HarnessMatch
Menu

Coding agent

mini-SWE-agent

Minimal bash-only coding harness for inspectable CLI use and reproducible research.

Status
Active
License
MIT
Evidence
Documented + code-verifiable + independently measured configuration, 23 sources
Product record checked
2026-07-27
Compare harnessesInspect sources

At a glance

A deliberately small open-source software-engineering agent with one linear bash-only loop, confirm, yolo, and human control modes, versioned trajectories, local-model routing, optional isolated environments, and batch evaluation runners.

Good choice if

  • Transparent baselines and controlled model-harness studies
  • Developers who want a simple CLI with confirmation before each proposed action
  • Pinned SWE-bench or ProgramBench batches with structured trajectory output

Check before choosing

  • The default local environment executes directly with the user's host privileges; isolation is opt-in
  • Confirm mode prompts per proposed action, but it is not a scoped permission policy and there is no rollback checkpoint
  • The bash-only surface omits MCP, browser tools, delegated agents, and persistent memory
See 2 more considerations
  • Local models require a compatible endpoint and may need manual cost-registry configuration
  • Interrupted batch runs are rerun manually; trajectories are written only after a task completes

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.

    EvidenceControl flow

  • Repository tool execution

    Documented

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

    EvidenceControl flow

  • Task-aware context management

    Documented

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

    EvidenceDefault agent implementation

  • Model-independent runtime control

    Documented

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

    EvidenceCLI modes

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 & integrationsBuilt-in toolsDocumented mechanism, not a performance score.
  • Context & stateBasic sessionDocumented mechanism, not a performance score.
  • Lifecycle & recoveryManual recoveryDocumented mechanism, not a performance score.
  • ObservabilityStructured tracesDocumented mechanism, not a performance score.
  • VerificationManualDocumented mechanism, not a performance score.
  • Governance & permissionsApproval promptsDocumented mechanism, not a performance score.

Public code audit

5/5public artifacts present
Security policy
Present at inspected commit
CI workflow
Present at inspected commit
Automated tests
Present at inspected commit
Evaluation assets
Present at inspected commit
Contributor documentation
Present at inspected commit

The v2.4.6 tree contains 42 Python test files and six GitHub Actions workflows. Its SWE-bench and ProgramBench runners, configs, and trajectory utilities are project-owned evaluation assets rather than independent performance evidence; the separately admitted Terminal-Bench 2.1 run is pinned to mini-SWE-agent 2.4.5 and complete Harbor metadata.

Inspect commit a83fcae82d2a08f0ee0c688f9d137b3566c097f8, checked 2026-07-27

Capability support

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

External tools (MCP)
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
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
Not documentedNo first-class support established by the current recordAbsence of current documentation is not proof that the capability is impossible.
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
DocumentedDocumented execution-isolation mechanismSource · checked 2026-07-27The source establishes the mechanism, not its quality or availability in every mode.
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
mini-SWE-agent 2.4.6 releaseCurrent audited release, published from the inspected commitOfficial repositoryRepository readme at inspected commitProject scope, installation, intentionally minimal design, models, environments, and evaluation workflowsOfficial repositoryPackage metadata at inspected commitVersion 2.4.6, Python requirements, package status, dependencies, entry points, and MIT license metadataOfficial repositoryCLI modesInteractive terminal usage, confirm, yolo, and human modes, interruption controls, and local historyOfficial docsControl flowLinear agent loop, bash-only action surface, termination, and exception handlingOfficial docsExecution environmentsHost execution, Docker, Singularity, Bubblewrap, ConTree, and managed SWE-ReX environmentsOfficial docsModel quickstartLiteLLM-backed provider configuration, API keys, model names, and gateway examplesOfficial docsLocal model configurationLocal OpenAI-compatible or vLLM endpoints, custom model names, and manual cost registrationOfficial docs
View 15 additional sources
Additional first-party evidence15 sources
Trajectory output formatVersioned trajectory records, full configuration, model statistics and calls, messages, exit status, and submission outputOfficial docsTrajectory inspectorLocal browser-based inspection of saved trajectory filesOfficial docsSWE-bench runnerBatch task selection, parallel workers, environment options, output, and interrupted-run behaviorOfficial docsProgramBench runnerProgramBench task selection, parallel execution, environment configuration, and outputOfficial docsYAML configurationExplicit agent, model, environment, and runner configuration for reproducible runsOfficial docsPython bindingsProgrammatic setup and invocation outside the interactive terminal UIOfficial docsContribution guideDevelopment installation, testing, documentation, and contribution workflowOfficial docsSecurity reporting at inspected commitProject security reporting contacts and disclosure channelOfficial repositoryTest workflow at inspected commitProject-owned automated test workflow across supported Python versionsOfficial repositoryInteractive confirmation implementationPer-action confirmation, yolo switching, human input, and interruption behaviorOfficial repositoryDefault agent implementationSingle linear agent loop, bash command parsing, step limits, and trajectory serializationOfficial repositoryHost environment implementationDirect local subprocess execution with inherited host accessOfficial repositoryExperimental Bubblewrap environmentOptional Linux namespace and filesystem isolation implementationOfficial repositorySWE-bench runner at inspected commitProject-owned batch runner, parallel workers, completed-trajectory handling, and result extractionOfficial repositoryProgramBench runner at inspected commitProject-owned ProgramBench batch and trajectory generation utilitiesOfficial repository