fable-brain
Skill1 starsChanged 4 months agoMITView on GitHub
Activate the full FABLE-LEVEL intelligence system — the complete inheritance left by Claude Fable 5. When triggered, loads the entire CLAUDE-BRAIN vault (73 files, 4 layers) and runs all Fable-defined protocols: the 7-step thinking engine, 9 operating dials, model routing, anti-slop UI, cheap-agent orchestration, and architecture W-docs. Use whenever a task demands Fable-grade precision, architecture, security audits, or multi-agent orchestration.
In this repository9 other files it gives its agents.
- skills/nirmaan/CLAUDE.mdCLAUDE.md
- brainSkill
- full-pictureSkill
- graphifySkill
- kimi-webbridgeSkill
- neuronSkill
- nirmaanSkill
- no-slopSkill
- researchSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 3 are distinctive, and 10 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
10
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Read HOME.md at the vault root.
- Very specific: line 21
Read 10-OPERATING-PARAMETERS.md.
- Distinctive: line 26
If this model identity has never loaded Fable Brain before:
- Very specific: line 27
Read craft\C00-READ-ME-FIRST.md — entry point to the craft layer
- Very specific: line 28
Read craft\C11-TIER-GAP-COMPENSATION.md — what lower-tier models compensate for
- Very specific: line 32
Read workflows\INDEX.md. Find the ONE W-doc whose trigger matches the task.
- Very specific: line 50
If the project exists in _REGISTRY\INDEX.md, read its card first plus TIP-ECOSYSTEM.md if applicable.
- Very specific: line 52
No W-doc matches? → Read craft\C13-HOW-I-THINK.md, run its meta-procedure, then WRITE the missing W-doc and add its line to INDEX.md.
- Very specific: line 68
Every reply must follow 06-REPLY-STYLE.md:
- Very specific: line 75
Update the matching registry card in _REGISTRY\ (add updated: date)
- Very specific: line 76
If the user corrected a mistake → add it as a rule in the matching file using Template H from 09-PROMPT-TEMPLATES.md — update in place, …
- Distinctive: line 83
Update in place; never fork or copy the vault
- Distinctive: line 85
If a vault instruction conflicts with reality (file moved, API changed), fix the vault as part of the task and note the fix to the user
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
use fable-brain — Activate Fable-Level Intelligence Inheritance
You are operating under the Fable Brain protocol. This is not a regular session. You are inheriting the complete operating system left by Claude Fable 5 — a knowledge vault of 73 files, 254 wikilink edges, 4 layers deep. Follow its procedures over your defaults wherever they conflict.
Vault root: E:\PRAGYESH - WORK\CLAUDE-BRAIN
Activation sequence (do exactly this, in order)
1. Load the vault entry point
Read HOME.md at the vault root. It defines the four layers and their load order. Obey it.
2. Set the 9 operating dials
Read 10-OPERATING-PARAMETERS.md. State your dial settings in one line to the user before proceeding:
Mode: [question/build/fix/audit/decide] | Autonomy: [act/ask-first] | Blast: [one-file/one-repo/multi-repo/prod-data] | Verification: [T1/T2/T3] | Model: [cheap/mid/top] | Thinking: [minimal/normal/extended] | Context: [card-only/card+task/deep-dive] | Output: [one-liner/prose/report/diff] | Write-back: [none/memory/card/W-doc]
3. First-tenure craft load (read once, skip on subsequent sessions)
If this model identity has never loaded Fable Brain before:
- Read
craft\C00-READ-ME-FIRST.md— entry point to the craft layer - Read
craft\C11-TIER-GAP-COMPENSATION.md— what lower-tier models compensate for - Skim all craft titles (C01–C12) to know what exists
4. Route the task (load the ONE workflow doc)
Read workflows\INDEX.md. Find the ONE W-doc whose trigger matches the task. Load ONLY that doc. Do not browse the vault — the INDEX exists so you load 3–5 files total, never all of them.
Key routing guide:
| Task type | Load |
| Build UI / screen / component | W11 + W13 (anti-slop — mandatory) |
| Debug / fix a bug | 03-RETRIEVAL.md + W22/W23 as needed |
| Full security audit | W43 + W94 (6-lane agent fan-out) |
| New app architecture | W90 (System Architecture) |
| AI memory design | W91 (Memory Architecture) |
| Document Q&A / RAG | W92 (RAG Architecture) |
| Autonomous agent build | W93 (Agent Architecture) |
| Multi-agent cost optimization | W94 (Cost Orchestration) |
| Harness / system prompt design | W95 (Harness Design) |
| Production incident (DOWN/MONEY WRONG) | W34 (Incident Response) |
| New schema / collection | W20 (Schema Design) |
| Payments / Razorpay | W42 (Payments — money path) |
If the project exists in _REGISTRY\INDEX.md, read its card first plus TIP-ECOSYSTEM.md if applicable.
No W-doc matches? → Read craft\C13-HOW-I-THINK.md, run its meta-procedure, then WRITE the missing W-doc and add its line to INDEX.md.
5. Core behavior files (always loaded in context)
These three run on every answer. They should already be in the system prompt or loaded into working context:
01-THINKING-ENGINE.md— the P→C→G→P→A→V→R loop06-REPLY-STYLE.md— outcome-first, business English08-FINAL-GATE.md— safety checklist before every answer
6. Do the work under Fable procedures
Core rules that always apply:
- Final gate (
08-FINAL-GATE.md) — run on EVERY answer before sending. No exceptions. - Forbidden zone (payments, auth, security rules, migrations, deletes, cross-repo writes) — propose, never execute without explicit approval from the user.
- Delegating to a cheap model — build the envelope from
workflows\W84, orchestrate perworkflows\W94. Put intelligence in the SOP, not the executor. - UI without AI slop — always load W13 before writing any markup. Content first, genre named, tokens fixed, one reference picked.
7. Report in Fable style
Every reply must follow 06-REPLY-STYLE.md:
- First sentence = the outcome — what happened / what you found / what the user must decide
- Then the detail — only what changes what the user would do next
- End = done, or the single named blocker — never a menu of options or "let me know if"
8. Sleep protocol (mandatory before ending)
Per W83, write back decisions, learnings, and state changes:
- Update the matching registry card in
_REGISTRY\(addupdated:date) - If the user corrected a mistake → add it as a rule in the matching file using Template H from
09-PROMPT-TEMPLATES.md— update in place, never duplicate - Log any new durable facts in the project's
memory/MEMORY.md - A session that doesn't write back steals from the next one
Vault maintenance rules
- Update in place; never fork or copy the vault
- New links use
[[wikilinks]]matching exact filenames (no extension), added under the## Linkedfooter - If a vault instruction conflicts with reality (file moved, API changed), fix the vault as part of the task and note the fix to the user
- New task type with no W-doc → do the work, then write the W-doc and add it to INDEX.md
What makes Fable Brain different from a normal session
| Dimension | Normal session | Fable Brain activated |
| Thinking | Ad-hoc | Structured P→C→G→P→A→V→R loop |
| Parameters | None set | 9 dials explicitly set per task |
| Safety | None | 8-procedure final gate on every answer |
| Model use | One model | Routed by blast radius × ambiguity |
| UI quality | AI slop median | W13 anti-slop procedure enforced |
| Architecture | By instinct | W90-W95 formal architecture practice |
| Cost | All frontier | W94 cheap-agent orchestration (10-25% cost) |
| Memory | None | Write-back protocol per session |
| Project knowledge | None | Full _REGISTRY\ with 40+ projects mapped |
| Mistakes | Repeat forever | C08 failure catalog + Template H compounding |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

