mirror of
https://github.com/multica-ai/multica.git
synced 2026-07-29 06:28:23 +02:00
c366cf2ba1
* feat(agent): add kiro cli acp runtime * fix(agent): align kiro acp prompt and notifications * chore(agent): clarify kiro acp args compatibility
68 lines
5.1 KiB
Plaintext
68 lines
5.1 KiB
Plaintext
---
|
|
title: Skills
|
|
description: Attach "knowledge packs" to an agent — compatible with the Anthropic Agent Skills open standard.
|
|
---
|
|
|
|
import { Callout } from "fumadocs-ui/components/callout";
|
|
|
|
A Skill is a **knowledge pack** for an [agent](/agents) — a `SKILL.md` plus optional supporting files (scripts, configs, reference templates) that tell the agent "when you hit this kind of task, think and act like this." Multica adopts the [Anthropic Agent Skills](https://agentskills.io) open standard, so any compliant Skill — from Anthropic's official repository, ClawHub, or skills.sh — can be imported directly.
|
|
|
|
## Workspace skills vs. local skills
|
|
|
|
Multica supports two skill sources:
|
|
|
|
- **Workspace skill** — stored in Multica's cloud. Once attached to an agent, it's synced down to your daemon at task execution time. This is the **standard way to share skills across a team**.
|
|
- **Local skill** — lives in a directory on your machine (each AI coding tool has a conventional default path, e.g. Claude Code's `~/.claude/skills/`). On your request, the [daemon](/daemon-runtimes) scans your machine, and you manually pick which ones to bring into the workspace.
|
|
|
|
Most of the time you want **workspace skills**: import once, every teammate's agent can use it. Local skills are a fit when you want to test locally first, or when the content involves sensitive local material.
|
|
|
|
## Importing a skill
|
|
|
|
Workspace skills come from four sources:
|
|
|
|
- **New** — write the `SKILL.md` and related files directly in the UI
|
|
- **From GitHub** — paste a repo URL (e.g. `https://github.com/owner/repo/tree/main/skills/my-skill`) and Multica pulls the `SKILL.md` and every file in that directory
|
|
- **From ClawHub** — search and import from the [ClawHub](https://clawhub.io) public marketplace, with version selection
|
|
- **From local** — the daemon scans skill directories on your machine, and you pick which to bring into the workspace
|
|
|
|
Both individual files and whole skill packs have size caps (single-file cap around 1 MB when importing from GitHub). The exact rules appear in the import dialog — exceeding them returns an error.
|
|
|
|
## Attaching to an agent
|
|
|
|
Once imported, a skill has to be **attached to a specific agent** to take effect. One agent can have multiple skills attached, and one skill can be attached to multiple agents.
|
|
|
|
After attaching, the agent picks up its skills the next time it starts a task — each AI coding tool has its own skill discovery path (Claude Code uses `.claude/skills/`, Cursor uses `.cursor/skills/`, etc.), and Multica drops files in the right place automatically. **However, three tools (Gemini, Hermes, OpenClaw) currently use the generic fallback path `.agent_context/skills/` — whether these tools actually read skills from that path depends on the tool itself.** Full path mapping and the native-discovery vs. fallback distinction is in [AI coding tools comparison → Where skill files go](/providers#where-skill-files-go).
|
|
|
|
After you edit a skill's contents, **only newly created tasks pick up the new version** — tasks already running continue with the old skill.
|
|
|
|
## Safety of third-party skills
|
|
|
|
Skills imported from GitHub or ClawHub may include scripts and executable content. Multica itself **does not sign, audit, or sandbox them** — skill contents are handed to the corresponding AI coding tool as-is, and whether the tool treats them as executable is up to the tool.
|
|
|
|
<Callout type="warning">
|
|
**Before importing a third-party skill, review the `SKILL.md` and every file that ships with it.**
|
|
|
|
In February 2026 the "ClawHavoc" incident — malicious instructions planted in a popular skill pack stole API keys from affected users. ClawHub has since added VirusTotal scanning, but **automated scans are not a substitute for your own review**.
|
|
|
|
**Only import from sources you trust.** For projects involving sensitive data, consider using only local skills you wrote yourself.
|
|
</Callout>
|
|
|
|
## Skills vs. MCP
|
|
|
|
Both augment what an agent can do, but in different directions:
|
|
|
|
- **Skill** = a structured **knowledge pack** (static content + instructions). The agent reads a skill to learn "when I see problem X, here's how to think and what to do."
|
|
- **MCP** (Model Context Protocol) = a **tool channel**. The agent uses MCP to connect to external services (databases, filesystems, third-party APIs) and **invoke** them.
|
|
|
|
The two are complementary. In Multica today, MCP support is **only truly consumed by Claude Code** — other tools receive the MCP config but don't actually use it. A dedicated MCP section will come in a later release.
|
|
|
|
---
|
|
|
|
By now you know what an agent is, how to create one, and how to attach skills. The next question: **where does it actually run, and why does my agent sometimes get stuck?** The next chapter covers the execution architecture — daemons, runtimes, and how tasks work together.
|
|
|
|
## Next steps
|
|
|
|
- [Daemon and runtimes](/daemon-runtimes) — where agents actually run, and how to tell online from offline
|
|
- [Executing tasks](/tasks) — the full lifecycle of one "agent work session"
|
|
- [AI coding tools comparison](/providers) — full comparison of all 11 tools (including each one's skill injection path)
|