@harnessa-fe/skill
Agent skill / playbook for using the Harnessa-FE MCP toolset. Distribute the same instructions to Claude Code, Cursor, Kiro, or any other MCP-aware AI agent.
This package is just data — a curated SKILL.md plus a tiny installer. It tells the agent when to invoke harnessa-fe tools, how to chain them for common debugging flows, and what the safety boundaries are.
Install
npx @harnessa-fe/skill install
npx @harnessa-fe/skill install cursor
npx @harnessa-fe/skill install kiro
npx @harnessa-fe/skill install plain
Refuses to overwrite. Delete the existing file first if you want to upgrade.
Inspect without installing
npx @harnessa-fe/skill print
npx @harnessa-fe/skill where
Programmatic use
import { SKILL_PATH, readSkill } from '@harnessa-fe/skill';
console.log(readSkill());
console.log(SKILL_PATH);
What the skill covers
- Mental model: project / build / tab / session — and how same-origin iframe identity inheritance works for micro-frontends.
- MCP tool catalog: page interaction, telemetry tails, rrweb replay, source intelligence, annotation tasks.
- Source-aware selectors: how to target elements by
comp (component) / loc (file:line) instead of CSS classes.
- Debugging decision flows: visual bugs, network bugs, micro-frontend bugs, post-crash forensics.
- Safety constraints:
page_evaluate is arbitrary JS; project_source is sandboxed; rrweb captures may contain secrets.
See the file itself for the full content: npx @harnessa-fe/skill print.
Prerequisite
The agent's host project must have the harnessa-fe MCP daemon configured. Minimal setup:
npm i -D @harnessa-fe/vite @harnessa-fe/runtime
import { harnessaFE } from '@harnessa-fe/vite';
export default defineConfig({ plugins: [react(), harnessaFE()] });
Then in .mcp.json (or your agent's MCP config):
{
"mcpServers": {
"harnessa-fe": { "command": "npx", "args": ["-y", "@harnessa-fe/mcp-server"] }
}
}
License
MIT