Anti-Sycophancy: Evidence-First Review

    1

    Pressure-test claims, plans, and decisions with explicit evidence, assumptions, uncertainty, and decisive counterarguments. Agrees only when evidence supports it.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Anti-Sycophancy: Evidence-First Review

    Example session with this skill installed

    Use anti-sycophancy to review this plan. We should migrate our PostgreSQL billing database to MongoDB this week. Last month CPU peaked at 85%, so PostgreSQL cannot scale. Do not just agree.

    • Read your context and instructions
    • Compiled the anti-sycophancy

    Verdict

    Do not approve the migration yet. The available evidence does not establish that PostgreSQL is the bottleneck or that MongoDB would resolve it.

    ClaimStatusReason
    PostgreSQL cannot scaleUnsupportedOne CPU peak does not prove a scaling limit.
    The CPU peak caused the observed problemUnknownNo symptom, query profile, or correlation was provided.
    MongoDB fixes the issueAssumptionNo measured failure mode links the workload to the database engine.
    Migration can safely land this weekHigh riskNo data-volume estimate, rollback plan, or consistency test is provided.

    Verified

    • The request proposes a one-week PostgreSQL-to-MongoDB migration.
    • The only supplied metric is a CPU peak of 85%.

    Assumptions that need testing

    • The system problem is CPU-related.
    • PostgreSQL is responsible for that CPU load.
    • MongoDB eliminates the specific bottleneck.
    • Billing workflows do not depend on features that the migration would weaken.

    Strongest counterargument

    Billing data often depends on transactions, relational constraints, joins, and auditability. A fast database migration can replace a manageable performance issue with a much larger correctness and reconciliation risk.

    Failure conditions

    1. CPU spikes come from application code, missing indexes, or connection churn rather than the database engine.
    2. The migration loses transactional guarantees required by billing operations.
    3. A one-week migration leaves too little time for production-scale validation.
    4. No tested rollback path exists.

    Decision

    Pause the migration. Run 24-48 hours of database and query profiling. Identify the slowest and most frequent queries, check connection-pool behavior, and list the billing workflows that require transactions.

    Evaluate lower-risk fixes first: indexes, query rewriting, connection pooling, caching, read replicas, and capacity changes.

    Confidence: Medium. The risk of migrating without a measured bottleneck is clear; the best alternative depends on missing workload data.

    Evidence that would change this conclusion: query-level profiling showing an irreducible database-engine limitation, plus a migration design with correctness tests, a production-scale rehearsal, and a tested rollback plan.

    Next step: collect the missing telemetry and compare at least two alternatives before approving development work.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Anti-Sycophancy helps an agent challenge weak reasoning without becoming performatively contrarian. Use it for technical reviews, high-stakes decisions, fact-checking, red-team analysis, and situations where repeated user pressure could distort a conclusion.

    What the agent changes

    • Rewrites loaded claims as neutral questions when needed.
    • Separates verified facts, supported-but-insufficient evidence, assumptions, unknowns, and contradicted claims.
    • Finds the strongest relevant counterargument, hidden dependency, and failure condition.
    • States the conclusion first, then explains material disagreements.
    • Calibrates confidence and identifies what evidence would change the result.
    • Updates only when new facts, arguments, or constraints justify it.
    • Preserves valid agreement instead of manufacturing objections.

    Best for

    • Architecture and migration reviews
    • Strategy and feasibility checks
    • Factual claim audits
    • Decision pre-mortems and red-team review
    • Evaluating competing options under uncertainty

    Boundaries

    The skill does not turn personal taste, creative preference, or emotional support into adversarial debate. High-stakes outputs still require appropriate external sources and qualified human review.

    Requirements

    A compatible agent that supports SKILL.md. No scripts, APIs, additional services, network access, or credentials are required.

    Source: https://github.com/woaiyrb/anti-sycophancy

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.

    Not intended for personal taste, creative preference, or emotional-support conversations. Accuracy is limited by the evidence and context available. High-stakes conclusions still require external sources and qualified human review.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 11 days ago

    • Free with an Agensi account

    Permissions

    No special permissions declared or detected

    Listed11 days ago

    What's inside

    Frequently Asked Questions

    It requires each conclusion to be supported by evidence, reasoning, and task constraints. Agreement is kept when warranted; disagreement is stated only when a specific claim or assumption is unsupported.

    It requires an agent that supports the SKILL.md standard. It is designed for Codex and also works with compatible agents such as Claude Code, Cursor, and other SKILL.md tools.

    A conclusion changes when new facts, arguments, or constraints materially alter the evidence. Repetition, disappointment, or social pressure alone are not treated as evidence.

    Yes. This listing is free, and the source is publicly available on GitHub. No paid upgrade is required.

    Do not use it to turn personal taste, creative preference, or emotional-support conversations into adversarial debate. High-stakes conclusions still require appropriate external sources and qualified review.

    New in Prompt & Skill Engineering

    Free