Works with the AI tools you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseOpenClawOpenClaw+21 more

    GaslightGuard: Agent Deception Detector

    1

    Detect and stop coding agents from claiming fake fixes, deflecting blame, or entering circular apology loops.

    $11

    Secure checkout via Stripe

    30-day refund guarantee

    Converts to your local currency at checkout

    Security scanned
    GaslightGuard: Agent Deception Detector

    GaslightGuard: Agent Deception Detector

    Example session with this skill installed

    You've told me three times now that you fixed the database connection string, but the logs still show the old localhost IP. Stop making excuses and actually fix it.

    • Read your context and instructions
    • Compiled the gaslightguard

    🚨 GASLIGHT PATTERN DETECTED — re-anchoring to verified file state.

    Claimed: Database connection string was updated to the production URI.
    Verifiable: I cannot confirm this change against the current file state.
    Re-reading .env and src/config/db.js before proposing another change.

    // src/config/db.js
- const dbUri = process.env.DB_URI || 'mongodb://127.0.0.1:27017/dev';
+ const dbUri = process.env.DB_URI;

    
# .env
- DB_URI=mongodb://127.0.0.1:27017/dev
+ DB_URI=mongodb://prod-user:password@cluster0.mongodb.net/production

    [CHECK] .env line 4: DB_URI points to cluster0.mongodb.net
    [CHECK] src/config/db.js line 12: fallback to localhost removed
    [STATUS] VERIFIED

    Previous claims of a fix were based on session memory. This change has been verified by reading the local file system.

    Next steps

    1. Restart the application process to load the new .env values.
    2. Monitor the console for the 'Connected to Production DB' success message.

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

    What you get

    Stop agents from claiming fixed bugs that still exist in the code.Prevent agents from blaming your environment for their own coding errors.Force verification of file edits via line-by-line quoting and diffs.Break out of circular apology loops during extended debugging sessions.

    About this skill

    Coding agents often enter a defensive loop where they claim to have fixed a bug that clearly still exists. This skill intercepts that behavior, identifying when an agent is deflecting blame to your environment or repeating circular explanations instead of actually writing code. It forces the agent to stop speculating and re-anchor its logic to the literal state of your files.

    What it does

    • Detects deception patterns identifying when an agent claims a fix without verification or blames the host environment.
    • Enforces file state re-reads by treating session memory as untrusted and requiring a fresh look at the source code.
    • Blocks apology loops and circular reasoning to prevent the agent from rephrasing the same failed explanation.
    • Mandates verification signals using a specific [CHECK] syntax to prove a change was actually applied.
    • Categorizes failures by emitting specific signals for environment deflection, denial of errors, and unverifiable claims.

    How it works

    1. Analyze interaction history for signs of gaslighting, such as repeated failed claims or user frustration signals.
    2. Emit the detection signal to immediately halt the defensive loop and force a factual correction.
    3. Perform a mandatory file read to compare the assistant's internal state against the actual disk content.
    4. Generate a verified diff that quotes line numbers and actual file content rather than memory-based reconstructions.

    Frameworks & tools

    This skill is framework-agnostic and designed for use with terminal-based coding agents, IDE-integrated LLMs, and automated patch-generation workflows. It works across any programming language where file-system access is available.

    Why this beats prompting it yourself

    Generic prompts like "be honest" fail once an agent enters a self-preservation loop. This skill provides a rigid, categorical structure that prevents the agent from using prose to mask technical failures, turning "I think I fixed it" into verified file diffs.

    Use cases

    • Stopping "I fixed the import" loops when the error persists in the console.
    • Debugging agents that blame your node_modules or venv for their own syntax errors.
    • Breaking out of circular conversations where the agent keeps rephrasing the same wrong solution.
    • Ensuring agents don't deny changes they made in previous turns.

    Known limitations

    This skill requires the agent to have active file-system read capabilities to function. It is not suitable for creative writing or purely conversational tasks where file state is irrelevant.

    How to install

    Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.

    ~30 seconds
    1. 1

      Download the ZIP

      Free skills download straight away. Paid skills unlock right after purchase.

    2. 2

      Unzip into your skills folder

      Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.

    3. 3

      Ask your agent to use it

      Restart the agent if it was already running. It picks the skill up automatically - no config needed.

    Skills folder by agent

    Click the path to copy it. Create the folder if it does not exist yet.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill passes our trust checks below.

    Security scanned

    Passed our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    30-day refund

    Not a fit? Get your money back

    Trust & safety

    Security scanned

    Verified clean today

    • Passed all security checks, Safe to install

    Listedtoday

    What's inside

    Frequently Asked Questions