Most LLM work focuses on text generation; turning prompts into repeatable, auditable actions is the next friction point. This repo centralizes tiny, metadata-first instruction bundles—"skills"—that let Codex map natural-language intents to deterministic multi-step workflows and external app actions.
What Sets It Apart
- Metadata-first skill format: each skill is a self-contained folder with a SKILL.md (name + description + execution guidance), letting Codex decide when to trigger without loading heavy docs.
- Action-capable integrations: paired with the Composio CLI/connectors, skills can perform real-world actions (Slack, GitHub, Notion, email, etc.), not just produce text — useful for end-to-end automation scenarios.
- Curated and installable: a curated list (high community uptake — ~1.3k stars) with an installer/installer-scripts to drop skills into $CODEX_HOME/skills for quick reuse across sessions.
- Progressive disclosure design: long references and scripts live in skill folders and are loaded only when the skill fires, keeping agent context small and focused.
Who It's For & Trade-offs
Great fit if you build or operate agent-driven workflows and want reusable, human-readable instruction bundles that trigger cross-app actions from the terminal. Useful for developer tooling (PR triage, CI fixes), meeting-note automation, and internal ops where traceability matters. Look elsewhere if you need a standalone LLM UI or a full orchestration/orchestration framework out of the box — this repo is a curated skill library that depends on the Codex/Composio ecosystem and third-party connectors; skill quality varies by contributor and some workflows require external API keys or connector setup.
