PromptZip

Stop pasting your entire codebase into AI tools. PromptZip picks only what matters — and compresses it.

You drop in a Jira ticket, bug report, or task description. PromptZip scans your local repository, finds the files that actually matter for that task, and produces a tight, context-aware prompt ready to paste into Copilot, Claude, GPT, or any coding assistant.

No bloat. No guesswork. Fewer tokens, better answers.

Runs locally. No source code is sent externally.

Why It's Worth Installing

Most developers paste raw task descriptions or entire files into AI tools — burning tokens on context that doesn't help. PromptZip does the triage for you: it reads your repo, understands what the task needs, and hands you a compressed prompt that fits within limits and stays on target.

Install

Install from the Marketplace or run:

code --install-extension architect-dhruv.prompt-compiler

Features

• Local repository indexing for common code and config file types
• Token counting with js-tiktoken using cl100k_base
• Jira/story/bug analysis for title, requirements, constraints, domain, task type, and complexity
• Conservative context selection that rejects weak matches instead of sending random files
• Low-confidence clarification loop that asks for missing paths, modules, symbols, or expected behavior before producing the final prompt
• Explicit file-path and symbol detection from the work item
• Repository convention detection from files such as AGENTS.md, CLAUDE.md, Copilot instructions, Cursor rules, package scripts, and test/build config
• Import-aware context expansion for nearby dependencies
• Reverse dependency and likely test-file expansion for safer prompts
• New-feature strategy inference with suggested files, pattern files, and wiring candidates
• Structured prompt packages with accuracy rules, selected-file rationale, and missing-context checks
• AI consumption forecast with rough prompt complexity, output-token, credit-tier, and credit-range estimates before using an AI assistant
• Interactive repository graph view with local import edges, search, and file details
• Repository graph summary with token, import, export, and symbol counts
• JSON export of the local repository graph
• Multi-root workspace folder picker
• Configurable include/exclude globs and file-size limits

Commands

Open the Command Palette and run:

• PromptZip: Index Repository
• PromptZip: Compile Prompt From Work Item
• PromptZip: Compile Prompt From Quick Task
• PromptZip: Show Repository Graph
• PromptZip: Export Repository Graph JSON

Typical Workflow

1. Open the repository you want to work in.
2. Run PromptZip: Compile Prompt From Work Item.
3. Paste a Jira ticket, bug report, story, or implementation note.
4. Review the selected files, confidence score, and extracted requirements.
5. Copy the optimized prompt into your AI coding assistant.

If PromptZip cannot identify strong local context, it opens a local clarification step and asks for more specific files, modules, symbols, commands, or expected behavior instead of producing an unreliable patch prompt.

Example

Input work item:

Implement Slack notifications following the existing notification service pattern.
Reuse current provider registration and add test coverage.

PromptZip output includes:

Strategy: extend-existing-pattern
Suggested new files: src/notifications/slackNotificationService.ts
Pattern files to follow: src/notifications/emailNotificationService.ts
Wiring/integration files to inspect: src/notifications/notificationRegistry.ts
Test files or patterns to inspect: src/notifications/emailNotificationService.spec.ts

The final prompt package also includes selected source previews, file rationale, missing-context checks, and accuracy rules for the LLM.

Settings

Example:

{
  "promptCompiler.maxFiles": 5,
  "promptCompiler.maxFileBytes": 50000,
  "promptCompiler.promptMode": "compact",
  "promptCompiler.includeGlobs": [
    "src/**/*.{ts,tsx}",
    "docs/**/*.md"
  ],
  "promptCompiler.excludeGlobs": [
    "**/{node_modules,dist,coverage}/**",
    "**/*.generated.ts"
  ]
}

Privacy

PromptZip reads files from the active workspace and keeps the repository index in extension memory. It does not send repository content, prompts, or token metrics to any external service.

The PromptZip: Export Repository Graph JSON command writes metadata to .prompt-compiler/graph.json in the selected workspace. That export contains paths, token estimates, imports, exports, and symbols, but not full source contents.

Limitations

• Context selection is heuristic and local; it does not understand runtime behavior.
• Very large files are skipped according to promptCompiler.maxFileBytes.
• Generated prompt packages may still need human review before coding.