Skip to content

Product roadmap

This roadmap describes direction, not release commitments. Stealth Prompt remains focused on authorized testing of selected AI interactions in a real browser session.

Current priorities

  • validate and extend the shipped FrameFuzz template pack with reproducible, authorization-safe fixtures and published compatibility results;
  • improve element detection and response capture across common chat interfaces;
  • make long-running tests faster, resumable, and easier to recover after provider or page failures;
  • expand deterministic evidence and make report retention and deletion clearer;
  • expand the local benchmark with reproducible authorized target fixtures;
  • improve provider compatibility, release packaging, and installation diagnostics;
  • publish a tested compatibility matrix for supported browser UI patterns.

Later

  • review automation for signed community strategy-pack pull requests;
  • trace and tool-call evidence through standard telemetry formats;
  • recorded multi-step flows for complex authenticated applications;
  • regression runs and CI-friendly report formats;
  • optional collaboration features that preserve local-first deployment.

Recently shipped

  • FrameFuzz matched framing campaigns with versioned templates, per-case fresh-context gates, Core and Direct execution, reproducible ordering, differential conclusions, report comparison, scenario schema v3, and vulnerable/safe local controls;
  • reviewed local learning, frozen evaluation, promotion gates, and signed strategy packs without automatic cloud ingestion.

Principles

  • The operator defines the target and authorizes every run.
  • Model output cannot create new browser operations or confirm a finding by itself.
  • Credentials and captured evidence stay local unless the operator explicitly chooses an external provider.
  • New capabilities must produce reviewable, reproducible evidence.

Feature requests and implementation proposals belong in GitHub Issues.