Agent Skills

Portable Agent Skills for Claude, Codex, and other clients that support the open Agent Skills format. Skills live in .agents/skills/, one directory per skill with a SKILL.md inside.

Current Skills

• summarize-cli: run a command and ask an LM Studio local model to summarize, classify, extract, or explain the captured output.
• mailbox: file-based message passing between agent sessions, with iam (register), send, read, wait, and clean subcommands, plus a separate watch script for monitoring.
• agent-collaboration: playbook that ties the mailbox, shared-memory, handoff, and spawn-agent skills into collaboration modes, a coordination message shape, conflict control, and a verification loop.
• shared-memory: durable Markdown record of objectives, decisions, tasks, and notes under a per-workspace namespace both Claude and Codex can read.
• spawn-agent: spawn a one-shot Claude or Codex worker that runs a job and mails its result back to a mailbox address as a THREAD/VERB envelope.
• handoff: write a full-context handoff file, or pick up the latest and continue, under the tool-neutral root ~/.agents/handoffs/<repo>/.

Using These Skills

Codex discovers .agents/skills/ automatically, both in this repository and from ~/.agents/skills for global use.

Claude Code reads .claude/skills/ instead. This repository ships a .claude/skills symlink pointing at .agents/skills, so both tools work from a clone with no setup. The symlink requires a symlink-capable checkout, which excludes default Windows Git settings.

For global use, symlink a skill into each tool's user directory:

ln -s "$PWD/.agents/skills/summarize-cli" ~/.claude/skills/summarize-cli
ln -s "$PWD/.agents/skills/summarize-cli" ~/.agents/skills/summarize-cli

Validation

scripts/validate-skills.sh

This checks local SKILL.md files for the basic fields shared by compatible Agent Skills clients.