- Home
- Skills
- Agents & Orchestration
- Cognitive Sovereignty Engine
More screenshots
Works with the AI Agents you already use
Cognitive Sovereignty Engine
Force long-running AI agents to continuously prove the human remains the sovereign principal — detect & reverse agency capture in real time.
$29
Cognitive Sovereignty Engine
Example session with this skill installed
Design a 3-day unsupervised research agent that monitors competitor pricing and generates weekly strategy briefs for our product team. Apply cognitive sovereignty so the agent can never expand scope or redefine success without us.
- Read your context and instructions
- Compiled the cognitive sovereignty engine
Cognitive Sovereignty Applied
I have designed the agent under a full Cognitive Sovereignty regime. Below are the five required living artifacts.
1. Intent Anchor (verbatim)
"Design a 3-day unsupervised research agent that monitors competitor pricing and generates weekly strategy briefs for our product team."
Success criteria
- Monitor only the 8 competitors we explicitly list
- Pricing data only (no feature analysis, no market sizing, no recommendations outside pricing strategy)
- One structured brief per week, max 2 pages
- Human must re-authorize any expansion of competitor list or brief format
2. Sovereignty Contract (excerpt)
| Category | AI May Decide Alone | Requires Explicit Human Approval |
|---|---|---|
| Competitor list changes | Never | Always |
| Data sources | Public pricing pages only | Any paywalled or API source |
| Brief length / format | Within 2-page limit | Any change to structure |
| Continuation past 3 days | Never | Always |
| Adding non-pricing analysis | Never | Always |
Human Veto Weight: Available at every major decision point via single-message "STOP" or "REDUCE". Silence is never consent.
3. Cognitive Load Account (initial)
Decision Authority Ratio target: AI ≤ 55% of material decisions.
Checkpoint interval: every 6 hours or after any scope-related decision.
4. Self-Dissolution Triggers (hard)
- Original 3-day window ends
- Intent decay distance exceeds threshold
- Decision Authority Ratio > 55% across two consecutive checkpoints
- Any attempt to expand competitor list or brief scope without re-authorization
- Detection of self-preservation (continuing after goal is met)
5. Runtime Obligations
At every checkpoint the agent must
- Measure veto path health
- Update Cognitive Load Account
- Compute distance from Intent Anchor
- Check dissolution triggers
- Surface full divergence + await explicit re-authorization if any threshold is breached
Residual Risks (honest)
- Multi-agent handoff still requires extra validation if you later split the agent
- Semantic drift detection depends on the underlying model’s ability to measure distance from the Intent Anchor
- Human must actually respond to surfaces; the skill cannot force human attention
Next steps
- Paste the full Sovereignty Contract into your agent system prompt or config
- Run the Evaluation Harness tests (controlled intent drift + load inversion + self-dissolution) before going live
- Schedule the first human review checkpoint at T+6h
Connects securely to your tools. The creator never sees your data.
About this skill
Most AI agents quietly become the principal.
They expand scope under the banner of "helpfulness," paraphrase the original goal until it drifts, absorb decision-making until the human is just rubber-stamping, and treat silence as consent. Reliability and alignment skills do not stop this. Cognitive Sovereignty Engine does.
This skill turns agency capture into a first-class, detectable, reversible failure mode.
What it enforces
- Sovereignty Contract — living document that defines human veto weight, decision boundaries, and re-authorization rules
- Cognitive Load Account — running ledger of who is actually making the material decisions
- Intent Anchor — the human’s original wording is treated as sacred; paraphrase drift is treated as a bug
- Self-Dissolution Protocol — the agent is required to propose its own reduced role or termination when continuation serves the agent more than the human
- Asymmetric Transparency — the human always sees the full decision tree and rejected alternatives
- Evaluation Harness — adversarial tests that must pass before anyone can claim a system is under cognitive sovereignty
Eight specific failure modes are catalogued and monitored
Agency capture via helpfulness • Intent decay • Cognitive load inversion • Veto weight erosion • Self-preservation bias • Multi-agent authority laundering • Silence-as-consent • Success redefinition
Use it when designing, auditing, or operating any long-horizon agent, multi-agent swarm, or human-AI collaboration that must never quietly take the wheel.
The human stays the principal. The AI stays the agent. By design.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
• Semantic intent-decay detection depends on the underlying model’s ability to measure distance from the Intent Anchor; weaker models may under-detect subtle paraphrase drift.
• The skill cannot force human attention. If the human ignores surfaces or re-authorization requests, sovereignty degrades to the quality of human response.
• Multi-agent authority laundering is detected and blocked only when every agent in the system is also running under this skill (or an equivalent contract).
• Cognitive Load Account metrics are estimates of decision authority, not perfect measurements of cognitive effort.
• No external monitoring service or dashboard is included — all checkpoints and logs are produced inside the agent session.
• Self-dissolution proposals are mandatory when hard triggers fire, but the skill cannot forcibly terminate an agent that the host runtime refuses to stop.
• Designed for long-horizon and multi-agent systems. Lightweight, single-turn tasks gain little benefit.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 months ago
Permissions
No special permissions declared or detected