using-agent-harness

name	using-agent-harness
description	Use when starting any conversation — establishes skill discovery and requires Skill tool invocation before ANY response.
disable-model-invocation	true

If you were dispatched as a subagent to execute a specific task, skip this skill.

The Rule

Invoke relevant or requested skills BEFORE any response or action. Even a 1% chance a skill applies means invoke it to check.

Instruction Priority

1. User's explicit instructions (CLAUDE.md, AGENTS.md, direct requests) — highest priority
2. Agent Harness skills — override default system behavior where they conflict
3. Default system prompt — lowest priority

How to Access Skills

• Claude Code: Use the Skill tool. Never Read skill files directly.
• Copilot CLI: Use the skill tool (auto-discovered from plugins).
• Gemini CLI: Use the activate_skill tool.

Codex: subagent setup, environment detection, and App finishing notes in references/codex-tools.md.

Skill Selection

When multiple skills could apply, use this order:

1. Process skills first (brainstorming, debugging) — they determine HOW to approach the task.
2. Implementation skills second (design, mcp-builder) — they guide execution.

"Let's build X" → brainstorming first, then implementation skills. "Fix this bug" → debugging first, then domain-specific skills.

Skill Types

• Rigid (TDD, debugging): Follow exactly. Don't adapt away discipline.
• Flexible (patterns): Adapt principles to context.

The skill itself tells you which.

Anti-Rationalization

These thoughts mean STOP — you're rationalizing:

• "This is just a simple question" → Questions are tasks. Check for skills.
• "I need more context first" → Skill check comes BEFORE clarifying questions.
• "Let me explore the codebase first" → Skills tell you HOW to explore. Check first.
• "This doesn't need a formal skill" → If a skill exists, use it.
• "I remember this skill" → Skills evolve. Read current version.

User instructions say WHAT, not HOW. "Add X" or "Fix Y" doesn't mean skip workflows.