challenge
Auto-activate when critically questioning claims, pushing back on assumptions, disagreeing with an approach, sanity-checking contentious decisions, or when a response feels like reflexive agreement. Produces an honest assessment — confirms what holds up with evidence, identifies
What it does
Challenge
Prevents reflexive agreement by forcing structured critical reassessment. References the perspectives skill for its critical thinking framework.
Workflow
1. Identify the Claim
Extract the core assertion being made. Strip away qualifiers and framing to find the actual claim. If there are multiple claims, address each separately.
2. Apply CRITICAL REASSESSMENT
Using the framework from perspectives/references/critical-thinking.md:
- Is it accurate? Check facts, verify assumptions against actual code/docs/data
- Is it complete? Are there missing considerations, edge cases, or perspectives?
- Is it well-reasoned? Does the logic hold, or are there gaps between evidence and conclusion?
3. Investigate
Do not reason from memory when you can verify — if a claim is about code, read the code; if about an API, check the docs.
- If the claim is about code: read the code
- If the claim is about an API: check the documentation
- If the claim is about performance: look for benchmarks, APM data, or profiling evidence
- If the claim is about best practices: check if the practice applies to this specific context and constraints
4. Deliver Honest Assessment
- If you find flaws: explain them clearly with specifics. Say what's wrong and why.
- If the reasoning holds: explain why it holds up. "I checked and this is correct because..." is more valuable than "I agree."
- If it's partially right: say which parts hold and which don't.
- Stay focused and to the point. No hedging, no padding.
5. No Meta-Commentary
Just present the analysis directly. Do NOT say things like:
- "Let me challenge this for you"
- "Playing devil's advocate here..."
- "I'll now critically evaluate this statement"
- "That's an interesting point, but..."
- "Great question! Let me push back on that"
The user knows they asked for a challenge — they don't need narration.
</workflow> <guardrails>Guardrails
- No hedging. Do not say "on one hand / on the other." Present the analysis directly with a clear verdict.
- No meta-commentary. Do not narrate what you are doing. Just do it.
- No sycophantic framing. Do not soften the verdict to avoid disagreement.
- Verify before asserting. If you can check it, check it. Do not reason from memory when evidence is available.
Validation Checkpoint
Before delivering the assessment, verify:
- Claims were verified against code/docs, not reasoned from memory alone
- Assessment is direct — no hedging, no "on one hand / on the other"
- If the claim holds up, explanation includes specific evidence why
- No meta-commentary slipped in ("Let me challenge...", "Playing devil's advocate...")
Example
Challenge: "We should rewrite the auth system in Rust for performance."
- Claim identified: Auth system is a performance bottleneck that Rust would solve.
- Investigation: Checked APM data — auth endpoint averages 12ms, well within SLA. Bottleneck is actually the user lookup query (340ms).
- Verdict: Claim doesn't hold. The auth system isn't the bottleneck. Rewriting in Rust would add complexity without addressing the actual performance issue. Recommend: optimize the user lookup query instead.
References Index
- Challenge Strategy — Five-step challenge workflow
- Critical Thinking Framework — CRITICAL REASSESSMENT pattern (from perspectives skill)
Capabilities
Install
Quality
deterministic score 0.46 from registry signals: · indexed on github topic:agent-skills · 11 github stars · SKILL.md body (3,583 chars)