AI Agent Hub
Back to plugins
🤖

vibe-intent-compiler

Model Inference Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Shhaaawwww/vibe-intent-compiler

Paste the following prompt into your AI chat to install this plugin:

Install the plugin by running the command in your DeepSeek Harness terminal; the project source is available at https://github.com/Shhaaawwww/vibe-intent-compiler .

About this plugin

In the DeepSeek Harness composer, a vibe coder's draft is often fragmented, self-correcting, and full of ambiguous references. Sending it as-is risks losing critical constraints or having the model fabricate implementation details that were never mentioned. vibe-intent-compiler addresses exactly this: it reads the full unsent draft, reorganizes only what the user actually said, and distills it into a concise, faithful, actionable instruction—never inventing project facts, files, or acceptance criteria.

Key capabilities include preserving protected text such as code, commands, paths, URLs, and identifiers, with deterministic post-rewrite validation that leaves the original draft untouched if verification fails; resolving vague references by consulting up to three explicitly selected files without searching the project or launching an agent loop; keeping short inputs short with output capped at roughly twice the source length; and never auto-executing tasks or sending the message.

It is built for developers who work in a vibe-coding workflow inside DeepSeek Harness: write a natural, even contradictory draft, hit Clarify, review and tweak, then send it yourself.

Use Cases

  • Distill fragmented, self-correcting drafts into concise, actionable instructions
  • Resolve vague references via up to three picked files without project search
  • Deterministic post-rewrite validation of protected text; original draft stays if it fails

Best For

  • Developers using the vibe-coding workflow inside DeepSeek Harness
  • Prompt writers who want to draft naturally, then refine before sending
  • Engineers who refuse to let models fabricate implementation details or acceptance criteria