Skip to content

The agent skill

tablewalk ships a skill in the Agent Skills format: a decision guide a coding agent loads when it builds or changes an App. It says which block fits an intent, how the page should look, which choices are hard to undo (data, access, secrets, public forms, environment), and the loop and rubric that decide when a page is done. Each rule links one line of a sample App. It points at tablewalk docs for the keys rather than repeating them.

From the App’s project, with tablewalk installed:

Terminal window
npx tablewalk agent init
Agent Reads --tool
Claude Code .claude/skills/tablewalk/ claude
Codex .agents/skills/tablewalk/ codex
Cursor .agents/skills/tablewalk/ cursor
GitHub Copilot .agents/skills/tablewalk/ (and .claude/skills/) copilot

With no --tool it writes both folders. It also adds a short, marked block to AGENTS.md naming the skill, for agents that read that file first; the rest of the file is left as it is. --dir <project> installs somewhere other than the current directory.

Run the same command after upgrading tablewalk. A copy it installed and nobody changed is updated, and one already current is said to be up to date. A copy the project has edited is left alone, the command says which files differ and exits 1; --force replaces it. Each copy records what was installed in .tablewalk-skill.json beside SKILL.md; commit it with the skill.

Ask the agent for the App in plain words (“an App for the clinic’s front desk: book a visit, check the patient in”). It loads the skill, settles the hard-to-undo choices with you, and then runs the loop in Build an App with a coding agent: scaffold, page by page with tablewalk check, tablewalk check --design and tablewalk shot, and a rubric answered on each screenshot.