Frontier habits, any model.

In mid-2026, Claude Fable 5 shipped hundreds of merged PRs in daily plan/build/review work. What made the output better wasn't raw intelligence. It was a repeatable set of operating habits. This plugin is those habits, distilled into six skills any capable model can run.

Install the plugin See the six skills

GRILL → SPEC → PLAN → BUILD → REVIEWNEVER BUILD BEFORE THE GATEDECISIONS LIVE ON DISKVERDICTS NEED LIVE PROBESDEGRADES HONESTLY ON WEAKER MODELSNOT AN OFFICIAL ANTHROPIC PRODUCT GRILL → SPEC → PLAN → BUILD → REVIEWNEVER BUILD BEFORE THE GATEDECISIONS LIVE ON DISKVERDICTS NEED LIVE PROBESDEGRADES HONESTLY ON WEAKER MODELSNOT AN OFFICIAL ANTHROPIC PRODUCT

01 / THE SIX SKILLS

001 fable-exe-mode The full operating character: judgment, planning, orchestration, verification, output discipline. Stays on for the session. SAY “FABLE.EXE MODE” 002 fable-exe-grill Structured facilitation: the staged interrogation, the war-game, the fork batch. Always ends in a decision ledger on disk. SAY "GRILL ME" 003 fable-exe-spec The two-tier spec pipeline: intake grill → complete design spec → review package for your smartest model → ingest its replacement spec. SAY "SPEC THIS OUT" 004 fable-exe-plan A staged plan-of-record a cold session can execute: walking skeleton first, every step carries a ground-truth Verify and a scope Fence. SAY "PLAN THIS BUILD" 005 fable-exe-review A verdict backed by live probes run this session: claim ladder, adversarial refutation, explicit confidence marks. Never a vibe read. SAY "IS THIS ACTUALLY DONE" 006 fable-exe-delegate The model-boundary crossing, both directions: dispatch with an explicit return path, ingest returned papers as claims, attribute who did what. SAY "DELEGATE THIS"

Your daily model runs the skills. Your smartest model gets a paper, never a skill.

The loop for a big idea: you and your daily model grill and draft the spec. Your smartest model reviews it and returns one complete replacement. Your daily model ingests, plans, and builds, and fable-exe-review gates the ship. Only one model? Run the review in a fresh session at high effort and say who reviewed it. Never skip the gate silently.

Through-line

grill → spec → plan → build → review

The hard rule

Never build before the gate LOCKED

Done means

Evidence, not claims LIVE PROBES

Weaker models

Shorter and more conservative. NEVER BLUFFING

Plays with

Layers on top of your existing CLAUDE.md and repo rules, never overrides them

02 / INSTALL

Two commands.

With Claude Code installed:

claude plugin marketplace add uhhexe/fable.exe
claude plugin install fable-exe@fable-exe

Then start a new session and say “fable.exe mode” to check it is live. Update later with claude plugin update fable-exe@fable-exe. Inside a session, /plugin marketplace add uhhexe/fable.exe works too.

Upgrading from v1? Remove fable@fable and the old fable marketplace once, then run the two commands above. Your project files stay untouched.

Never build before the gate.

MIT licensed. The plugin's own design spec was written with its own pipeline. Read docs/SPEC.md.

View on GitHub