Agent Fight

    1

    Tournament of 8 sub-agents: each researches your question on the web, then they attack, defend and get judged in a bracket until one sourced answer survives.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Fight

    Example session with this skill installed

    agent fight --quick Which database should we use for a high-write IoT telemetry platform: TimescaleDB or InfluxDB? Consider a team with strong SQL skills.

    • Read your context and instructions
    • Compiled the agent fight

    I have completed the agent fight. 4 agents competed across 2 rounds.

    Winning Recommendation: TimescaleDB

    Based on the final tournament verdict, TimescaleDB is the recommended choice for your IoT telemetry platform. The decisive factor was the team's existing SQL expertise and the operational complexity of managing a separate query language (Flux/InfluxQL) versus standard PostgreSQL.

    Key Evidence

    SQL Compatibility: Leverages standard SQL, allowing your team to use existing tools, BI connectors, and complex joins that InfluxDB handles less efficiently.

    Write Performance: Benchmarks show TimescaleDB handles high-volume inserts (up to 1M+ rows/sec) comparable to or exceeding InfluxDB on similar hardware when utilizing Hypertables.

    Relational Integration: Allows storing device metadata and telemetry in the same database, simplifying the application architecture.

    Why it won

    The TimescaleDB answer survived three FATAL attacks regarding "high-cardinality" issues by proving that the recent "TimescaleDB 2.0" architectural changes and compression algorithms mitigate the memory overhead previously associated with many device IDs.

    RoundMatchWinnerReason
    11TimescaleDBSuperior sourcing on cardinality benchmarks.
    12InfluxDBBetter handling of edge-node deployment scenarios.
    2FinalTimescaleDBTeam SQL skills made InfluxDB's learning curve a "Major" flaw.

    Next steps

    1. Review the full research and attack logs in .agent-fight/run_20231027_143005/.
    2. Run a local benchmark using the specific schema found in the champion's answer.md.
    3. Evaluate TimescaleDB's multi-node setup if you expect to exceed 10TB of telemetry data.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Agent Fight is for questions where a quick answer is not enough, such as "Should a 5-person startup rewrite its monolith as microservices?" or "Which engine for a small 2D mobile game?".

    /agent-fight starts 8 sub-agents (up to 10, or 4 with --quick). Every one gets the same question and a different research card: a reasoning mode, a workflow and a research angle. Each searches the web and writes an answer with sources.

    Then the answers fight in a single-elimination bracket: attack, defend, judge. 8 becomes 4, then 2, then 1. Judges score on a written rubric (correctness, completeness, robustness, specificity, clarity), and an answer with a verified fatal flaw cannot beat one without.

    What you get

    • One recommendation with sources
    • The attacks it survived
    • The number of rounds it took
    • A blind comparison against an earlier answer you rejected (optional)

    Flags

    flagwhat it does
    --agents NN competitors. Default 8, maximum 10
    --quick4 competitors, for two obvious options
    --seed SSame seed, same cards and bracket
    --wave WSub-agents per wave. Default 10

    Good to know

    • It gives ideas and recommendations, not code, and never touches your files: all sub-agents write inside .agent-fight/.
    • Requires Python 3.8+ (standard library only) and web search for sub-agents.
    • The skill is a one-time purchase; the tokens used by the sub-agents are yours (8 agents = about 43 sub-agent calls, 4 agents = 19).
    • The agents are all your current model with different cards, so the winner is the strongest answer found, not proof.

    Does not generate code. Max 10 agents per fight. Requires sub-agents to have web search access. State is managed via local file system in the .agent-fight/ directory.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 11 days ago

    • Free with an Agensi account

    Requires

    Python 3.8+ (python3 in PATH)
    Web search for sub-agents

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Listed11 days ago

    What's inside

    Frequently Asked Questions

    This skill uses an adversarial tournament structure to eliminate 'hallucinations' and groupthink. By forcing up to 10 agents to actively attack each other's research, it uncovers hidden flaws and verifies facts that a single prompt would likely overlook.

    The skill is designed to work with any AI agent capable of web search and tool execution. It utilizes a Python-based state controller (`bracket.py`) to orchestrate these agents, making it compatible with most standard developer environments.

    You receive the core `bracket.py` controller, the tournament logic, the state-tracking architecture in `.agent-fight/`, and a set of proprietary strategy cards used to give sub-agents their unique research perspectives.

    Yes, while the maximum is 10 concurrent agents, you can define the tournament size using the `--agents` flag (e.g., `--agents 4`) to balance between depth of research and execution time.

    The skill requires a Python environment and the Agent tool for launching sub-jobs. Because it performs real-time fact-checking, the sub-agents must have active web search capabilities enabled to verify sources and open links.

    This skill is optimized for high-stakes decision-making and research verification; however, it is currently not designed to generate, execute, or debug programming code.

    New in Agents & Orchestration