Improve propose skill with brainstorming-inspired discovery process

[HumanBean17]

Summary

• Add Hard Gate preventing writing proposals without discovery
• Add systematic 8-step process with task tracking requirement
• Agent determines clarifying questions dynamically (not scripted)
• Present 2-3 solution approaches with trade-offs
• Draft sections incrementally with approval
• Add self-review checklist after writing
• Add user review gate before completion

Why Now

The current propose skill lacks the questioning discipline that makes the brainstorming superpower effective. Agents never ask clarifying questions before writing proposals - they just write directly.

Approach

Combined the best of both skills:

• From brainstorming: questioning phase, task tracking, incremental validation, self-review, user gates
• From propose: repository-specific structure, quality bar, examples, templates, writing rules

Key insight: Questioning discipline over specific questions. The agent determines what clarifying questions to ask based on context - questions are NOT scripted.

Changes

• Updated .agents/skills/propose/SKILL.md with discovery process
• Added process flow diagram
• Added anti-pattern warning about "too simple to need discovery"
• Added 8-step checklist with task tracking requirement
• Expanded "The Process" section with step-by-step guidance
• Added self-review section with 4-point quality check
• Added user review gate before completion
• Kept all existing structure, examples, templates, and repository context

Test Plan

• Code review passes (dispatched subagent, awaiting results)
• Verify skill loads correctly in agents
• Test skill behavior on sample proposal task

🤖 Generated with Claude Code