“I'm an OpenClaw user. What can intouch do for me?”
One question, asked of a stock install. Nothing here is staged copy — this is the answer the assistant gave, played back.
Good question — since you already think in skills and agents, here's the direct translation:
What you already have: SKILL.md files on ClawHub — YAML frontmatter + markdown instructions that an AI agent re-reads and improvises every single run, paying LLM tokens each time.
• Discover & install from ClawHub — search, preview the SKILL.md before it lands, install it. Skills land at INTOUCH_HOME/skills/<name>/SKILL.md on your own server.
• Convert it — this is the actual point. I read the skill once and generate *native* automation instead of running the prompt raw:
• One reusable operation → an IML tool
• A multi-step pipeline → a Workflow File
• An AI step only survives where reasoning is genuinely still needed
After conversion it runs deterministically — same steps every time, zero per-run model cost.
• Credentials stop being shell exports. Anything in the skill's requires.env that's a secret becomes an encrypted vault credential, injected at execution, never printed or put in my context. Everything else becomes a runtime env var.
• It gets a schedule and a watchdog. Converted automation can run on a schedule (7 types), fire off a trigger file, or sit inside a monitor with conditions + alert actions. A skill that used to need you to invoke it manually now runs at 6am and only speaks up when something's wrong.
• Full run history — every execution logs inputs, outputs, duration, and result code. That's the difference between "a clever agent did something" and something you can actually trust unattended.
Pattern mapping (how your skill's guts become native tools):
It answered like this because it knew who was asking.
The same assistant answers a shop owner about supplier price files and a sysadmin about cron sprawl. Same engine, same tools, same audit trail — sharpened to the person asking.
The same assistant, asked by someone else.
Every one of these is a stock install answering a different person’s first question. Nothing was staged and nothing was edited.