skills.rest
legioncodeinc
legioncodeincOFFICIAL
Explore all skills by this author →
WebsiteGitHub

harness-integration-stinger

Wire capabilities across Claude Code, Cursor, Codex, and Cowork harnesses.

84|37|Updated May 23, 2026
Search Tags:#claude code#hooks#codex#mcp server#cursor#plugin manifest#cross-harness integration
One-click install
npx skills add https://github.com/legioncodeinc/vibe-coding-tools --skill harness-integration-stinger-legioncodeinc
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: harness-integration-stinger
Source: https://github.com/legioncodeinc/vibe-coding-tools/tree/main/src/skills/harness-integration-stinger
Command: npx skills add https://github.com/legioncodeinc/vibe-coding-tools --skill harness-integration-stinger-legioncodeinc

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Shipping one capability (a skill, agent, hook, or MCP tool) across multiple AI coding harnesses fails silently because each harness has different component locations, hook event surfaces, MCP config formats, and plugin manifests. This Skill encodes the per-harness integration decisions so a capability works correctly in Claude Code, Cursor, ChatGPT Codex, and Claude Cowork without copy-paste breakage. ## Core Features & Use Cases - Wiring-Mechanism Decision Framework: Choose between hooks, MCP servers, native extensions, or plain instruction files per harness using a decision matrix grounded in each harness's actual capabilities. - Per-Harness Reference Guides: Component placement tables, hook lifecycle event surfaces (including the five-event shared floor between Claude Code and Codex), MCP registration syntax (JSON vs. Codex's TOML mcp_servers), capability detection, graceful degradation, and spec-six portable skill frontmatter. - Worked Six-Host Case Study: A complete real-world integration (Hivemind across Claude Code, Codex, Cursor, Hermes, pi, OpenClaw) showing detection probes, idempotent wiring, tool contract stability, and distribution gates like ClawHub's static scanner. - Use Case: You need to register an MCP server across all four harnesses. The Skill shows the JSON form for Claude Code and Cursor, the TOML form Codex requires, and warns that Cowork connectors must be publicly reachable through Anthropic's cloud rather than localhost. ## Quick Start Ask the agent to wire your skill or MCP server into Claude Code and Cursor, or to audit whether an existing harness adapter handles capability detection and graceful degradation correctly.

Frequently Asked Questions about harness-integration-stinger

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I register an MCP server across Claude Code, Cursor, and Codex?▼

Claude Code and Cursor use JSON with an mcpServers key in .mcp.json or plugin manifests, while Codex requires TOML with an underscored mcp_servers key in config.toml. A JSON block copy-pasted into Codex silently fails with no parse error, so write the Codex TOML form as its own case.

What hook events work across different AI coding harnesses?▼

The verified shared floor across Claude Code and Codex is five events: SessionStart, UserPromptSubmit, PreToolUse (Bash-only on Codex), PostToolUse, and Stop. Claude Code exposes 26 events total, but 21 have no Codex equivalent, so design hook-driven capabilities around the five-event intersection first.

Does a Claude Code skill work in Cursor, Codex, and Cowork?▼

Yes, if the SKILL.md frontmatter uses only the six spec fields: name, description, license, compatibility, metadata, and allowed-tools. Claude-Code-only fields like hooks or paths cause hard packaging errors on claude.ai uploads, the Skills API, and Cowork's save pipeline.

Why does my MCP server fail to connect from Claude Cowork?▼

Cowork connectors run through Anthropic's cloud infrastructure, not the local network, so a localhost or stdio-only MCP server silently fails to connect. The server must be reachable over the public internet from Anthropic's IP ranges, unlike Claude Code, Cursor, and Codex which run locally.

How should a capability degrade when a harness lacks a feature?▼

Classify the gap first using preserve, translate, degrade, or drop at build time, or OK, DEGRADED, BLOCKED at runtime. For example, a commands-shaped capability compiles natively for Claude Code and Cursor but must be re-expressed as a skill for Codex, which has no commands surface.

When should I use hooks versus MCP versus a native extension?▼

Use hooks when the capability reacts to lifecycle events, MCP when the model needs to call a tool with a stable typed contract, and a native extension only when a dedicated UI surface is required. Default to a skill or rules file for everything else since it is the most portable path.

Related Skills

Explore More Software Engineering Skills→
unclenateunclenate
3

harness-mcp

Enforce governance-compliant MCP server behavior with tool tier mappings and threat documentation.

Community
Advanced
vellum-aivellum-ai
1.2k

plugin-builder

Scaffold, verify, and publish Vellum plugins bundling hooks, tools, skills, routes, and apps.

Official
Advanced
waybarrioswaybarrios
507

insecure-defaults

Detects fail-open insecure defaults like hardcoded secrets and weak authentication in application code.

Community
Intermediate
ayehiaaayehiaa

ruzy

Fuzz Ruby code and C extensions with ASan/UBSan instrumentation.

Community
Advanced