Methodology

What every entry contains

Each harness entry uses the same template: a table of facts (maintainer, host agents, component counts, license), a feature table using the definitions on the compare page, and a description of its workflow, components, how it steers the agent, and the files it writes. The terms are explained on how harnesses work. Entries list every source they used.

How entries are researched

Harnesses are open source, so entries are researched from the source code at a specific commit, which each entry records. An agent clones the repository, counts components with shell commands, and answers a fixed set of questions (one per template section and one per feature), citing the file that supports each answer. Anything the files do not settle is marked unknown.

DeepWiki, which indexes repositories and answers questions about them, is used as a second opinion when available. Each entry links to its DeepWiki page so you can ask your own questions. An entry's sources say whether DeepWiki was used.

Status

• Draft: written from the sources listed, but not yet checked by a person.
• Verified: a person has checked the claims against the sources.

Every entry shows the date it was last researched. Harnesses change quickly; treat anything older than three months with caution.

Feature values

• yes: supported and documented.
• partial: supported with limits. The note says which.
• no: not supported.
• n/a: does not apply to this kind of harness.
• unknown: not yet researched.