agent-skill-groups

name	agent-skill-groups
description	Manage large local Agent Skills libraries with runtime-aware scenario groups for Codex, Claude Code, OpenCode, GitHub Copilot, and generic SKILL.md roots.

Agent Skill Groups

Use this skill when the user wants to inspect, group, profile, enable, disable, or clean up local SKILL.md directories across agent runtimes.

Workflow

1. Identify the runtime: codex, claude-code, opencode, or generic.
2. Inspect the current skill inventory before moving anything.
3. Generate or load a groups.json file.
4. Preview changes with plan.
5. Generate persistent agent instructions with memory.
6. Create a backup before moving directories.
7. Switch profiles only after the plan is understood.
8. Validate the final filesystem state.

For a safe end-to-end rehearsal, run agent-skill-groups demo --json first. It creates sample active and disabled skill roots, writes groups.json and an agent memory block, switches a profile, restores the backup, and reports ok: true when the reversible workflow succeeds.

Commands

agent-skill-groups runtimes
agent-skill-groups demo --json
agent-skill-groups analyze --runtime codex --json
agent-skill-groups suggest --runtime codex --json
agent-skill-groups init --runtime codex --output groups.json
agent-skill-groups status --config groups.json --runtime codex --json --details
agent-skill-groups route --config groups.json --runtime codex --task "solve a web CTF with IDOR" --json
agent-skill-groups route --config groups.json --runtime codex --task-file task.txt --emit-command session-load
agent-skill-groups plan --config groups.json --runtime codex research
agent-skill-groups memory --config groups.json --runtime codex --write AGENTS.md
agent-skill-groups backup --config groups.json --runtime codex
agent-skill-groups profile --config groups.json --runtime codex research
agent-skill-groups validate --config groups.json --runtime codex --json
agent-skill-groups discovery --json

Use AGENT_SKILL_GROUPS_CONFIG when scripts or hooks need a stable config path.

Safety

• Do not edit skill contents during grouping.
• Do not delete skills as part of a profile switch.
• Keep .system and other protected skills in core.
• Prefer --json output when another agent or script consumes results.