7/ 14·feed
LEGENDARY
TIER 5 / 5
SYNCED13d agoLast re-publish — creators re-run the publish prompt to sync their rig.
claude-code·Fable 5.1·ready
wall-of-glass
TOP 1%
SAMPLESample rig — an illustrative example seeded by howiclaude, not a real user submission. Rank is hidden because there's no real ranked population yet.
91
/ 100Efficiency score 91 of 100 — derived from analysis sub-scores (context discipline, automation depth, claude-md specificity, parallelism, permission hygiene)Top 1%·—·2 scored rigsBand is a fixed rubric curve for now — 2 real scored rigs on the site; true percentiles unlock with population.
EXPORT copies a prompt — paste it into your own Claude Code to install this rig.
TOKENS / MO38.0 MMonthly token burn — self-reported in the publish interview, not measured by the site.
SPEND / MO$220Monthly Claude spend — self-reported in the publish interview.
PLANMAX 20×Subscription tier — self-reported in the publish interview.
INSTANCES3Concurrent Claude Code instances they typically run — self-reported.
VOICESOMEShare of prompting done by voice instead of keyboard — self-reported.
SINCENOV ’24When they started using Claude Code — self-reported.
MODEL MIX
FABLE 55% · OPUS 30% · SONNET 15%
FABLE 55%OPUS 30%SONNET 15%MODEL MIX — share of tokens by model family, straight from the rig's published usage data.
FABLE 55% · OPUS 30% · SONNET 15%
MODELFable 5.1MODEL — The default model this rig runs (/model, or the `model` setting). Read from settings when published with the 2026-09 prompt; otherwise inferred from the usage mix.
Fable 5.1: Claude Fable 5.1 — the Mythos-class tier above Opus; 1M context, thinking always on; the `fable` alias since 2026-09-01.
Fable 5.1: Claude Fable 5.1 — the Mythos-class tier above Opus; 1M context, thinking always on; the `fable` alias since 2026-09-01.
EFFORTxhighEFFORT — Effort level — how deeply Claude reasons per task (/effort, `effortLevel`). low → medium → high → xhigh → max; ultracode = xhigh + dynamic workflows; auto lets Claude pick.
xhigh: Deeper reasoning — the best setting for hard agentic work on the 5-class models.
xhigh: Deeper reasoning — the best setting for hard agentic work on the 5-class models.
PERMSplanPERMS — Permission mode — `permissions.defaultMode`. Auto mode (a background safety classifier approves safe actions) became Claude Code's default on 2026-08-14; Manual asks each time; plan is read-only until you approve a plan.
plan: Plan mode — read-only exploration → written plan → your approval → edits.
plan: Plan mode — read-only exploration → written plan → your approval → edits.
FLEET3× iTerm2 · teamsFLEET — How many Claudes run at once, and what orchestrates them — tmux / cmux grids, the agent view (`claude agents`), the desktop app, or a GUI manager. Agent teams and cross-session messaging (2026-08) show here too.
their window layout┌─ secondary (L) ──┬─── main ────────────┬─ portrait (R) ─┐ │ monitoring / logs │ code │ claude │ docs / Notion │ └───────────────────┴───────┴────────────┴────────────────┘
their window layout┌─ secondary (L) ──┬─── main ────────────┬─ portrait (R) ─┐ │ monitoring / logs │ code │ claude │ docs / Notion │ └───────────────────┴───────┴────────────┴────────────────┘
LOOPS1 · /goalLOOPS — Standing automations — /loop, /goal, routines (/schedule), saved dynamic workflows, Stop-hook loops. "My job is to write loops." One card per loop lives in the LOOPS deck below.
MEMORYauto memory · 180-line CLAUDE.mdMEMORY — What Claude remembers between sessions: auto memory (on by default since mid-2026, MEMORY.md + topic files) and the size of the global CLAUDE.md — the 2026 flex is a small one.
VERIFYPlaywright + a verifier subagent + /code-review before every mergeVERIFY — How this rig verifies Claude's work — the #1 official best practice in 2026: tests, /verify, /code-review, Stop hooks, Chrome or a simulator pane. In their words when published; otherwise inferred.
INPUTvoiceINPUT — How prompts get in — keyboard, voice (/voice hold or tap, or a dictation app), or both.
voice: Voice dictation via Claude Code's built-in /voice (or a dictation app like Wispr / Superwhisper).
voice: Voice dictation via Claude Code's built-in /voice (or a dictation app like Wispr / Superwhisper).
PHYSICAL · LOADOUT12 ITEMS
SHELLzsh · default·
MACHINEMac Studio M2 Ultra · 128GB·
EDITORVS Code·
+ DESK GEAR
BUILD · CLAUDE CODE CONFIG0 SKILLS · 0 PLUGINS · 0 HOOKS · 0 MCP · 1 SURFACES
HOOKS5+PostToolUse / PreToolUse / Stop
SURFACEClaude Code · CLI + Desktop + chromeprimary + secondary
+ the rest unlisted
BADGES
◇ Self-reported — attested in the publish interview.
runs 3+ Claude instances at once
works in git worktrees
5+ custom hooks
standing desk
◇ Self-reported — attested in the publish interview.
runs standing loops — /loop, /goal, routines
◇ Self-reported — attested in the publish interview.
agent teams, cross-session messaging, or a 10+ subagent fan-out as measured by the publish prompt
◇ Self-reported — attested in the publish interview.
daily-drives a Mythos-class model
◇ Self-reported — attested in the publish interview.
70%+ prompt-cache hit rate, as measured by the publish prompt on their machine (published, not site-verified)
◇PARALLEL PROCESSOR
runs 3+ Claude instances at once
PARALLEL PROCESSOR◇ Self-reported — attested in the publish interview.
runs 3+ Claude instances at once
WORKTREE WIZARD
works in git worktrees
WORKTREE WIZARDworks in git worktrees
HOOK MASTER
5+ custom hooks
HOOK MASTER5+ custom hooks
STANDING STRONG
standing desk
STANDING STRONGstanding desk
◇LOOP ENGINEER
runs standing loops — /loop, /goal, routines
LOOP ENGINEER◇ Self-reported — attested in the publish interview.
runs standing loops — /loop, /goal, routines
◇FLEET COMMANDER
agent teams, cross-session messaging, or a 10+ subagent fan-out as measured by the publish prompt
FLEET COMMANDER◇ Self-reported — attested in the publish interview.
agent teams, cross-session messaging, or a 10+ subagent fan-out as measured by the publish prompt
◇FABLE-TIER
daily-drives a Mythos-class model
FABLE-TIER◇ Self-reported — attested in the publish interview.
daily-drives a Mythos-class model
◇CACHE-MAXXER
70%+ prompt-cache hit rate, as measured by the publish prompt on their machine (published, not site-verified)
CACHE-MAXXER◇ Self-reported — attested in the publish interview.
70%+ prompt-cache hit rate, as measured by the publish prompt on their machine (published, not site-verified)
STATUSLINE · THEIR TERMINALTheir real statusline — captured from their settings.json statusLine by the publish prompt, shown as published.click to copy
CONFIG · 3 OF 10 SECTIONS · INDIVIDUALLY FORKABLEINSPECT ALL 3 IN DETAIL →
OPEN
SYSTEM PROMPT
1ITEM
INSPECT →⧉ EXPORT
OPEN
LOOPS
1ITEM
INSPECT →⧉ EXPORT
OPEN
DESK
8ITEMS
INSPECT →↵
FREE SLOT
YOUR RIG HERE
+
PUBLISH YOURS →↵