design, 10,000-session production evaluation, real and synthetic false-negative rates, false positives, and high-stakes limitationOfficial announcement
Agents, state and recovery9 sources
Parallel agentsSubagents, agent view, experimental teams, batch delegation, and worktree choicesOfficial docsAgent viewResearch-preview background session manager, parallel dispatch, worktrees, permission inheritance, local lifecycle, and quota scalingOfficial docsSubagentsSpecialized delegated contexts, per-agent tools, permissions, hooks, and worktreesOfficial docsAgent teamsExperimental coordinated sessions, shared tasks, messaging, permissions, and known overheadOfficial docsDynamic workflowsScripted multi-subagent orchestration for audits, migrations, and cross-checked researchOfficial docsWorktree isolationParallel session and subagent file isolation, base refs, cleanup, and boundariesOfficial docsCheckpointing and rewindAutomatic file checkpoints, rewind, conversation restore, and external-side-effect limitationsOfficial docsSession managementLocal session persistence, resume, naming, branching, export, and retention controlsOfficial docsProject memoryCLAUDE.md hierarchy, default-on local auto memory, storage, loading limits, and user auditOfficial docs
Automation and extensions8 sources
Hooks referenceLifecycle hooks, pre-tool enforcement, subagent events, transcripts, and managed controlsOfficial docsMCP integrationLocal and remote MCP servers, authentication, scopes, resources, and security warningsOfficial docsPlugin marketplacesPlugin distribution, agents, hooks, MCP and LSP components, source pinning, updates, validation, and managed marketplace restrictionsOfficial docsRoutinesScheduled local and remote tasks, credentials, result delivery, and operational constraintsOfficial docsGitHub ActionsCI automation, pull-request and issue triggers, authentication, permissions, and securityOfficial docsManaged Code ReviewResearch-preview multi-agent pull-request review, verification and deduplication, triggers, plan scope, neutral checks, and optional user-defined CI gatingOfficial docsGitLab CI/CDGitLab pipeline automation, authentication, and CI security controls for Claude CodeOfficial docsAgent SDK overviewProgrammatic agent loop, tools, permissions, sessions, subagents, MCP, and deployment surfaceOfficial docs
Enterprise and operations13 sources
Managed MCP accessOrganization allowlists, denylists, and managed MCP server restrictionsOfficial docsSettingsUser and project settings hierarchy, sandbox keys, permission defaults, and managed deliveryOfficial docsServer-managed settingsCentral policy delivery, precedence, fail-closed refresh option, audit hooks, and limitationsOfficial docsAdmin setupEnterprise rollout map for providers, managed policy, monitoring, and data handlingOfficial docsAuthenticationClaude subscription, Console API, enterprise providers, credentials, and long-lived tokensOfficial docsFeature availabilityFeature matrix across subscription plans, Console, Bedrock, Agent Platform, and FoundryOfficial docsEnterprise network configurationProxy, custom CA, mTLS, and required network domains for enterprise deploymentsOfficial docsLarge codebasesMonorepo and large-tree guidance with nested CLAUDE.md, sparse worktrees, and package focusOfficial docsUsage monitoringOpenTelemetry metrics and events, configuration, privacy boundaries, and dashboardsOfficial docsData usageWhat Claude Code transmits, retention posture, telemetry services, and opt-out controlsOfficial docsZero Data RetentionQualified Enterprise enablement, direct-inference scope, excluded product surfaces and integrations, and model availability changesOfficial docsAgent SDK secure deploymentIsolation, credential management, and network controls for hosted Agent SDK agentsOfficial docsRuntime errors and retriesRuntime failure classes, automatic retries, unattended watchdog behavior, timeouts, partial responses, recovery, and provider-specific errorsOfficial docs
Releases and public code audit4 sources
Claude Code changelogVersioned release notes for features, permission modes, sandboxing, and surface changesOfficial docsClaude Code 2.1.220 releaseCurrent public distribution release used for product-version verificationOfficial announcementClaude Code public repositoryRelease artifacts, plugins, examples, support automation, license, and absent core sourceOfficial repositoryAnthropic sandbox runtimeOpen Seatbelt and bubblewrap runtime for process filesystem and network isolationOfficial repository