All products

Write project instructions for AI agents

Turn project-specific decisions and verified commands into a shared instruction file that a new contributor can use on a real task.

Write project instructions for AI agents around the things a new colleague cannot infer from a directory listing. Perhaps a test command starts a service that sends mail, or the public pages are built separately from the admin tools. The project-memory skill helps document the conditions that matter before someone changes the wrong part.

Start by identifying the instruction file that already governs the project. If another client imports it, keep the decision in the canonical file. A written convention gives the next agent context, while an enforced rule requires a separate mechanism that can actually block an action.

What it does

  • Record working commands alongside their scope.
  • Separate shared instructions from directory-specific exceptions.
  • Capture the reasoning behind a project decision.
  • Identify competing copies of the same guidance.
  • Use an optional shared issue ledger when authorised collaborators need one.

Install and use

  1. Give a file-capable agent authorised access to the project, its current instructions and the decisions to preserve. A GitHub ledger is optional and requires repository permission.
  2. Ask, “Make a handover for someone changing the storefront. Which test commands apply, and what would they need to know about the separate deployment?”
  3. Check the instructions against that task and correct statements that no longer match the project.

FAQ

Source

Made by Joe Amditis. Original skill: project-memory, published on GitHub under the MIT licence.

View on GitHub