What problem does it solve? Coding agents underperform when their environment is underspecified: they re-derive how to run the app every session, break sibling repos because nothing registered the seam between them, and drift across long sessions with no memory. This Skill builds and maintains the harness around the agent - AGENTS.md maps, boot scripts, acceptance specs, seam registries, and cross-repo execution plans - for clusters of tightly-related repositories. ## Core Features & Use Cases - Audit and doctor: Score a repo cluster against a 12-dimension rubric (map, bootability, inter-session memory, contract registry, deployment topology, and more) and detect drift such as dangling cross-repo links, duplicated skills, and plans that disagree with actual git branches. - Maintain loop (garden): A SessionEnd hook records raw session facts as JSON; the garden command batches that evidence into cited proposals for missing harness artifacts, which you review and apply onto branches - nothing is ever pushed or merged automatically. - Cross-repo exec-plans: Create plans with explicit merge orders for changes spanning multiple repos, then diff the plan against real git state with plan status. - Use Case: You manage a docs repo, a backend, and an e2e suite where spec moves keep breaking consumers. Run init to generate the cluster manifest, audit to find that no seam registry exists, then register each shared document in contracts.yaml with a verification command so drift fails a check instead of silently breaking e2e. ## Quick Start Ask the agent to run python harness.py init --workspace <your workspace dir> --name <cluster-name> from this skill's scripts directory, then run python harness.py audit to see the top harness gaps.
