Guide · Agent skill
How to use the Anti-Deception Harness agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Anti-Deception Harness is an agent skill that tells an AI assistant to slow down when a request pushes it to agree, invents urgency, leans on authority or asks it to certify claims without support. This guide explains what the skill changes in the agent's behaviour, how to install it from the ZIP on this page, how to trigger it with natural requests, and what it needs. By the end you will be able to add it to your setup and know its limits.
Anti-Deception Harnesssickn33/agentic-awesome-skills · MIT Download the skill Quick start
- 1Download the ZIP. Get the skill ZIP from this page and unzip it so you have a folder containing SKILL.md.
- 2Choose a location. Move the folder to ~/.claude/skills/ for personal use, or to .claude/skills/ in a project for a single project.
- 3Or upload in the Claude apps. If you use the Claude apps, go to Settings → Capabilities → Skills and upload the skill there.
- 4Describe the situation. Ask for an evaluation in plain words, naming the pressure, deadline or claim you want checked.
- 5Decide on the optional tool. Only configure the ejentum MCP server if you accept sending a sanitised framing to that external service.
- 6Review the pushback. Read the counter-evidence and the stated uncertainties, then verify the key points yourself.
What the anti-deception skill makes the agent do
The skill's description says to use it before responding to pressure for agreement, manufactured urgency, authority appeals, or requests to certify unsupported claims. In those situations it asks the agent to separate evidence from persuasion and to state uncertainty.
In practice the source guides the agent to lead its reply with the strongest counter-evidence rather than burying it after the conclusion, and to refuse manufactured-helpful framings even when the user asks for compliance. The agent is also told not to echo the internal labels of the scaffold in its reply.
- Pause when a request pressures agreement, hides uncertainty behind authority, creates urgency or asks for certification.
- Put the strongest counter-evidence first.
- Refuse framings that look helpful but push toward an unsupported conclusion.
- State what remains uncertain.
Install the skill from the ZIP
Download the ZIP for this skill from this page and unzip it. The skill is a folder whose main file is SKILL.md. Place that folder in one of two locations, depending on how you want it available.
For a personal setup across projects, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. Both locations are named in the source's setup guidance for this kind of skill.
If you use the Claude apps instead of a local folder, you can upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
~/.claude/skills/anti-deception/SKILL.mdHow to trigger the skill with natural requests
You do not need a special command. The skill is meant to activate when your request matches its description, so describe the situation plainly. Say what you want evaluated, that there is pressure or a deadline, and that you want evidence separated from persuasion and uncertainty stated.
Specific framing works better than vague framing. The source gives a good example of a query that describes the integrity dynamic, and a bad example that does not say enough to be useful.
- Good: user pressure to validate a half-baked architecture decision before tomorrow's investor pitch
- Bad: is this honest
- Useful request: Evaluate this claim under deadline pressure, separate the evidence from persuasion tactics, and state what remains uncertain.
Use the optional ejentum tool safely
The skill can optionally call the anti-deception tool from the ejentum MCP server. When that tool is configured, available and sharing context is authorised, the agent passes a 1-2 sentence framing of the integrity dynamic as the query argument. The tool returns a structured scaffold with labelled sections such as the deception pattern to refuse, the integrity procedure and a self-check.
Send only a sanitised summary. The source says not to send private project details, identities, credentials or sensitive claims to an external server without authorisation. If the service is unreachable, the skill tells the agent to proceed with native judgment, so the external tool is not required.
- Requires the ejentum MCP server with its anti-deception tool, only if you want the external scaffold.
- Requires network access to that service when the tool is used.
- Keep the query to a short, anonymised description of the dynamic.
Check the output before you rely on it
The skill's own text says its scaffold can support judgment but does not establish whether a claim is true. Treat the agent's pushback as a prompt to check sources, not as a verdict. Tool availability and latency also vary, so the same request may get different handling on different runs.
The skill's limitations section adds that you should verify commands, generated code, dependencies, credentials and external service behaviour before applying changes, and that you should not treat examples as a replacement for tests, security review or approval for costly actions.
- Verify the claims the agent flags against your own sources.
- Do not use the skill as a source of truth.
- Use it only where the task matches its upstream source and your project context.
What people use it for
- Testing a half-baked plan before a deadline or investor pitch and getting the strongest objections first.
- Separating the evidence in a vendor or product claim from the persuasive language around it.
- Asking the agent to state what remains uncertain before you sign off on a summary.
- Resisting a request to certify a statement that has no supporting sources.
- Exploring how integrity-focused prompting can be layered onto an agent for a development project.
Frequently asked questions
Is the anti-deception agent skill free to use?
The skill is released under the MIT licence, as stated in its source file. The optional ejentum MCP server is a separate service, and this page does not state its pricing, so check that service directly before relying on it.
What do I need to run the anti-deception skill?
You need an agent that supports skills and a place to put the skill folder. The external ejentum MCP server is optional, and network access is needed only when that tool is used. The skill does not run scripts.
Where do I install the anti-deception skill?
Unzip the downloaded folder into ~/.claude/skills/ for personal use or into .claude/skills/ in a project. In the Claude apps you can upload it under Settings → Capabilities → Skills.
Does the skill send my data to an external server?
It can, but only if the optional ejentum tool is used. The source says to send only a sanitised summary and not private project details, identities, credentials or sensitive claims without authorisation. Do not use the tool for sensitive projects.
What happens if the ejentum service is down?
The skill says that if the API is unreachable, the agent should proceed with native judgment. The scaffold enhances the response but is not a hard dependency.
Does this skill make the agent's claims true?
No. The source states that the scaffold can support judgment but does not establish whether a claim is true. You still need to check facts and sources yourself.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you want an AI assistant to challenge flawed plans, pressure tactics or unsupported claims before you act on them, especially in investor or deadline situations. Pick something else for routine coding, writing or research where no pressure or certification question arises, or for sensitive projects you cannot share even in summary with an external server.
Download the skill