Pick My Harness

Pick My Harness

Turns "which app should I use" into one named recommendation. A model is only half a local AI setup; this picks the app that gives it skills, MCP servers and your files.

Before you start

Run the check in references/environment-check.md first. These skills need a shell on the user's own machine. If the environment is a sandbox or container, stop and tell the user to run this in Claude Code on the computer they want to set up. Do not report detected specs from a sandbox: wrong specs are worse than none.

Steps

Track progress:

Task Progress:
- [ ] 1. Ask the four questions
- [ ] 2. Match against the matrix
- [ ] 3. Name one tool and its limits
- [ ] 4. Render the HTML report

1. Ask the four questions

Ask all four at once, as a short numbered list, in plain language. Never ask about VRAM here.

1. What is the main job: chatting with documents, writing code, general chat, or running automated tasks?
2. Is this just for you, or for a team who all need access?
3. How comfortable are you editing a config file if it unlocks more control? (happy / rather not)
4. Does everything need to stay on your machine with no cloud fallback?

If the user already answered some in their request, do not re-ask. Use what they gave.

2. Match against the matrix

Apply the decision rules in references/harness-matrix.md. That file holds the six tools, their real capability flags, and the tie-breakers.

3. Name one tool and its limits

Output exactly one recommendation, never a shortlist. Give: the tool, one sentence on why it won for their answers, its three capability flags, and one line on what it cannot do. Then give the single command or download link to get it.

If their answers make a second tool genuinely necessary alongside the first (most commonly LM Studio underneath something else), say so explicitly as a pair rather than presenting a choice.

4. Render the HTML report

Deliver the recommendation as a rendered HTML page, not as chat text. Build it from references/report-template.md using the pick-my-harness layout in section 4, save it to the Desktop, and open it. Keep the chat reply to two lines plus the file path.

Human checkpoints

Stop after step 3 and ask whether to proceed with installing it. Do not install anything from this skill. If they say yes, route to /local-ai-setup, or to /install-openwebui when Open WebUI is the pick. Open WebUI and Odysseus are the two browser harnesses and /local-ai-setup asks which one before installing, so route there rather than deciding for them.

Self-improvement

This skill is never finished. Improve it as you use it.

• Results are delivered as HTML pages. When the layout in references/report-template.md proves unclear for a real result, fix the layout there rather than working around it in chat.
• When a tool ships or loses a capability, update the flags table in references/harness-matrix.md and change its date line.
• When the user disagrees with a recommendation and explains why, add that as a tie-breaker rule in the matrix file.
• When a question turns out not to change the recommendation, delete it. Four is already the maximum.
• Keep this small: when you add something, cut anything that no longer changes behaviour.

Routing