dont-ask-melisted

Get a second opinion from a different model family when the user wants to validate or stress-test Claude's answer before trusting it. The skill is user-triggered (not automatic) — Claude activates it when the user signals uncertainty or explicitly asks for a cross-check. Use when the user says any of these (or semantically similar): SECOND-OPINION TRIGGERS — 'second opinion', 'give me a second opinion', 'sanity check', 'sanity check this', 'cross-check', 'cross-check this', 'am I missing something', 'stress-test this', 'review this', 'critique this', 'devil's advocate', 'thoughts?', 'is this right?', 'poke holes in this', 'what could go wrong', 'another perspective please'. FULL-REVIEW TRIGGERS — 'this is important', 'run a full review', 'big decision', 'high-stakes', 'high-stakes review', 'check before I send', 'before publishing', 'boardroom debate', 'dual validation', 'two independent opinions', 'don't let me ship something dumb', 'this can't be wrong', 'double-check this'. ROUND-TABLE TRIGGERS — 'help me
awrshift/skill-dont-ask-me · ★ 3 · AI & Automation · score 79

Install: claude install-skill awrshift/skill-dont-ask-me

# Don't Ask Me Cross-check Claude's own answer with a second AI before bringing it to the user. Three review styles, picked automatically by what the user typed. Two model families (Gemini and an isolated Claude Opus subagent) catch different blind spots than one model alone. ## When to invoke Four trigger conditions. If any of them fires, pick the matching style below. 1. User has two or more viable paths to choose between (architecture options, technology choice, content angle) — use Round-Table Discussion. 2. High-stakes design decision (ADR, architecture, launch copy, anything where rollback costs something real) — use Boardroom Debate as the default. 3. User stuck on the same problem after two or more attempts (failed fix, failed retry, about to add another flag or condition) — use Devil's Advocate. 4. Non-trivial proposal awaiting user approval (new module, refactor plan, multi-step migration) — use Devil's Advocate or Boardroom Debate depending on stakes. If none of these fires, don't invoke the skill. A single-model answer is enough for routine work. Out of scope: factual lookups like "what's the latest X version" or "is Y still free". Use Claude Code's native WebSearch for those. This skill is for critique and adversarial validation, not fact retrieval. ## The three review styles ### 1. Devil's Advocate (single critique) Trigger phrases: "sanity check this", "am I missing something", "stress-test this", "critique this", "give me a second opinion". Two imple