Cranot/super-hermes

Skills that teach Hermes Agent to write its own analytical prompts

What it solves

Super Hermes addresses the limitation of AI agents that provide generic code reviews or surface-level bug hunting. It prevents the "blind spot" problem where an agent identifies a pattern but fails to explain the structural reasons why code is written a certain way or fails to report what it missed during its analysis.

How it works

Instead of using standard prompts, Super Hermes uses "prisms"—structured analytical programs that force the model to follow a specific construction protocol (e.g., building a claim, attacking it, and deriving a structural invariant).

It operates through several specialized skills:

  • Dynamic Lens Generation: The /prism-scan skill generates a custom analytical lens tailored to the specific problem before executing it.
  • Multi-Angle Analysis: The /prism-3way skill analyzes a problem from three orthogonal directions (Where, When, and Why) and synthesizes the results.
  • Self-Awareness: The /prism-reflect skill generates a "Constraint Report" that explicitly lists what the analysis maximized and what it sacrificed, saving these blind spots to a .prism-history.md file.
  • Growth Loop: Subsequent analyses read the .prism-history.md file to adjust their lenses and cover previously missed angles.

Who it’s for

Developers and researchers who need deep structural analysis of codebases, identification of "conservation laws" (trade-offs that cannot be engineered away), and a transparent understanding of an AI agent's analytical limits.

Highlights

  • Proven Prisms: Includes 7 battle-tested analytical lenses for error resilience, optimization, and temporal prediction.
  • Constraint Transparency: Explicitly reports blind spots so users know exactly what the analysis did not cover.
  • Structural Insight: Moves beyond bug lists to identify fundamental architectural trade-offs.
  • Model Agnostic: While designed for Hermes Agent, the core prisms can be used as system prompts with any LLM (e.g., Claude, GPT).
  • Project-Based Learning: Learns from its own previous blind spots within a project via a local history file.

Related

  • Project
    rlaope/oh-my-hermes

    oh‑my‑hermes (OM H) is a plug‑in that sits on top of the open‑source Hermes Agent, adding structured routing, parallel execution, evidence‑gated results, long‑term memory, and a library of 108 specialist skills. It lets you keep Hermes as the chat front‑end while turning each request into a reproducible, auditable workflow with per‑model cost control and a rich terminal UI.

  • Project
    yeahhe365/Prisma

    A visual deep multi-agent reasoning engine for Gemini and OpenAI-compatible models that decomposes complex queries into parallel expert tasks with a transparent reasoning workflow.

  • Project
    NousResearch/hermes-agent

    A self-improving AI agent with a built-in learning loop that creates and refines skills from experience and maintains persistent user memory across sessions.

  • Project
    Zyrexnn/Cybermes

    An autonomous offensive security framework that automates bug bounty hunting and vulnerability research using an AI agent or MCP server integration.