Skip to main content
This guide shows you how to equip an AI coding agent to operate pcl, from creating an assertion project through deploying it with the Phylax platform. Phylax publishes reusable instructions in the agent-skills repository. These skills give compatible agents the project structure, safety boundaries, and pcl workflows needed to complete a task instead of relying on a large one-off prompt.

​
Prerequisites

Before you begin, make sure you have:
  • pcl installed
  • An agent that supports the Agent Skills format, such as Codex or Claude Code
  • A local protocol repository the agent can inspect and edit
  • Clearly defined protocol invariants
For deployment, you also need a target chain, an RPC endpoint, and access to the protocol manager wallet. Authenticate with the platform for your target network before starting the agent. See How to Deploy Projects Through Agentic PCL for the Linea and Ethereum commands.
An agent can edit code, create platform resources, and broadcast on-chain transactions. Review its plan and local changes before authorizing remote mutations. Require a successful dry run and explicit approval before any production deployment.

​
Install the Phylax agent skills

Install the two PCL workflow skills from GitHub:
The commit pin ensures that both commands install the reviewed skill versions shown in this guide. The agent-skills repository also includes focused skills for invariant design, assertion implementation, testing, backtesting, trigger optimization, and troubleshooting.
Restart your agent after installation if it does not discover the new skills in the current session.

​
Choose the workflow

Use the skill name in your request so the intended workflow is unambiguous.

​
Start with a high-level prompt

You can describe the intended outcome and let the skills determine the required pcl workflow. Include the protocol, target environment, authoritative sources, release contents, and verification boundary. For example:

​
Orient the agent with the installed CLI

Current versions of pcl expose machine-readable guidance for agents. Include this instruction in your prompt:
This prevents the agent from inventing flags or relying on a workflow from another pcl version.

​
Keep a human in the approval loop

Authenticate with the target platform yourself before starting the deployment agent. Then give the agent autonomy for inspection, local file changes, builds, and tests. Require it to pause for:
  • Any choice between staging and production
  • Access to a signer or keystore
  • Platform mutations
  • On-chain broadcasts
Do not paste a private key into the conversation or commit it to the repository. Prefer a Foundry keystore and provide its password through a file outside the project.

Create a Project with an Agent

Bootstrap and validate an assertion project locally

Deploy Projects Through Agentic PCL

Create a platform project and activate a release on-chain