OUTCOMES
What this Prompt helps you deliver
- The industry Agent has a clear responsibility
- Skills are reusable with acceptance examples
BEFORE YOU START
Before you start
- The industry process and human standard are clear
- Representative inputs and expected outputs are ready
What AI should inspect first
packages/apps.agents/skillsdocs/agent/tooling-skills.mddocs/agent-runtimeREADY TO RUN
Complete Prompt
Copy the complete Prompt with its execution and acceptance boundaries.
Turn industry expertise into TabTin Agents, Skills, and execution rules that complete real work with evidence and hand control to a person when required—not a chat persona. Read AGENTS.md, docs/agent/tooling-skills.md, docs/agent-runtime, .agents/skills, and packages/apps. Collect tasks, inputs, good/bad examples, evaluation, required approvals, and systems, then decide whether each capability is persona, Skill, App/connector, or ordinary workflow.
Define responsibility, prohibitions, context, tools, and handoff. Capture stable methods as small Skills with trigger, steps, input, output, verification, and failure behavior. Register tools through App manifest/current discovery rather than UI coupling. Keep human approval for high-risk writes, external messages, deletion, and payment. Record decisions, tool results, artifacts, and final evidence.
Run three representative and two failure cases against human standards. Missing context must trigger a question/stop, unavailable tools must produce a real blocker, and handoff must preserve context. Turn acceptance examples into repeatable tests or a harness. Report Agent/Skill/tool boundaries, locations, results, failure behavior, approvals, and missing domain material. Never place customer-private data in a public Skill or fixture.