- Home
- Skills
- Agents & Orchestration
- agent consensus until
Works with the AI Agents you already use
agent consensus until
A multi-perspective debate engine that iterates until real consensus is reached or a stall is detected.
$5
agent consensus until
Example session with this skill installed
Deep consensus: Should we migrate our primary database from PostgreSQL to a NoSQL solution for the new analytics module? Don't stop until you agree.
- Read your context and instructions
- Compiled the agent consensus until
I have completed the deep consensus protocol regarding the database migration. The debate ran for 3 rounds before reaching a unanimous agreement.
Consensus: Maintain PostgreSQL with optimized indexing and a materialized view strategy instead of migrating to NoSQL.
Confidence: High
Agreement: unanimous
Rounds: 3 Stop reason: consensus
Agreed points
- PostgreSQL can handle the projected 20% annual data growth using partitioning.
- The cost of rewriting the existing reporting ORM outweighs the performance gains of NoSQL.
- Materialized views provide the required latency reduction for the analytics dashboard.
Resolved objections
- Write latency concerns → Resolved by implementing a dedicated follower instance for heavy analytics queries (round 1).
- Schema flexibility → Resolved by utilizing JSONB columns for the unstructured event data portion (round 2).
Key assumptions
- Projected write volume remains under 5,000 events per second.
- The team maintains current expertise in SQL tuning and vacuum management.
Full debate transcript
Round 0 — Initialization
Advocate: We should move to NoSQL to ensure horizontal scalability. Current analytics queries are slowing down the production API.
Critic: NoSQL lacks ACID compliance for our financial audits and will require a total rewrite of our reporting engine.
- Synthesizer: There is a clear tension between scalability and consistency. We need to explore hybrid options.
Round 1 — Iteration
Critic (new objections): Moving to NoSQL introduces eventual consistency risks that the Advocate hasn't addressed.
Advocate (response): I concede the consistency risk. We could use a follower database instead to isolate analytical load.
- Synthesizer (update): Shifted focus toward SQL optimization and read-replicas. Unresolved count: 1.
Convergence log
- Round 0: unresolved=2
- Round 1: unresolved=1 (↓1)
- Round 2: unresolved=0 → consensus
Next steps
- Review the
scripts/load-test.shto validate the follower instance performance. - Update the
db/schema.sqlto include the proposed JSONB fields.
Connects securely to your tools. The creator never sees your data.
About this skill
High-stakes decisions often fail because AI agents agree too quickly or hallucinate a consensus that hasn't been pressure-tested. This skill forces a multi-perspective debate that refuses to terminate until genuine agreement is reached or a terminal stall is detected.
What it does
- Persistent iteration runs a cyclical debate loop that continues until all objections are resolved or the safety cap is hit.
- Multi-role debate coordinates an Advocate, Critic, Synthesizer, and Verifier to attack and defend the proposal from every angle.
- Convergence tracking measures the reduction of unresolved objections round-over-round to detect progress or deadlocks.
- Parallel initialization launches initial positions concurrently to ensure the Advocate and Critic remain unbiased by each other's early reasoning.
- Fact-check verification triggers a final review of core claims to prevent the group from agreeing on factual inaccuracies.
How it works
- Initialize by launching the Advocate, Critic, and Synthesizer in parallel to establish the baseline unresolved objection count.
- Iterate through debate rounds where the Critic finds new flaws and the Advocate must either rebut them or update the consensus draft.
- Analyze convergence after each round to check if objections are hitting zero, stalling for two rounds, or reaching the safety cap.
- Verify the final output by fact-checking the top three claims before delivering the full audit log and consensus report.
Frameworks & tools
This skill utilizes read_file, search_files, and list_directory to gather context. It employs a fan-out subagent architecture for parallel execution and iterative loops for state management.
Why this beats prompting it yourself
Standard prompting leads to "groupthink" where the LLM tries to please the user by agreeing with itself immediately. This skill enforces a red-team protocol that forbids the Critic from repeating resolved points and tracks mathematical convergence to prove the answer is defensible.
Use cases
- Evaluating high-stakes architectural changes in complex codebases.
- Stress-testing strategic business proposals where a simple "yes" is dangerous.
- Resolving conflicting technical requirements between multiple stakeholders.
- Auditing mission-critical logic that requires a deep audit trail for compliance.
Known limitations
The process is resource-intensive and slow, capped at 8 rounds to prevent infinite loops. It requires specific subagent capabilities for optimal parallel initialization.
Resource intensive. Maximum safety cap of 8 rounds. Requires subagent support for parallel execution. Stalls after 2 rounds of no progress.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 16 days ago