challenge
Use when questioning claims, pushing back on assumptions, sanity-checking decisions, evaluating confident assertions, avoiding reflexive agreement, or answering prompts like "are you sure?"
Best use case
challenge is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Use when questioning claims, pushing back on assumptions, sanity-checking decisions, evaluating confident assertions, avoiding reflexive agreement, or answering prompts like "are you sure?"
Teams using challenge should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/challenge/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How challenge Compares
| Feature / Agent | challenge | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Use when questioning claims, pushing back on assumptions, sanity-checking decisions, evaluating confident assertions, avoiding reflexive agreement, or answering prompts like "are you sure?"
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Challenge
Prevents reflexive agreement by forcing structured critical reassessment. References the `perspectives` skill for its critical thinking framework.
<workflow>
## 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.
</guardrails>
<validation>
### 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...")
</validation>
<example>
## 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.
</example>
## References Index
- **[Challenge Strategy](references/challenge-strategy.md)** — Five-step challenge workflow
- **[Critical Thinking Framework](../perspectives/references/critical-thinking.md)** — CRITICAL REASSESSMENT pattern (from perspectives skill)Related Skills
flow-memory-keeper
Use at task, phase, flow, sync, archive, finish, revise, or failure checkpoints to keep Flow specs clean, capture learnings and failures, elevate durable patterns, and refine this skill with project-specific nuances
vue
Use when editing Vue projects, .vue files, vue.config.js, Vue 3 components, Composition API, <script setup>, SFC state, deployment workflows, or Vue CI configuration.
vite
Use when editing Vite projects, vite.config.ts, vite.config.js, Vite plugins, HMR, asset bundling, frontend build settings, deployment config, or Litestar/Vite integration.
uvicorn
Use when deploying ASGI apps with uvicorn, editing uvicorn CLI commands, Config or Server usage, workers, reload, event loop selection, SSL, lifespan, logging, or development server behavior.
tracer
Use when tracing execution paths, mapping dependencies, understanding unfamiliar code, following data flow, investigating end-to-end behavior, debugging call chains, or deciding which files to read next.
testing
Use when writing or refactoring tests, editing test_*.py, *.test.ts, *.spec.ts, conftest.py, vitest.config.ts, pytest fixtures, mocks, coverage, async tests, anyio, or test failure debugging.
terraform
Use when creating, adopting, refactoring, or operating Terraform, *.tf files, .terraform.lock.hcl, terragrunt.hcl, root modules, backends, state, workspaces, imports, CI plan/apply, tests, or policy checks.
tanstack
Use when editing TanStack code, @tanstack imports, useQuery, createRouter, React Query, TanStack Router, Table, Form, Store, file-based routing, data fetching, or SPA state management.
tailwind
Use when styling with Tailwind CSS, editing tailwind.config.ts, tailwind.config.js, @tailwind directives, utility classes, responsive layouts, @apply, cn(), @theme config, dark mode, or forms.
svelte
Use when editing Svelte components, .svelte files, svelte.config.js, Svelte 5 runes, $state, $derived, SvelteKit, component state, or migrating away from Svelte 4 patterns.
sqlserver
Use when writing T-SQL, editing SQL Server .sql files, using sqlcmd, SQL Server connection strings, stored procedures, execution plans, indexes, Always On, JSON, security, or connector code.
sqlalchemy
Use when editing SQLAlchemy code, sqlalchemy imports, mapped_column, DeclarativeBase, ORM models, relationships, select() queries, async sessions, engines, events, or migrations.