Key concepts and terminologies for Patched
ReadPR
step reads a Pull Request from a repository, and the CallLLM
step prompts an LLM endpoint with the given input.
Steps are the building blocks for creating automations in Patched. They abstract away implementation complexity and provide a simple interface to carry out an operation. This does not mean that steps are limited in functionality or flexibility - they can be modified, extended, or recreated as needed.
Fix this code which is breaking my build: {{code}}
is a prompt that asks the LLM to generate a fix in a code snippet that is provided dynamically based on the code
provided in the step’s input.
Prompts are critical to create high-quality, reliable automations. They provide a way to execute an infinite array of context-aware tasks, including code generation, reasoning, data transformation, and more. The best results are achieved when prompts are well-defined, clear, and concise - with necessary context and examples.
PRReview
patchflow automates the process of reviewing a Pull Request, and the Generate Docstrings
patchflow automates the process of generating docstrings for a repository.
Patchflows are a directed graph that represents the flow of data between steps. Each node in the graph represents a step, and each edge represents the data flow.
PRReview
creates a comment on the target Pull Request with the review details, and running Generate Docstrings
creates a Pull Request with the generated docstrings. Each run is considered a patch
.
Patches are the atomic unit of execution in Patched. They are created by running a patchflow with the required input data. Patches can be triggered manually, scheduled, or triggered by external events like webhooks or API calls.