# Harness

> Maintain the shared coding-agent prompt and skill layout, including canonical prompt storage, compatibility symlinks, and invocation metadata.

- Skill: `d-kimuson/harness` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add d-kimuson/harness`
- Raw SKILL.md: https://api.skillmd.com/api/skills/d-kimuson/harness/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: d-kimuson (https://skillmd.com/u/d-kimuson)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/d-kimuson/harness

---


# Harness maintenance

- Store shared prompts canonically in `.agents/prompts/`.
- Keep tool-specific prompt directories as symlinks to the canonical directory:
  - `~/.claude/commands` → `~/.agents/prompts`
  - `~/.pi/prompts` → `~/.agents/prompts`
- Unless a prompt is explicitly intended for the agent to invoke autonomously, set `disable-model-invocation: true` in its frontmatter.
- Preserve the prompt's user-facing invocation with `user-invocable: true` unless the prompt is intentionally hidden from users.
- Keep one canonical source; do not duplicate prompt files under tool-specific directories.
- End every prompt body with `$ARGUMENTS` on its own trailing line so user-supplied command arguments are always received. Never omit it, even for prompts that currently take no arguments.