agent-plugins

Agent skill collection for parallel multi-branch development workflows. Compatible with Claude Code (/skill-name) and any agent that reads plain Markdown (Codex, GPT, etc.).

Zero-install usage: add this repo as a submodule and point your project's AGENTS.md at AGENTS.md (the skill index). Any agent then discovers and loads every skill on demand — no per-skill install, and future skills appear automatically. The Claude Code marketplace below is an optional native-command layer.

Skills

Skill	Description
scrumban	Durable planning: write-before-do, SPRINT vs BACKLOG, resume-cold entries, queue follow-ups — so a reboot//clear/parallel agent resumes from the board cold
bugs	Durable bug handling: a BUGS.md ledger (status/repro/SHA/notes) + a fix loop (reproduce in the real harness, faithful regression test), coordinated via rozum
multi-agent	Coordination protocol for parallel agents: claim/heartbeat/triage/release + autonomous loop
multi-repo	Workspace management for repos listed in REPOS.md: status, sync, update, clone
spec-dev	Spec-driven development: write spec → implement → verify, keep spec in sync
plan-mode-bypass	Restore bypassPermissions after approving a plan in Claude Code plan mode

Usage

Option A — as a git submodule (recommended for projects)

# Add once to your project
git submodule add https://github.com/sergey-scherbina/agent-plugins .agents/plugins
git submodule update --init .agents/plugins

Submodules are only initialized in the shared main checkout. Worktrees do not need their own submodule init. From any worktree, read skills via the main repo root:

MAIN=$(git worktree list | head -1 | awk '{print $1}')
# read: $MAIN/.agents/plugins/<skill>/commands/<skill>.md

Reference this in AGENTS.md:

Skill files: `$MAIN/.agents/plugins/<skill>/commands/<skill>.md`
where MAIN=$(git worktree list | head -1 | awk '{print $1}')

Option B — install to ~/.claude/commands/

git clone https://github.com/sergey-scherbina/agent-plugins
cd agent-plugins
./install.sh              # all skills
./install.sh multi-agent  # one skill

Files land in ~/.claude/commands/<skill>.md. Works for Claude Code slash commands and any agent that resolves ~/.claude/commands/.

Option C — Claude Code plugin marketplace

claude plugins marketplace add github:sergey-scherbina/agent-plugins
claude plugins install multi-agent

Updating

# Submodule
git submodule update --remote .agents/plugins

# Installed copy
git -C agent-plugins pull && ./install.sh

# Claude Code marketplace
claude plugins update multi-agent

Adding a new plugin

my-plugin/
├── commands/
│   └── my-plugin.md      # skill definition — plain Markdown, no tool-specific frontmatter
├── hooks/                # optional: shell scripts for Claude Code hooks
│   └── auto-allow.sh
└── README.md

install.sh auto-discovers any directory that has commands/<name>.md.