Skip to content

Changelog — Guide

All notable changes to Repository Harnesses for AI Coding Agents: A Practical Guide are tracked here.

Skills version independently. Each skill has its own changelog under .agents/skills/<skill>/CHANGELOG.md.


1.1.0 (2026-08-05)

Changed

  • Introduction (guide/00-Introduction.md): reworked Why This Matters — replaced the "For engineers / For managers" split with unified prose, and added the human-reviewed-context argument (checking vetted context into the codebase makes agent responses more predictable across sessions and reduces the repeated repository-level rediscovery each session would otherwise do).
  • Install docs — canonical page (guide/README.md): guide/README.md → Skills is now the single source for per-tool install/update instructions. Reframed the GitHub Copilot section around the Claude-format marketplace manifest (same repo-harness plugin via copilot plugin marketplace add / install, or the chat.plugins.marketplaces setting), with the standalone .vsix as a niche fallback. Labeled the VS Code Agent Plugins path as Preview, and scoped the "not automatic" update guidance per install method (native scan tracks the checkout, copies must be re-copied, Cursor uses the scan).
  • Deduplicated install docs (README.md, harness-plugin/README.md, AGENTS.md): removed the copied per-tool install blocks (per INVARIANTS.md "link, don't copy") — the root README and plugin README now link to the canonical page, and the AGENTS.md Module Context pointer replaces its stale /plugin install repo-harness command with a link.

Added

  • Update instructions (guide/README.md): documented that plugin/marketplace updates are not automatic by default (except VS Code's periodic extension check) and added the per-tool update commands — Claude Code auto-updates or manual /plugin update, and copilot plugin update (Copilot CLI).

1.0.0 (2026-08-01)

Initial public release.

  • Ten-chapter guide (00–09) covering the case for repository harnesses, the five control layers, seven harness components, how agents navigate a harness, a reference repository layout, and step-by-step guides to build, migrate, and maintain a harness — plus a reading list, FAQ, and this changelog.
  • Two skillsharness-setup (build or migrate a harness; pairs with chapters 7–8) and harness-inspect (health check or full audit of an existing harness; pairs with chapter 9) — installable via the Claude Code and Cursor plugins, or read natively by any agent that supports .agents/skills/.
  • Plugin distribution: Claude Code (.claude-plugin/) and Cursor (.cursor-plugin/) marketplace manifests, plus a harness-plugin/ package for VS Code / GitHub Copilot.
  • Reference layout: this repository is itself structured as the harness it describes — AGENTS.md, INVARIANTS.md, .agents/skills/, and a CLAUDE.md shim.