Mid-Conversation Direction Check
Reassess a stalled conversation against the real goal, expose weak assumptions, and recommend whether to continue, adjust, or stop.
Before & after
Same prompt, same agent — the only difference is the skill.
That's a good question to step back and ask. It sounds like you're feeling some doubt about the direction. Let me help you zoom out and think about this more strategically. I'd want to understand what's making you feel like it might be overcomplex, what the original goal was, and whether there are simpler paths to the same outcome. Sometimes when we're deep in the details, we lose sight of the bigger picture. What specifically is making you question the approach? That would help me give you a better sense of whether we should stick with the plan or explore alternatives.
STRATEGIC REASSESSMENT FRAME CHECK • Original problem: Reduce checkout abandonment by 15% • Approach chosen: Full checkout flow rebuild • Question: Is flow rebuild the minimum viable path? ASSUMPTION AUDIT • Assumption 1: Current UX is the primary abandonment driver → Evidence: User testing showed friction at payment entry (3/5 dropped) → Risk: Mobile users abandon after shipping calc, not payment • Assumption 2: Rebuild is faster than optimization → Reality: 8 weeks rebuild vs. 2 weeks targeting payment step BIAS CHECK • Sunk-cost trap: We've already sketched a full design • Complexity bias: Sophisticated solution feels more professional • Feature-flag phobia: Could A/B test payment-step changes on 10% first SIMPLER PATH • Week 1: Deploy payment-entry copy test (reduces cognitive load, 2-4% uplift) • Week 2: Add one-click payment option for repeat users (historical CTR 22%) • Week 3: Measure and decide if 8-week rebuild is needed RECOMMENDATION: Reverse the sequence. Test low-risk changes first; rebuild only if quick wins plateau.
About this skill
“We've gone deep, but are we actually getting closer?” This skill pauses detail work, reconstructs the current goal, and checks whether the chosen approach still deserves more effort. You get a direct recommendation grounded in the conversation, including when the right answer is to keep going.
What it does
- Separates the desired outcome from the implementation that has started to dominate the conversation.
- Checks five dimensions: goal, approach, assumptions, outsider perspective, and the future cost of continuing versus changing course.
- Distinguishes confirmed constraints from assistant suggestions that were never validated.
- Identifies observable signs of sunk cost, anchoring, scope drift, or premature optimization without inventing motives.
- Resolves consequential uncertainty with one bounded test and an explicit decision rule.
- Recommends continuing, adjusting, or abandoning the current path, with one concrete next action.
What you get
A concise direction check with a verdict, supporting evidence, the assumption most likely to change the decision, and the next useful move. Larger reviews include a compact comparison of the current path and a credible alternative.
Best for
Builders, writers, operators, and teams working through long Claude conversations where implementation detail, repeated revisions, or mounting effort may have displaced the actual goal.
SKILL.md preview
---
name: mid-conversation-direction-check
description: Reassess an ongoing conversation through five lenses: goal, approach, assumptions, outsider perspective, and future cost. Separate confirmed facts from accumulated suggestions, then recommend continuing, adjusting, or abandoning the current path with one actionable next step. Use when the user says "reflect", "take a step back", "are we on track?", "are we missing something?", "we keep going in circles", "are we overthinking this?", "this is taking forever", or "sanity-check our approach".
---
# Mid-Conversation Direction Check
… (sign up to view the full skill)Featured in
More productivity skills
View all Productivity skills →Apple-Style UI Engineer
Use when building any new UI component, screen, or page in React + Tailwind from a plain-text description, and the result must look like App
Zapier, Make and n8n Automation Recipe Builder
Designs no-code automations that run unattended: exact trigger, mapped fields, dedupe, error alerts, approvals, and monthly task cost.
AI Productivity System
A complete AI-driven system for planning your day, triaging tasks, and shipping more without burning out.
Daily Standup Writer from Git and Tickets
Turns your commits, PRs, reviews, and ticket moves into a 15-second standup post that never says shipped when it only merged.