forsy-ai/forsy-trace-skill

Forsy Trace Skill is a small open source tool for recording what an AI agent did while completing a task, not just the final result. It captures the whole process as a structured file: what the agent was asked to do, what tools and context it had, each action it took, what it observed afterward, where it failed or retried, what feedback it got, and what it eventually produced. The idea behind it is that looking only at an agent's finished output leaves out most of the useful information. If you also have a record of the steps, tool calls, mistakes, and corrections along the way, that record can be studied, checked for quality, used to train future agents, or reused as a kind of memory of past work. You install it with a single command that copies a skill file and a JSON schema into your own project. The skill file, skill.md, is meant to be handed to an AI agent as instructions for producing a properly structured trace of a workflow it just completed. The schema folder defines exactly what fields a trace file should contain. The project also ships a set of example traces covering different kinds of work, such as scientific computing, legal research, and hardware planning, plus scripts for checking that a trace file matches the schema and for exporting traces into line delimited JSON files for further analysis. A typical use looks like this: add the skill file to an agent's environment, ask the agent to do or redo a piece of real work, save the resulting trace, validate it against the schema, and then export it alongside other traces for review or research. The installer only copies files onto your machine. It does not call any outside service, run anything on its own, or send your data anywhere. This project is an early, evolving version of a trace format the maintainers plan to build a larger platform around, for exchanging real world agent work experience. It is released under the MIT license.