severity1/claude-code-prompt-improver v0.5.4 on GitHub

Changes

Second UserPromptSubmit hook (scripts/workflow-guidance.py) that injects model-routing and plan-mode-first HITL guidance when a dynamic workflow is requested
Triggers on the workflow/workflows keyword (non-slash prompts), the /deep-research command, /effort ultracode, and saved workflow commands under .claude/workflows/ (cwd) or ~/.claude/workflows/
Guidance is model-agnostic: reserves the session model for planning, strategy, and orchestration and routes implementation to a smaller, cheaper model; /effort ultracode appends a clause applying the routing session-wide
Test suite (tests/test_workflow_guidance.py) covering trigger detection, the * and # bypass prefixes, the /workflows no-output case, the conditional guard, path-traversal safety, empty/invalid stdin, and the guidance token budget

# (memorize) prefix now bypasses the workflow-guidance hook, mirroring improve-prompt.py, so a prompt like "# remember the deploy workflow doc" no longer false-triggers
Natural-language keyword search restricted to non-slash prompts so the /workflows run-management command no longer false-triggers (slash prompts keep explicit /deep-research, /effort ultracode, and saved-workflow detection)
CORE_GUIDANCE leads with a conditional ("If this prompt will run as a dynamic workflow...") so broad keyword matches like "fix the CI workflow file" self-cancel at the model rather than injecting inert guidance

Plugin registers two UserPromptSubmit hooks (improve-prompt.py and workflow-guidance.py) alongside the PreToolUse plan-guidance hook
Bumped plugin version to 0.5.4

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