BoldBlack Harness

BoldBlack Harness is an open-source command-line environment for running coding agents inside containers. The current project wraps three agent backends—pi, opencode and Hermes—behind one CLI, supports local LM Studio models by default and can route supported agents to multiple cloud providers through environment files. ¹

The product is deliberately narrower than a general agent platform: it is an execution wrapper and isolation layer. It mounts a selected directory or file into the agent container, supports persistent interactive sessions or ephemeral one-shot runs, and can expose user skills and context files when requested. ¹

Harness supports three agent backends, local or cloud model routes, environment-file configuration, user skills/context files, persistent state and single-file or project-directory mounts. This gives it meaningful flexibility inside developer workflows while keeping the surface intentionally CLI-centric rather than providing broad enterprise connectors. ¹

The package is public, MIT-licensed and installable through npx/npm-style package runners. A container runtime is required; local-model use also requires a compatible local inference service such as LM Studio. The software cost is therefore low, but users still bear model/API and local-compute costs where applicable. ^{2 1}

The repository, package metadata and release history provide strong first-party evidence of the current product surface and maintenance cadence. The fixed 50-source panel found no qualifying exact-product independent evidence: 43 sources returned no relevant result and seven were inaccessible during lawful discovery. Missing or inaccessible coverage is neutral rather than negative, but it materially limits rating confidence.

No eligible exact-product numeric user-rating input was found, so no external numeric adjustment is applied. AiToolMap did not perform a hands-on test.

The project is young but actively maintained: the repository was created in April 2026 and the current release line reached v1.10.0 in September 2026, with recent releases documenting fixes, integration tests and release-pipeline changes. ^{3 4}

The README documents capability dropping, no-new-privileges, seccomp restrictions on Docker, bounded mounts, home-directory protections, cosign image verification, SLSA provenance and dependency cooldowns. These are meaningful design disclosures and the code is publicly inspectable, but AiToolMap treats them as project claims/design evidence rather than independent penetration-test or security-audit results. ¹

Best fit: developers who want to run supported coding agents with a smaller host-exposure surface, especially local-first users switching among agent/model providers. Weaker fit: teams needing managed enterprise governance, a GUI, mature support guarantees or independently validated security/reliability evidence.