@rag-rat/skills

One-command installer for rag-rat's agent skills — reusable instruction sets that teach your coding agent to get the most out of a rag-rat-indexed repo.

npx @rag-rat/skills

That installs the skills into whichever agents it detects (Claude Code → .claude/skills/, Codex → .codex/skills/, Cursor, opencode, and 70+ others).

What you get

| Skill | What it does | |---|---| | using-rag-rat | The working rule for any rag-rat repo: reach for the MCP tools (semantic_search, symbol_lookup, impact_surface, the call graph, important_symbols) to find and understand code before grep, and record durable, non-obvious learnings as cross-agent rag-rat memories before finishing. | | dream-review | Triage and resolve the rag-rat dream memory-maintenance worklist: per finding kind, fix the underlying memory / coverage gap (resolving the finding at the root) or record an accept / dismiss verdict. | | init-rag-rat | Set up rag-rat in a not-yet-indexed repo (a dormant MCP server): scan the repo, guide the embedding-backend choice (local FastEmbed, a local infinity Docker server, or an ephemeral Modal/RunPod GPU worker), preview + write the config after confirmation, run the first index, and prompt an MCP restart. | | configure-rag-rat-dream | Enable and operate the optional AI memory-maintenance passes (dream --verify/--compact): configure the [llm.dream] chat model (Connect, or ephemeral Modal/RunPod via cookbook — vllm/ollama), and run it on demand or on a schedule (systemd timer). Pairs with dream-review, which triages the findings it produces. |

Commands

npx @rag-rat/skills                 # install (default)
npx @rag-rat/skills update          # refresh rag-rat's skills to the latest
npx @rag-rat/skills list            # list installed skills
npx @rag-rat/skills remove          # remove rag-rat's skills

update and remove (plain, or with only -g/-y) default to rag-rat's own skills (using-rag-rat, dream-review, init-rag-rat, configure-rag-rat-dream) — they won't touch unrelated skills you've installed. Pass your own targets — a skill name, --agent, or --all — to drive the underlying skills CLI directly instead.

Flags are forwarded to the underlying installer:

npx @rag-rat/skills -a claude-code       # only Claude Code
npx @rag-rat/skills -s using-rag-rat     # only one skill
npx @rag-rat/skills -g                   # install to your home dir (global), not the project
npx @rag-rat/skills --copy               # copy files instead of symlinking
npx @rag-rat/skills -y                   # non-interactive

How it works

This package is a thin wrapper over the skills CLI, pinned to rag-rat's curated public skill bundle (plugin/skills in the rag-rat repo). It doesn't reinvent the multi-agent installer — skills already knows how to place a SKILL.md into every supported agent, symlink-or-copy, project-or-global. npx @rag-rat/skills is just the branded, single-command entry point; it forwards your flags verbatim.

Equivalent to running:

npx skills add https://github.com/cq27-dev/rag-rat/tree/main/plugin/skills

The public skills are copied from the repository's .agents/skills source of truth into plugin/skills/. Repository-only contributor skills remain under .agents/skills and are deliberately absent from this installer.