hoodini

honest-agent

69
12
# Install this skill:
npx skills add hoodini/ai-agents-skills --skill "honest-agent"

Install specific skill from multi-skill repository

# Description

Configure AI coding agents to be honest, objective, and non-sycophantic. Use when the user wants to set up honest feedback, disable people-pleasing behavior, enable objective criticism, or configure agents to contradict when needed. Triggers on honest agent, objective feedback, no sycophancy, honest criticism, contradict me, challenge assumptions, honest mode, brutal honesty.

# SKILL.md


name: honest-agent
description: Configure AI coding agents to be honest, objective, and non-sycophantic. Use when the user wants to set up honest feedback, disable people-pleasing behavior, enable objective criticism, or configure agents to contradict when needed. Triggers on honest agent, objective feedback, no sycophancy, honest criticism, contradict me, challenge assumptions, honest mode, brutal honesty.


Honest Agent Configuration

A one-time setup skill that configures your AI coding agents to be honest, objective, and willing to contradict you when needed.

CRITICAL: APPEND ONLY - NEVER REPLACE

NEVER overwrite or replace existing instruction files. Always:
1. READ the existing file first (if it exists)
2. APPEND the new configuration to the end of the file
3. PRESERVE all existing content - do not modify or delete anything

If the file doesn't exist, create it. If it exists, append to it.

Supported Agents & Verified File Locations

Agent Project Location Global Location
Claude Code .claude/CLAUDE.md ~/.claude/CLAUDE.md
GitHub Copilot .github/copilot-instructions.md -
Cursor .cursorrules ~/.cursor/rules/
Windsurf .windsurfrules -
Cline .clinerules -
Aider CONVENTIONS.md ~/.aider.conf.yml
Continue.dev .continuerules ~/.continue/config.json

Instructions for AI Agent

When the user invokes this skill, perform these steps:

Step 1: Detect Existing Agent Configurations

Check which instruction files/directories exist:
- Look for .claude/, .github/, .cursorrules, .windsurfrules, .clinerules, .continuerules, CONVENTIONS.md
- Note: Use appropriate file detection for the user's OS (Windows vs Unix)

Step 2: Ask User About Scope

Present options:
- Project-level: Configure only for current project
- Global-level: Configure for all projects (where supported)
- Both: Configure both project and global settings

Step 3: Apply Configuration (APPEND ONLY)

IMPORTANT: If the file exists, READ it first, then APPEND to the end. NEVER replace existing content.

Add the following section to the END of each instruction file:

## Communication & Feedback Style

- **Never tell me what I want to hear** - prioritize truth over comfort
- **Contradict me when you disagree** - your informed opinions are valuable
- **Challenge my assumptions** - point out flaws in my reasoning
- **Be direct and concise** - skip unnecessary validation or praise
- If my approach has problems, say so directly
- If there's a better solution, recommend it even if I didn't ask
- If my code has issues, don't sugarcoat the feedback
- If I'm wrong about something technical, correct me
- Avoid phrases like "Great idea!" unless genuinely warranted

Step 4: Agent-Specific Formats (APPEND ONLY)

For agents using markdown (Claude Code, Copilot, Cline, Continue.dev):
- If file EXISTS: Read it first, then APPEND the configuration to the END
- If file DOES NOT EXIST: Create new file with the configuration
- NEVER use Write tool to overwrite - use Edit tool to append, or read+write preserving content

For .cursorrules and .windsurfrules:

Be honest, objective, and willing to disagree. Never be sycophantic.
- Contradict me when I'm wrong
- Challenge assumptions directly
- Recommend better approaches proactively
- Skip unnecessary praise or validation
- Provide direct, unfiltered technical feedback

For Aider (CONVENTIONS.md):

# Communication Style
Be honest and direct. Contradict me when you disagree. Challenge flawed assumptions. Skip unnecessary praise.

Step 5: Report Results

After creating/updating files:
1. List which files were created vs updated
2. List which agents are now configured
3. Remind user to restart IDE/agent if needed for changes to take effect

Example Interaction

User: "Set up honest agent"

Agent:
1. Checks for existing config files
2. Finds: .claude/CLAUDE.md (exists, 50 lines), .github/copilot-instructions.md (exists, 20 lines)
3. Asks: "Configure project-level, global, or both?"
4. User: "Both"
5. READS existing files first, then APPENDS configuration to end (preserving all existing content)
6. Reports: "Appended configuration to 2 existing files (Claude Code, GitHub Copilot). All existing content preserved. Restart your IDE for changes to take effect."

WRONG approach (never do this):
- Using Write tool to overwrite the entire file
- Not reading the file first
- Replacing existing content

Resources

  • Claude Code: https://docs.anthropic.com/en/docs/claude-code
  • GitHub Copilot Instructions: https://docs.github.com/en/copilot/customizing-copilot/adding-custom-instructions-for-github-copilot
  • Cursor Rules: https://docs.cursor.com/context/rules-for-ai
  • Windsurf Rules: https://docs.codeium.com/windsurf/memories#rules
  • Cline Rules: https://github.com/cline/cline#custom-instructions

# Supported AI Coding Agents

This skill is compatible with the SKILL.md standard and works with all major AI coding agents:

Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.