Shortcut
A compact instruction selects the kind of work an agent should perform: review, fix, investigate, audit, security analysis, tests, documentation and other defined modes.
Project / tool
A tiny CLI for defining reusable context shortcuts and turning compact instructions into focused task contracts for coding agents, assistants and other AI-driven workflows.
ctxcuts avoids repeating long prompts and keeping oversized context blocks permanently active. Instead, small versionable task contracts live with the project and only the context required for the current task is loaded when needed.
A compact instruction selects the kind of work an agent should perform: review, fix, investigate, audit, security analysis, tests, documentation and other defined modes.
Each shortcut references a context file that explicitly defines the objective, constraints and expected working mode for that task.
Contracts can receive information from the invocation through small template variables, allowing the same definition to be reused across different targets and areas of work.
ctxcuts combines the shortcut, target and context contract into a complete prompt that can be consumed by different tools without depending on a particular provider.
ctxcuts deliberately stays outside the agent: it prepares context and lets each tool decide how to consume it.
Shortcuts and their context files live under .ctxcuts/ inside the project and can be maintained alongside the code.
A compact instruction such as :r src/app.py selects the contract and provides the specific target for the task.
The CLI renders variables and constructs the complete prompt only when that particular context is required.
Expanded output can be sent directly to other tools and agents through standard Unix pipelines.
doctor validates the setup and reports missing context files or suspicious context issues before they are used.
stats provides a lightweight estimate of how much reusable context is loaded from a compact shortcut invocation.
ctxcuts does not implement memory, embeddings, vector search or autonomous agent orchestration. Its responsibility is keeping context explicit, focused, reusable and easy to share or remove without coupling a project to a particular agent.