skillkitskillkitskillkitskillkitskillkitskillkitskillkit

skillkit is a catalog of skills for your coding agent, each a concrete, repeatable capability rather than a prompt you retype every session. Four of the five address the same gap: an agent is good at doing the work and bad at leaving a trail behind it. debug-decisions records architectural choices as versioned Markdown with a revert plan attached, spec-versioning puts specs and implementation plans where the repo can find them, memory-org settles which memories belong to the project and which belong to you, and sync-skills keeps the whole set in step across machines. mind-gym is the odd one out and stays on purpose: it trains the human, not the agent.

An interactive installer (npx github:sondalab-ai/skillkit, no clone needed) lists what is available, lets you multi-select, and copies each chosen skill into your agent’s skills directory. Skills that ship extra wiring, slash commands or hooks, are set up on request, with your settings backed up first and the change made idempotently. The uninstaller runs each skill’s teardown, removes its commands and hooks, and deletes the directory, but never touches recorded user data, which you remove by hand if you want it gone.

Two of those five now have a second home. claude-loadout implements the same jobs as commands, writing the identical decision records into the identical directory and following memory-org’s layout for its note store, and does per session what the skills leave to you by hand. If you run the launcher, take debug-decisions and memory-org from there. skillkit stays the standalone route, no Python and no wrapper around your agent command, and it keeps the one thing a launcher will not ship: the guided revert of a decision it recorded. spec-versioning, sync-skills and mind-gym live here and nowhere else.

The shared toolbox the rest of the suite reaches for when an agent needs to do something specific rather than reason it out from scratch. Public and open.