fablize-for-hermes

What fablize-for-hermes does

fablize-for-hermes is a port of the fablize project, originally a Claude Code plugin, to Hermes Agent's skill system. It keeps the procedures that fablize says were verified: grounding work by running what you build, splitting multi-step tasks into stories with evidence checkpoints, investigating bugs systematically, and refusing to stop early without tool-backed proof.

The repository contains a Hermes skill at skills/fablize/SKILL.md covering foundation rules, a multi-story loop, deep investigation (reproduce, form three or more competing hypotheses, trace the causal chain, verify before and after), verification grounding and an honest escalation rule. A stdlib-only script, goals.py, supports create, next, checkpoint and status commands and stores state in ~/.hermes/fablize/ so it survives session death. Two text packs hold the grounding loop and the investigation protocol, and setup.sh installs the skill. The README is available in English, Chinese and Portuguese.

Key features

• Hermes skill with foundation rules such as leading with the outcome and no promises without tool calls
• goals.py goal engine that refuses completion without evidence
• Investigation protocol: reproduce, competing hypotheses, causal chain, before and after checks
• Verification grounding for HTML, SVG, games, UI and charts
• State saved in ~/.hermes/fablize/
• setup.sh with skill-only, local and global options

When to use it

• Breaking a multi-part build into stories with a verification gate on the last one
• Debugging by recording rejected hypotheses instead of guessing
• Making an agent run and observe a generated page before declaring it done

Who it is for: Hermes Agent users who want the agent to prove its work with evidence on coding and build tasks.

How it fits with Hermes Agent

It is built for Hermes Agent, adapting an existing Claude Code plugin into a Hermes skill with state under ~/.hermes.

FAQ