killall-agents

Four lane agents and workflow skills for Claude Code, Codex and opencode. Ordinary workflow skills live in killall-skills.

Install

Claude Code:

/plugin marketplace add killallgit/killall-agents
/plugin install killall-agents@killall-agents

Codex:

codex plugin marketplace add killallgit/killall-agents
codex plugin add killall-agents@killall-agents

Skills CLI (setup and rules only):

npx skills add killallgit/killall-agents

Includes setup-agents, orchestrate, follow-the-rules.

Install agents

From a project's git root:

python3 skills/setup-agents/scripts/install.py [--tools claude,codex,opencode] [--remove] [--force]

Writes one file per tool into .claude/agents/, .codex/agents/ and .opencode/agents/, each with that tool's model from factory/models.toml, so no agent inherits the session's model. Keeps the copies out of Git through .git/info/exclude and lists them in .worktreeinclude so Claude Code worktrees get them. Rerun to refresh; it removes copies from older versions and keeps every file it did not write. --force replaces a same-named agent it did not write. In session: /killall-agents:setup-agents in Claude Code or $setup-agents in Codex.

Lane agents

Edit factory/models.toml to change a model, then rerun the installer.

Agent	Job	Claude	Codex	opencode
planner	Turns spec into lanes, contracts, acceptance checks; read-only	Opus / high	gpt-6-astra / high	Opus
grunt	Relays task fitting 64K tokens to local Pi agent	Haiku / low	gpt-6-luna / low	Haiku
builder	Implements one lane in own worktree	Sonnet / medium	gpt-6.1-sol / medium	Sonnet
reviewer	Checks lane behavior, follow-the-rules and comments; runs gate independently	Sonnet / high	gpt-6.1-sol / high	Sonnet

Installing plugin adds no agents. To land finished work, use /lets-finish from killall-skills.

opencode agents run as subagents with the model pinned; the tools restriction is Claude Code only.

Ask top-level session to use orchestrate skill to run a spec. Skill is project-manager loop: plan with planner, run lanes with grunt and builders, check with reviewer. Agents read gate, branch, worktree rules from project's own AGENTS.md.

Grunt needs pi on PATH with ready provider. Builders and reviewers need included follow-the-rules skill.

Development

factory/<name>.md is source, in Claude Code's format. factory/models.toml maps each agent to a model per tool. The installer renders both into each tool's file at install time.

python3 skills/setup-agents/scripts/test_install.py
python3 scripts/test_release.py

Release with task tag:patch, task tag:minor or task tag:major: it tests origin/main, bumps the version in all three plugin manifests, tags it, and pushes.

See pending work.