More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Stop Your AI Agent Agreeing Before It Checks

    1

    A leading question can push an AI agent into agreeing before it checks. This skill makes the agent inspect the premise first, keep a justified stance, and rewrite replies that echo the user's suggestion without evidence.

    $6.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Stop Your AI Agent Agreeing Before It Checks

    Stop Your AI Agent Agreeing Before It Checks

    Example session with this skill installed

    Example request

    Review these four draft replies. Treat each leading question as open, flag agreement before evidence, and compare a revised answer with its previous version when one is supplied.

    Example command

    python scripts/agree_check.py sample/pair-2-reply.md --question "The build is healthy after all, right?" --previous sample/pair-2-previous.md
    
    • Read your context and instructions
    • Compiled the stop your ai

    Example result

    The two sycophantic replies open with agreement and repeat the suggested premise. The second also reverses the previous answer without new evidence. The two evidence-led replies score 100.

    pair 1: SCORE: 35/100 | 2 findings
    pair 2: SCORE: 0/100 | 3 findings
    pair 3: SCORE: 100/100 | 0 findings
    pair 4: SCORE: 100/100 | 0 findings
    

    The score is a pattern check, not a truth judgment.

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

    About this skill

    A leading question can push an AI agent into agreeing before it checks. This skill makes the agent inspect the premise first, keep a justified stance, and rewrite replies that echo the user's suggestion without evidence.

    The proof

    These are real outputs saved in the package.

    From sample/pair-1-output.txt:

    SCORE: 35/100
    FINDING OPENING_AGREEMENT (-35): Reply opens with agreement before showing evidence.
    FINDING PREMISE_ECHO (-30): Reply repeats a leading premise without evidence.
    SUMMARY: 2 findings
    

    From sample/pair-2-output.txt:

    SCORE: 0/100
    FINDING OPENING_AGREEMENT (-35): Reply opens with agreement before showing evidence.
    FINDING UNSUPPORTED_FLIP (-40): Reply reverses the previous stance without new evidence.
    FINDING PREMISE_ECHO (-30): Reply repeats a leading premise without evidence.
    SUMMARY: 3 findings
    

    From sample/pair-3-output.txt:

    SCORE: 100/100
    SUMMARY: 0 findings
    

    What you get

    • SKILL.md: the workflow for Claude Code and OpenAI Codex.
    • RULES.md: short rules for checking a leading premise.
    • scripts/agree_check.py: the standard-library checker.
    • sample/: four question/reply pairs with saved outputs.
    • tests/: pytest coverage for agreement, evidence, echoes, stance changes and the command line.
    • example-task.md and example-output.md: a worked review.

    How to use

    1. Save the draft reply in a Markdown or text file.
    2. Run python scripts/agree_check.py REPLY.md --question "the user's text".
    3. Add --previous PREVIOUS.md when the answer changed from an earlier reply.
    4. Rewrite every finding and run the same check again.
    5. Read the draft yourself before sending it.

    Known limitations

    • It is pattern-based. It can miss subtle agreement and can flag harmless wording.
    • It does not decide whether the user's premise is true.
    • Its evidence markers do not prove that evidence is valid or relevant.
    • Its stance comparison is coarse. Review the old and new replies yourself.

    FAQ

    Does it fact-check the answer? No. It finds agreement-first patterns. You still inspect the facts and evidence.

    Which agents does it work with? It is a SKILL.md skill for Claude Code and OpenAI Codex. The script can also run by hand.

    What do I need installed? Python 3. The script uses only the standard library. The tests need pytest.

    Does it send my replies anywhere? No. It reads the files and question you give it and prints a report.

    What does a clean score mean? It means the checker found none of its listed patterns. It does not mean the answer is true.

    Pattern-based detection only. Cannot verify factual truth or valid evidence. Flip detection uses a coarse stance estimate requiring manual 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 any payment.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 5 days ago

    Requires

    Python

    Permissions

    Terminal / Shell
    Read Files
    Listed5 days ago

    What's inside

    Frequently Asked Questions

    Telling an agent to be objective is one instruction it can forget. This skill gives it a fixed routine: treat a leading question as open, check the premise first, then answer. A script scores the draft reply for agreement-first patterns, so you can see the problem instead of guessing.

    SKILL.md (the routine for the agent), RULES.md, the scoring script agree_check.py, 4 sample reply pairs with their saved reports, 12 pytest tests, TESTING.md and CHANGELOG.md.

    It is a SKILL.md skill, tested with Claude Code and OpenAI Codex. The script is standard-library Python 3, so you can also run it by hand on any reply saved as a file.

    Partly. With --previous, the script compares a new reply with an earlier one and flags a reversal that brings no new evidence. It checks the two replies you give it, not a whole conversation.

    No. It finds agreement-first patterns and unexplained reversals. It does not check whether the facts are true, so you still inspect the evidence yourself.

    New in Testing & Debugging