Skip to content

add: AI-assisted prompt-based context selection #4

@akadeepesh

Description

@akadeepesh

What kind of feature is this?

New CLI flag or option

What problem does this solve?

Currently, users must manually decide which files should be included in an AI context package.

This becomes difficult in medium or large repositories where a simple change request may involve multiple related files across frontend, backend, shared components, utilities, or configuration layers.

Example workflow today:

contextzip src/components/Login.tsx src/lib/toast.ts

The user has to manually discover and select relevant files before asking an AI model for help.

This creates friction in AI-assisted development workflows and often results in:

  • missing context
  • oversized context packages
  • unnecessary token usage
  • repeated retries with AI models

The problem becomes more noticeable when the requested change is described in natural language instead of explicit file paths.

Example:

contextzip --prompt "Change toast color on failed login"

The tool currently has no way to infer which files are likely related to the request.

Proposed solution

Introduce a new prompt-aware mode that uses lightweight AI-assisted relevance selection to automatically determine which files should be included in the generated context package.

Example usage:

contextzip --prompt "Change toast color on failed login"

Proposed workflow:

  1. Scan the project structure

  2. Build a lightweight project map using:

    • file names
    • directory structure
    • optional import relationships
  3. Send the project map and user prompt to a lightweight/local AI model

  4. Receive a ranked list of likely relevant files

  5. Automatically generate:

    • context zip
    • prompt.txt
    • optional metadata

Example selected files:

app/login/page.tsx
components/ui/toast.tsx
lib/auth.ts
server/auth/login.ts

Initial implementation can remain intentionally simple:

  • filename-based relevance matching
  • directory semantics
  • framework-aware heuristics

Future improvements may include:

  • import graph analysis
  • semantic embeddings
  • git-aware weighting
  • AST/symbol relationships
  • local/offline model support

The goal is not to solve the coding task itself, but to intelligently reduce the amount of manual context selection required before sending code to ChatGPT, Claude, or other AI systems.

If this is a new framework — what should be excluded?

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions