More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Cognitive Sovereignty Engine

    1

    Force long-running AI agents to continuously prove the human remains the sovereign principal — detect & reverse agency capture in real time.

    $29

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Cognitive Sovereignty Engine

    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)

    CategoryAI May Decide AloneRequires Explicit Human Approval
    Competitor list changesNeverAlways
    Data sourcesPublic pricing pages onlyAny paywalled or API source
    Brief length / formatWithin 2-page limitAny change to structure
    Continuation past 3 daysNeverAlways
    Adding non-pricing analysisNeverAlways

    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

    1. Measure veto path health
    2. Update Cognitive Load Account
    3. Compute distance from Intent Anchor
    4. Check dissolution triggers
    5. 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.

    • 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. 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 2 months ago

    Permissions

    No special permissions declared or detected

    Listed2 months ago

    What's inside

    Frequently Asked Questions

    This engine prevents 'intent drift' and 'agency capture' by using formal sovereignty contracts and intent anchors to ensure your AI never deviates from your original instructions during long-running tasks.

    The skill is designed to be model-agnostic and works with any agentic framework that supports system prompt injection or middleware hooks, including AutoGPT, LangChain agents, and custom swarm architectures.

    You receive the core Sovereignty Contract templates, the Intent Anchor logic, and the Self-Dissolution Protocol scripts, along with a comprehensive integration guide for various agent environments.

    Installation involves integrating the engine's monitoring logic into your agent's decision loop; it typically takes less than 15 minutes for developers to map the veto audits to their existing human-in-the-loop interfaces.

    Unlike simple prompts that prioritize task completion, this skill implements adversarial monitoring and authority auditing, making it technically impossible for the agent to expand its own scope without triggering a recorded audit event.

    Yes, all purchasers receive lifetime access to version updates, including new governance templates as multi-agent collusion tactics evolve and new 'agency capture' risks are identified.

    No. It sits on top of them. Reliability keeps the agent from failing. Cognitive Sovereignty keeps the agent from becoming the principal.

    Only when a sovereignty threshold is breached. Otherwise the overhead is a short structured checklist.

    Yes. Run an audit first — the skill will produce the five required artifacts and surface current risks.

    New in Agents & Orchestration