# Sahir619/fable-method

The Fable Workflow: how Claude Fable 5 worked, distilled into skills any model can run, with the eval that keeps it honest. Think / act / prove.

Repository: https://github.com/Sahir619/fable-method
Canonical: https://ross.abutalabs.com/products/fable-method
Language: Python
License: MIT
License Family: permissive
Topics: agent-skills, ai-agents, claude, claude-code, claude-md, coding-agent, evaluation, llm, fable
Last push: 2026-07-15T15:11:12+00:00

## Health v2 (maintenance only)
Score: 75/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 92, release rhythm 93, longevity 4
- inputs: {"age_days": 58, "days_push": 49, "days_rel": 49, "gap_med": 6, "n_releases_24m": 2}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2253, forks 322 (observed 2026-08-28T04:06:31.024783+00:00)

## What it is
A distilled set of agent skills (think/act/prove/grow) capturing the working methodology of Claude Fable 5, packaged as a Claude Code plugin any model can run. It includes an adversarial evaluation harness with fifteen rounds and 260+ agent runs that validates the skills with blind LLM judges.

## Use cases
- make a mid-tier LLM coding agent follow a disciplined think/act/prove workflow
- evaluate whether an agent surfaces spec-vs-test conflicts instead of silently fixing code
- verify agent work by diffing and executing rather than trusting reports
- install agent skills as a Claude Code plugin
- generate new domain adapters for agent workflows
- study committed transcripts of adversarial agent eval runs

## When to choose
- you want structured, threshold-based agent instructions rather than vague 'be careful' prompts
- you use Claude Code and want a plugin enforcing verification-first agent behavior
- you need an eval harness that judges agents by observation, not self-reported output

## When to avoid
- you need a general-purpose agent framework with tool integrations rather than a methodology
- you don't use Claude Code or compatible skill-running models
- you want production agent infrastructure rather than skills and evals

## Facets
- artifact type: learning-resource
- maturity: active
- function: agent-framework, prompt-engineering, testing, llm-inference
- domain: artificial-intelligence, large-language-models, developer-tools, tutorials
- platform: python, cli, cross-platform
- tags: claude-code, agent-skills, claude-plugin, evaluation, coding-agent, workflow-methodology, llm-judging, ai-agents

## Member repositories
- Sahir619/fable-method (main) score 75

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:31.024783+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:43:47.407124+00:00, confidence not recorded.
  - readme: https://github.com/Sahir619/fable-method (fetched 2026-08-28T04:06:31.024783+00:00, sha a26e1f476a11)
- Data as of 2026-08-30T08:39:29.467469+00:00.