Skills for AI agents

Sombra's MCP tools let an AI agent save, search, and organise your library. Skills go one step further: they package repeatable workflows — project tracking, feature specs, technical research — into slash commands your agent runs directly.

Everything a skill produces is stored in Sombra, so your work persists across conversations. Close the chat, open a new one next week, and pick up exactly where you left off.

The skills are open source at github.com/sombra-hq/sombra-skills. For a copy-paste setup guide and live examples, see the Skills page.

Installing

Claude Code

/plugin marketplace add sombra-hq/sombra-skills
/plugin install sombra

On first use you'll be prompted to connect your Sombra account via OAuth. Free accounts work — no credit card required.

Other MCP clients

The Sombra tools that power these skills are available to any MCP client — Claude Desktop, Claude.ai, ChatGPT, OpenCode, and more. Connect to https://sombra.so/mcp via Streamable HTTP with OAuth 2.1. See the MCP docs for step-by-step setup. The slash-command packaging is specific to Claude Code's plugin system, but the underlying workflows work anywhere the tools are available.

The skills

/sombra:guide — general Sombra usage

The entry point for everyday use. Save web pages, browse collections, distil context, and search your library. Sombra tools also activate automatically when you mention research, saving, or looking something up — no slash command required.

/sombra:guide save this page

/sombra:project — living project tracker

Track any multi-phase effort across conversations: house purchase, renovation, content strategy, product launch, learning goals. Creates a collection with structured phases, checkboxes, a decisions log, and open questions. Updates are surgical — one changed item doesn't rewrite the whole document.

/sombra:project Kitchen Renovation

Works in Claude Desktop, Claude.ai, ChatGPT, Claude Code, and any MCP client.

/sombra:spec — living feature spec

Everything in project, plus git branch context, code-path references, and commit-based progress detection. Tracks which branch, which files, and which items are done — and cross-references your git log to suggest what to check off. For software features; requires git and file access (Claude Code, OpenCode).

/sombra:spec Auth Rework

/sombra:pickup — pick up where you left off

Loads an existing project or spec, shows what's changed, and highlights the next step. Adapts automatically — git-aware in a repo, plain status report elsewhere.

/sombra:pickup
/sombra:pickup Auth Rework

/sombra:update — force-sync a tracker

Push updates into an existing tracker in one pass. Gathers signals from your input, git history, and conversation context, then marks items complete, adds new items, logs decisions, and creates supporting artifacts.

/sombra:update batch 4 done, KR-6 cancelled — plumber unavailable

/sombra:research — build a technical reference

Deep research on APIs, libraries, and frameworks. Saves sources to Sombra and distils them into a dense context document any agent can load to write correct code without stale training data. Works anywhere with web search access.

/sombra:research Pedestal HTTP routing

/sombra:verify-distill — verify a distilled context

Scan a collection's distilled context and classify every substantive claim against its saved sources — anchored, loose, or unsourced — then walk through cleaning up the unsourced ones. Keeps your distilled context grounded in real artifacts.

/sombra:verify-distill Lacinia GraphQL Reference

How they fit together

Each skill uses Sombra collections as its persistence layer:

• Collections organise related material.
• Context documents are the distilled, structured view — the spec, the project plan, the API reference — read in full at the start of every session.
• Artifacts hold the depth: item specs, reports, decision history, archived phases, and saved web pages, fetched only when needed.

Trackers stay fast to load no matter how long a project runs: completed phases collapse to one-line stubs, the decision log keeps only recent entries inline, and updates are surgical rather than full rewrites.

Compatibility

For a copy-paste install guide and per-skill examples, see the Skills page. To connect a client, see the MCP docs.