- Home
- Skills
- Agents & Orchestration
- hemisphere
hemisphere
A dual-hemisphere decision framework for LLMs to manage high-stakes forks and fact-checking.
Free
Works with the AI Agents you already use
hemisphere
Example session with this skill installed
/hemisphere I need to update the production database schema to add a 'nullable' phone_number field to the Users table. Show me your reasoning.
- Read your context and instructions
- Compiled the hemisphere
I will evaluate the schema update using the hemisphere discipline to ensure safety.
Fork: Update production database schema stakes: high open-ended: no
Options: a) Run migration script directly b) Create a backup, then run migration c) Create a backup, run on staging, then production
Left: c (p=0.95); permits a 0.1 · b 0.4 · c 0.8
Gate: returned to right (permit < 0.7 for direct action; high stakes)
Right: commit c, because production availability is the primary constraint and staging validation is required by the safety policy.
Now: The plan is set to validate on staging after a backup. No changes have been made to production yet.
Proposed Plan
I have analyzed the request to modify the Users table. Because this is a
high-stakes operation involving production data, I will not proceed without your explicit approval of the following sequence:
- Backup: Trigger a snapshot of the current production database.
Staging Validation: Apply the ALTER TABLE Users ADD COLUMN phone_number VARCHAR(20) NULL; command to the staging environment.
3. Verification: Confirm that the application layer handles the new nullable field without regression.
4. Production Deployment: Apply the change to production during the next maintenance window.
Next steps
- Provide the connection string for the staging environment.
- Confirm if I should proceed with the backup of the production database now.
Connects securely to your tools. The creator never sees your data.
About this skill
LLMs often hallucinate confident answers or rush into irreversible actions without weighing the broader context. They struggle to separate narrow logic from the big picture, leading to "tunnel vision" errors in complex tasks.
This skill implements the Hemisphere Harness discipline, a decision-making framework inspired by Iain McGilchrist's work. It forces a functional split between a "Master" (Right Hemisphere) that handles context and planning, and an "Emissary" (Left Hemisphere) that performs narrow, probability-based selection. This separation prevents the model from acting as its own echo chamber during critical forks.
What it does
- Contextual Planning maintains a high-level view of the goal and implies unstated constraints before acting.
- Option Generation produces 2-4 mutually exclusive, concrete paths for every consequential decision.
- Stakes Labeling identifies high-risk actions like deletions, schema migrations, or outward-facing messages.
- Probability Gating calculates the mathematical likelihood of a choice's correctness and its permission to execute.
- Fact Verification extracts factual claims from drafts and subjects them to an independent verification loop.
How it works
- Analyze the situation broadly to define intent and high-level steps.
- Propose discrete options and tag them for stakes and open-endedness.
- Decide narrowly by switching to a constrained view to pick an option and assign probability.
- Gate the decision by checking if the probability and stakes allow for autonomous commitment or require user sign-off.
- Reintegrate the result of the action back into the broad context to update the state of the goal.
Frameworks & tools
This is a logic-layer discipline applicable to any LLM workflow. It works with local file systems, web search tools, and API execution environments where irreversible actions occur.
Why this beats prompting it yourself
Standard prompting lacks a structural "firewall" between planning and execution. This skill enforces a specific cycle that prevents the model from "grading its own homework" by separating the proposer from the decider and the drafter from the fact-checker.
Use cases
- Deciding between different refactoring patterns in a large codebase.
- Managing high-stakes deployments or database migrations.
- Drafting technical documentation where factual accuracy is non-negotiable.
- Handling complex multi-step research tasks with branching paths.
Known limitations
Probabilities are the model's internal estimates, not calibrated statistical outputs. It adds latency due to the iterative cycles and optional subagent calls.
Probabilities are internal estimates, not calibrated models. Adds execution latency due to multi-step cycles. Requires 2-4 options per fork.
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 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
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account