Skip to content

Version History

seasonalhawk edited this page Jun 14, 2026 · 4 revisions

Version History

Compatibility

Platform Phase Status
Claude Desktop for Mac Phase 1 Supported. Tested on 1.12603.1 (3df4fd), 2026-06-11.
Claude.ai Phase 2 Planned. Not yet tested.
Claude Code Phase 2 Planned. Pending ask_user_input_v0 confirmation in CLI contexts.
Claude (API) Phase 2 Planned.
VS Code / Cursor / IDE extensions Phase 2 Planned. Interface module under design.

Phase 1 is Claude Desktop for Mac only. All other platforms are planned for Phase 2. The engine requires ask_user_input_v0 to be available in the chat context. If the tool is unavailable, the engine stops and displays a clear error rather than falling back silently.


v4.0.0 (current)

Breaking change: Replaced the sequential one-question-per-call loop with a single batched ask_user_input_v0 call.

Changes:

  • All questions now appear together in one native Claude UI interaction
  • Users submit all answers in one action instead of cycling through questions one at a time
  • Added Part C integration guide for skill and plugin authors
  • Self-test updated to use the batched call
  • Sequential loop retired

Migration from v3: No skill-side changes required. The QUESTIONS array format is unchanged. The widget's execution model changed internally. If your skills used the old sequential flow, they now get the batched experience automatically on upgrade.


v3.0.0

Merged the engine and self-test into a single SKILL.md. Eliminated the need for a separate self-test file.

Changes:

  • Self-test QUESTIONS drawn from a randomized 20-question bank (retired in v4)
  • Single file install confirmed as the canonical model

v2.1.0

Removed the hardcoded 8-question cap. The engine now iterates dynamically over the full QUESTIONS array regardless of length.


v2.0.0

Breaking change: Replaced the show_widget HTML rendering engine with native ask_user_input_v0.

Changes:

  • No HTML required
  • Questions rendered as native Claude choice cards
  • One question per call (later retired in v4.0.0)
  • Faster, more reliable, no external rendering dependency

v1.1.0

Template embedded directly in SKILL.md to eliminate file-read failures at runtime. Mandatory no-fallback rule added: if the tool is unavailable, the engine stops and reports an error instead of falling back to prose.


v1.0.0

Initial release. Custom HTML widget rendered via show_widget. Question template stored as a separate file.


Copyright 2025-2026 Kenneth Benavides. MIT License.

Clone this wiki locally