Build with an AI coding agent
Point Claude Code, Cursor or any coding agent at Plico and get a working agent project.
Plico is a folder of files with fast, key-free checks, which suits a coding agent: it can write the
files, run plico validate and plico eval, and read the result without a model key.
Give the agent the docs
https://plico.sohab.dev/llms.txtis an index of every docs page with a one-paragraph summary. Each page is plain Markdown.https://plico.sohab.dev/llms-full.txtis all of the docs in one file.- Any docs page as Markdown: request it with
Accept: text/markdown, or use the/llms.mdx/docs/<page>/content.mdURL thatllms.txtlists.
Start from the scaffold
npm create plico@latest my-agent writes an AGENTS.md into the project, and a CLAUDE.md that
imports it. It explains the layout, how to add a tool and an eval, and the commands that check
the work. Claude Code, Cursor and other agents read these files on their own.
Add the Claude skill
The repository has a plico skill for Claude Code. Install it into your project:
npx skills add ITZSHOAIB/plicoOr copy skills/plico from the repository
into .claude/skills/plico.
A prompt that works
Build a support agent with Plico (https://plico.sohab.dev/llms.txt). It needs a
lookup_ordertool and anissue_refundtool that waits for human approval. Add evals for the approved, denied and held cases. Runplico validateandplico evaluntil they pass. No API key is needed.
Example: a refund agent is what this produces.
What an agent should check
| Command | What it proves |
|---|---|
npx plico validate . | The layout, config, tool and skill names. |
npx plico eval . | Every eval case passes. Exit 1 on a failure or no cases. |
npx plico run --dry --script <file> . | One scripted run. Exit 2 means it is waiting for approval. |
npx plico readiness . | What is still missing, such as a model. |