Skill System

A Skill is a capability package for an Agent: a directory built around a SKILL.md file, containing instructions, scripts, and reference material. An Agent reads these on demand while running a task, enabling it to do something it otherwise couldn't—like generating a Word document with a table of contents and headers.

What A Skill Is

Each skill is a directory containing at minimum a SKILL.md: the frontmatter holds name and description, and the body is the instructions the Agent reads. The directory can also include attachments (scripts, reference docs, config templates) and declare dependencies and MCP servers.

---
name: weekly-report-style
description: 生成周报 Excel 时统一使用团队的配色与表头规范
---

# 使用说明

正文写给 Agent 看：什么时候用这个技能、
按什么步骤做、附带的脚本和参考文件怎么用。

Skills use progressive disclosure: normally the Agent sees only each skill's name and description, and reads in the full body and attachments only when it decides a skill applies to the current task. That way, even with many skills installed, not everything gets loaded into every conversation.

System Skills and User-Installed Skills

The platform's preinstalled general-purpose system skills fall into two main categories: Office document generation (docx / xlsx / pptx / pdf) and development skills (frontend, full-stack, iOS, Android, Flutter, React Native, and more). The Skill Management page in the console shows the full list, and opening any skill reveals its version, attachments, and the full SKILL.md.

Office Document Generation

This group of system skills lets Agent steps produce office document files directly, not just Markdown text:

A typical use: an Agent step in a daily-report workflow uses the xlsx skill to produce a formatted weekly report spreadsheet, downloaded with the execution's artifacts or delivered straight to Slack or email. Artifacts can also be published as public links; seePublic Artifact Sharing.

Skill Marketplace: Installing and Versions

The ClawHub Skill Marketplace page in the console connects to the ClawHub registry. Search by keyword, view each skill's version, author, download count, and last update, and click install to add it to your account.

# 搜索市场
curl "https://braidrun.com/api/skills/market/search?query=testing&limit=10" \
  -H "Authorization: Bearer <token>"

# 安装（安装状态只记在你自己的账号下）
curl -X POST https://braidrun.com/api/skills/market/<slug>/install \
  -H "Authorization: Bearer <token>"

# 升级到最新版本
curl -X POST https://braidrun.com/api/skills/market/<slug>/upgrade \
  -H "Authorization: Bearer <token>"

• Skills carry version numbers. For installed skills, click Check for Updates on the Skill Management page; upgrading redownloads the latest version
• Only skills installed from the marketplace go through the upgrade flow; skills imported from ZIP have no upgrade source—just reimport
• Uninstalling only removes the skill from your own account and doesn't affect other users who installed the same skill
• You can also upload a ZIP to import a skill you wrote yourself: the package must contain a SKILL.md with valid name and description fields, and the total unzipped size must not exceed 100MB

Enable/Disable Per User

Each skill on the Skill Management page has its own toggle. Enabled/disabled state is recorded on your account: once you disable a system skill, it's no longer visible to your Agents, while other users are unaffected. Skills privately installed by other users are neither visible to you nor loaded for you.

Agent Scope: Limiting Which Skills an Agent Can Use

Beyond the account-level toggles, you can define a narrower skill scope for a single Agent in its configuration. Common fields:

• Skills subsystem toggle—turn it off and this Agent uses no skills at all
• Allowlist mode—when on, only allowlisted skills can be detected by this Agent, and built-in skills are added to the allowlist automatically
• Disable list—a denylist that still applies when allowlist mode is off
• Max skills per request—limits how many skills are injected into a single request
• Auto search and download—off by default; when on, the Agent automatically searches ClawHub and downloads new skills when it finds a capability missing locally. This introduces external downloads, so use it with caution in enterprise settings

At execution time, the platform merges the Agent configuration with your user-level state into the effective skill scope: a skill you disabled on the Skill Management page is unavailable to every Agent in that execution.

Runtime Configuration

Some skills need parameters or credentials filled in before they work, such as an API token for a third-party service. A skill with a config template shows Configured or Not configured on its card; click Configure Skill and fill in the fields.

• Sensitive fields (token, password, private key, and so on) are masked after saving and never shown in plaintext; submitting with the mask unchanged keeps the old value
• You can configure skills you installed yourself; only administrators can change the runtime configuration of system skills

Common Endpoints

Related Pages

• Agent Configuration Guide
• Built-In Module Library
• Public Artifact Sharing