severity1/claude-code-prompt-improver v0.6.1

v0.6.1: User-decision and plan-mode nudges

on GitHub

Changes

This release adds two UserPromptSubmit nudges that cover decision axes the existing set did not: an explicit user-owned fork (a lexical signal, so keyword-gated) and task complexity (no lexical signal, so always-evaluate). Both are pure-data rows - one JSON file each, zero Python.

Added

• ask-user-question nudge (nudges/UserPromptSubmit/40-ask-user-question.json, priority 40): keyword-gated on decision/fork language (which/should/prefer/choose/decide/options/recommend/better/vs/trade-offs/pick). Routes genuine user-owned decisions through the AskUserQuestion tool with concrete options so the user can react and think critically, grounds questions in research when context is thin, and defers to a sensible default for minor or reversible choices. Self-cancels false positives via a leading condition
• plan-mode nudge (nudges/UserPromptSubmit/50-plan-mode.json, priority 50): non_slash-only criteria with no match patterns, so it always evaluates. Injects an eval-then-branch instruction - judge whether the task is complex, multi-step, ambiguous, or architectural enough to warrant a reviewed plan, enter plan mode if so, otherwise proceed. Complexity is not lexical, so it cannot be keyword-gated; the open gate evaluates every task regardless of wording while the engine's default bypass keeps */#/empty silent
• tests/test_engine.py gains a plan-mode pair (always-fires on a no-keyword prompt, silent under * bypass) and an ask-user-question pair (fires on decision language, silent on a plain prompt)

Changed

• approach-assessment nudge: trimmed the plan line from its guidance text and description so it owns "which approach" (subagent vs orchestration vs just do it) while plan-mode owns "whether to plan at all"; the two no longer overlap
• Bumped plugin version to 0.6.1

Documentation

• README synced to the full nine-nudge set: ask-user-question and plan-mode added to the capability table, architecture registry, and directory tree; the always-on framing reframed to two always-evaluate nudges (improve, plan-mode) plus seven conditional; the stale approach-assessment plan wording and priority example corrected

Installation

```bash
claude plugin marketplace add severity1/severity1-marketplace
claude plugin install prompt-improver@severity1-marketplace
```