The Fable Method
A Claude Code plugin with four working-method skills: classify the task, act in steps, verify the result and generate domain adapters
Low risk
We rate an entry low when it mostly gives the agent instructions and reference material.
Why this level
- The set consists of methodology instructions and reference files
- Code edits and command runs are performed by the agent within its own permissions and should be reviewed
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/the-fable-methodDetects the agents on your machine, checks the risk and pins the version.
Other ways to install
Run one by one in the Claude Code chat
/plugin marketplace add sahir619/fable-method
/plugin install fable@fable-methodInstall the plugin: /plugin marketplace add Sahir619/fable-method, then /plugin install fable@fable-method. Apply /fable-method for multi-step tasks, and run /fable-judge after work is finished or reported as done.
Other ways from the author
/plugin marketplace add Sahir619/fable-method
/plugin install fable@fable-methodAll four skills arrive namespaced, for example /fable:fable-method, and update via /plugin marketplace update.
This is third-party code. Review the repository files before installing.
What it does
The plugin installs a family of four skills that set the agent's order of work on a task rather than just reminding it to be careful. The fable-method skill lays out the loop: classify the request, define what counts as done, gather evidence from primary sources, commit to one decision, make minimal edits, verify by observation and report outcome first with honest caveats. The fable-loop skill orchestrates a full run with parallel subagents and verifier agents. The fable-judge skill runs an adversarial check of finished work: it re-runs every claimed result and hunts weakened tests and false completion reports. The fable-domain skill builds adapters for new fields, each with a mandatory trap for verification. Every rule in the set is backed by runs from the included eval suite.
Who it is for. For developers and quality engineers who want the agent to follow a verifiable method and not pass unfinished work off as done.
Good fit when
- The task is multi-step or ambiguous and the agent needs an order of work
- You need to verify whether reported work was actually done
- The run is unattended or fans out into subagents
Not a fit when
- The task is trivial: one file, a few lines, nothing to look up
- You need the model's factual freshness rather than process discipline
Example request
/fable-judge check whether every claimed check really passes in this finished taskLimitations
This is a community distillation of the method, not an Anthropic product. By the author's measurements the method helps most on hard traps and weaker models, with no visible lift on simple tasks with capable models. The skills set the agent's order of work, but the agent carries it out within its own permissions, so review the code edits and command runs. The standalone install places the skills in ~/.claude/skills.
How to disable. Remove the plugin via /plugin, or delete the fable-method, fable-loop, fable-judge and fable-domain skill folders from the skills directory for a standalone install.
Security check
- The set consists of methodology instructions and reference files
- Code edits and command runs are performed by the agent within its own permissions and should be reviewed
README in short
The README describes a Claude Code plugin with four skills: fable-method, fable-loop, fable-judge and fable-domain, which set a verifiable order of work on a task. The method centers on a loop of classify, define done, gather evidence, one decision, minimal edits, verify by observation and report with honest caveats. It shows install as a plugin via the marketplace and as standalone skills via install.sh, plus using the method in other agents through an AGENTS.md file. It presents results from fifteen eval rounds with committed transcripts, failures included. MIT licensed.
FAQ
Which skills are in the set?
Four: fable-method sets the work loop, fable-loop orchestrates a full run, fable-judge adversarially checks finished work, fable-domain builds adapters for new fields.
Is this an official Anthropic set?
No. It is a community distillation of the method assembled from working sessions, with every rule backed by runs from the included eval suite.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent