Changes
Skill evals gain a second, agentic harness (claude plugin eval: real tool
use, with/without-plugin delta) alongside the existing single-turn one,
piloted on using-git-worktrees.
Fixed
using-git-worktreesnow fires for "remove this worktree and its branch"
cleanup requests, not just creation — the routing description previously
never mentioned removal, even though the skill has a full cleanup section.
Added
claude plugin evalagentic eval harness:scripts/evals/prepare-plugin-evals.py,
make plugin-evals[-prepare]. Piloted onusing-git-worktrees(4 cases);
seedocs/skill-evals.mdfor the harness layout and the pilot's
with/without results.
Plugins
| Plugin | Description |
|---|---|
| browser | Browser-based testing, validation, screenshots, recordings, and quick automation |
| dev-flow | Code review, fixes, commits, release flows, and linting for development workflows |
| discovery | Research, docs lookup, utility routing, skill authoring, instruction review, and agent configuration audits |
| git-flow | Git worktrees, branch cleanup, hooks, secret scanning, config hygiene, and guardrails |
| infra-ops | Kubernetes, Terraform, Helm, GitHub Actions, AWS, GCP, and Linux infrastructure operations |
| programming | Idiomatic development across C# /.NET, Go, Java/Kotlin, Python, Rust, TypeScript, shell, and web |
| spec-flow | Lightweight spec loop: plan one slice, execute one task, checkpoint or close |
Distribution
Artifacts are rendered by Agent Bundler from the repository root:
agbun build --root .