Pegboard
Docs menu

Changelog

What changed in each Pegboard release.

Releases and binaries are on GitHub. Versions follow Semantic Versioning. Before 1.0, a minor version can change behavior.

0.2.0

November 4, 2026. Shared stacks, safer skills, and checks for CI.

Added

  • include pulls other stacks into yours, from a folder, a git repo (owner/repo//path?ref=v1), or a zip. Git and zip stacks are pinned in pegboard.lock and cached by content hash.
  • Approval of MCPs from shared stacks. An MCP that a git or zip stack brings needs apply --approve, and the plan shows its full command and where it comes from. The lock records what you approved, so a team reviews it once, in a pull request.
  • Trust boundaries. A shared stack can’t use ~/, absolute paths, or .. that leaves its folder.
  • policy.allow_skill_sources, next to allow_mcps. Policies add up across included stacks and can’t be relaxed.
  • Hidden Unicode checks. Skills with bidirectional overrides or Unicode tags are stopped in the plan, and pegboard audit scans every skill and context file your agents read.
  • SBOM export. pegboard lock export --format cyclonedx|spdx.
  • pegboard import --from apm.yml reads a Microsoft APM manifest. APM-style sources such as owner/repo/path#v1 work as skill sources.
  • GitHub Copilot CLI is the tenth supported agent.
  • GitHub Action to run drift, plan, or audit in CI.
  • pegboard update refreshes included stacks and lists new or changed MCP commands.

Changed

  • include is no longer ignored with a warning.
  • A skill dropped by remove.skills no longer counts toward the duplicate-name check, so remove can settle a clash between two sources.
  • The built-in skill tells agents never to pass --approve or --allow-hidden-unicode without asking you.

0.1.0

November 3, 2026. First release.

  • init, plan, apply, drift, and import for MCPs on Claude Code, Codex, opencode, Hermes, Pi, OMP, OpenClaw, Cursor, and Gemini CLI, in global and project scopes, with state and backups.
  • ${env:…} and ${file:…} secrets, kept as references where the agent supports them.
  • Skills from git and local sources, cached once, linked into each agent, and pinned in pegboard.lock.
  • context.agents_md and the @AGENTS.md bridge in CLAUDE.md.
  • Ad-hoc mode: install, uninstall, --save, and adopt.
  • replicate, to copy one agent’s setup to others.
  • The built-in skill, and install.sh with checksum verification.