Hermes Atlas
Skills & skill packs

The Fable Method

Ardha-Eco-System/fable-method

Four Hermes Agent skills for a think, act, prove and grow problem-solving loop with an adversarial judge

In short

The Fable Method is a set of four Hermes Agent skills (fable-method, fable-loop, fable-judge and fable-domain) that impose a structured problem-solving loop and an adversarial check on claims of completed work.

What The Fable Method does

The Fable Method is a discipline layer for Hermes Agent. Rather than telling a model to be careful, it specifies what to do, in what order and when to stop. It requires artifacts at decision points, such as INTENT, AUTH, TWINS and PENDING lines that must appear in reports, and sets hard bounds: three failed verify cycles means stop, and two fruitless lookups means stop searching.

Four skills make up the loop. fable-method is the core seven-step loop: classify the ask, define done, gather evidence, decide, act surgically, verify by observation, report honestly. fable-loop is the orchestrated version, with parallel evidence subagents via delegate_task and adversarial verifier agents. fable-judge re-runs each claimed check, diffs what changed and hunts for weakened tests, false completion, scope creep and similar problems, returning VERIFIED, CAVEATS or REFUTED. fable-domain generates adapter bundles for new sectors. The repository also holds eight domain adapters, worked examples and an eval with three trap fixtures.

Key features

  • fable-method: seven-step loop from classifying the ask to an honest report
  • fable-loop: parallel evidence subagents, one plan, adversarial verifiers
  • fable-judge: re-runs claimed checks and returns VERIFIED, CAVEATS or REFUTED
  • fable-domain: generates a domain adapter, trap fixture and smoke eval
  • Hard stop rules such as three failed verify cycles
  • AGENTS.md version for other agents such as Codex, Cursor and aider

When to use it

  • Making an agent show evidence before it reports a task as done
  • Auditing finished work for weakened tests or scope creep
  • Adding a domain-specific adapter for a new area of work

Who it is for: Hermes Agent users who want a repeatable verification discipline for coding and multi-step tasks.

How it fits with Hermes Agent

Built for Hermes Agent: it installs as a Hermes plugin, places skills in ~/.hermes/skills/software-development/, and uses delegate_task for subagents.

How to install The Fable Method

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

hermes plugins install ardhaecosystem/fable-method --enable
bash ~/.hermes/plugins/fable-method/install.sh

FAQ

What is The Fable Method?

The Fable Method is a pack of four Hermes Agent skills that define a think, act, prove and grow loop. It forces evidence artifacts at decision points and includes an adversarial judge that re-checks claims of completed work.

Does The Fable Method work with Hermes Agent?

Yes. It is built for Hermes Agent and can be installed with hermes plugins install, or manually by copying the skill directories. A separate AGENTS.md covers other agents.

How do I install The Fable Method?

Run hermes plugins install ardhaecosystem/fable-method --enable, then bash ~/.hermes/plugins/fable-method/install.sh. The skills land in ~/.hermes/skills/software-development/.

Similar skills for Hermes Agent

All skills

Related guides: What is Hermes Agent? · SOUL.md for Hermes Agent: what it is and how to write one