← ClaudeAtlas

ki-repo-harnesslisted

Audit, conform, and design Knowledge Islands compatible harnesses — source repositories that co-locate skills, subagents, MCP servers, evals, and hooks while publishing a verified installed capability payload. Use when creating a harness, checking its five-part source layout, validating its declared capability prefix and skill identities, reviewing its CLAUDE.md orientation, confirming its `.ki.toml` harness declaration, or distinguishing source shelves from the directly installed payload. Triggers: "audit the harness", "scaffold a new harness", "does this repo follow the harness standard", "refresh the harness standard", "is this a compatible harness". Governs the container and publication boundary, not its contents: skill quality → `ki-skills`; agent quality → `ki-subagents`; repository roadmap → `ki-work-roadmap`; MCP code → `ki-repo-mcp`; engineering toolchain → `ki-engineering`; repository settings → `ki-repo`; CLI installation and activation → `tools-ki`.
knowledgeislands/ki-agentic-harness · ★ 0 · AI & Automation · score 76
Install: claude install-skill knowledgeislands/ki-agentic-harness
# Knowledge Islands compatible harnesses This skill governs two related shapes: the five-part **source harness** where capabilities are authored together, and the smaller verified **compatible payload** installed from it. The installed payload, never a checkout or repository-local executor, is the source used by the `ki` host. The complete contract is in [the compatible harness standard](references/standards-compatible-harness.md). Its structured TypeScript catalogue under `scripts/rubric/items/` is canonical; [the generated rubric](references/rubric.md) is the readable publication. [Sources](references/sources.md) record provenance, and [exemplars](references/exemplars.md) illustrate the source-versus-installed distinction. ## Operating modes The universal modes are **AUDIT · CONFORM · EDUCATE · REFRESH**. ### Mode AUDIT — check a source harness and publication boundary - Run `ki repo audit --skill ki-repo-harness --repo <path>`. The host executes declared dependencies before the harness delta and reports mechanical results. - Review the generated rubric's judgment criteria, especially capability and payload boundaries. Source layout does not establish a verified payload, local-development selection, activation, resolved capability, component result, or execution; obtain those states from `tools-ki` or report them unavailable. - Run separately coverage-detected owning audits where applicable: `ki-engineering` for the development toolchain, `ki-repo` for repository gove