Fable 5 Brain

Public, source-cited Obsidian brain for Claude Fable 5, Claude Mythos 5, Project Glasswing, Claude Code, and current agent-operations practice.

This is a public research repository. It is not affiliated with, endorsed by, or sponsored by Anthropic, and its content is derived from public sources.

Public research repository. Not affiliated with, endorsed by, or sponsored by Anthropic. Content is derived from public sources. Raw operator captures, private notes, API credentials, and account-specific exports are omitted from this public release.

Status

• Maturity: market-ready by python scripts/audit_brain.py --require market-ready
• Current wiki pages: 109
• Current source ledger entries: 96
• Current claim-ledger rows: 187
• Canonical operator entrypoint: SKILL.md
• Canonical vault entrypoint: wiki/hot.md

What This Is

Fable 5 Brain is a working knowledge vault and agent operating layer. It turns dated public sources, public-safe corpus metadata, and verified claims into:

• Obsidian notes with frontmatter, wikilinks, source URLs, and confidence tags.
• Operator playbooks for Claude Fable 5, Mythos-class models, and Claude Code.
• Source and claim ledgers that make factual drift visible.
• Audit scripts that gate the brain before it is called market-ready.
• A sanitized static landing page in site/ for GitHub Pages-style publishing.

The brain is advisory. It does not mutate accounts, production systems, pipelines, books, customer records, or live infrastructure.

Knowledge Graph

Every note carries frontmatter, wikilinks, and source citations, so the vault forms one densely connected Obsidian graph. Navigation stays fast and factual drift stays visible.

Buyer, Outputs, Boundaries

The buyer for this repository is an operator, research owner, or agent team that needs a verifiable second brain for Fable, Mythos, Claude Code, and related agent practice.

The expected outputs are a navigable Obsidian vault, source and claim ledgers, operator playbooks, audit scripts, reusable templates, and a sanitized GitHub Pages landing page.

The boundaries are explicit: this brain is advisory, source-cited, and read-only by default. It does not grant rights to third-party material, publish raw captures, make legal claims, or automate changes to live systems without a separate approval and rollback process.

Publish Safety

This repository contains research about public prompt material and public discussion of prompt mirrors. Public availability does not automatically grant copyright, trademark, confidentiality, or redistribution rights. This is not legal advice.

Rules before pushing or publishing:

• Keep .raw/ source captures out of Pages, releases, and public ZIPs.
• Do not publish API keys, account exports, cookies, tokens, private user data, or local machine secrets.
• Keep GitHub Pages sanitized. The site/ folder is designed for public-safe summary content only.
• Treat Anthropic, Claude, Fable, Mythos, and Project Glasswing names as third party marks and use them only descriptively.

See PUBLISHING_NOTICE.md for the rights and provenance policy.

Repository Map

.
|-- .github/                 # manual GitHub Pages workflow
|-- .claude-plugin/          # Claude Code plugin and marketplace metadata
|-- SKILL.md                 # agent entrypoint and operating procedure
|-- CODEX.md                 # vault runtime instructions
|-- wiki/                    # Obsidian brain, notes, hubs, hot cache
|-- references/              # source ledger, claim ledger, canon, adapters
|-- docs/                    # operator docs and product boundaries
|-- agents/                  # agent templates and handoff material
|-- specs/                   # Brainstein project spec
|-- fable_5_brain/           # Python package and CLI
|-- scripts/                 # audit, package, scaffold, visual/report tools
|-- tests/                   # import/export and pipeline checks
|-- examples/sample-vault/   # deterministic demo vault
|-- assets/                  # template brain assets and SVGs
|-- images/                  # repository cover and graph images
|-- site/                    # sanitized Quartz site and content mirror
|-- install.sh               # local install helper
|-- uninstall.sh             # local uninstall helper
`-- .raw/                    # public stub; raw captures omitted

Folder guides:

• wiki/README.md
• references/README.md
• site/README.md
• .raw/README.md

How It Works

1. Public source material enters references/source-ledger.json; private raw captures stay outside this public release.
2. Claims are verified into references/claim-ledger.md.
3. Wiki notes cite sources and carry confidence tags.
4. Hubs, wiki/hot.md, wiki/index.md, and wiki/overview.md keep navigation current.
5. Audits enforce source coverage, Obsidian hygiene, adapters, tests, docs, and product clarity.

The working doctrine is simple: no source, no claim. No verification path, no release.

Core Topics

• Claude Fable 5 model behavior, pricing, refusals, fallback, and adaptive thinking.
• Claude Mythos 5 and the Mythos-class model tier.
• Project Glasswing, trusted access, and dual-use safety posture.
• Claude Code workflows: CLAUDE.md, hooks, subagents, skills, permissions, sandboxing, MCP, and headless usage.
• Prompting, context engineering, source-led research, and multi-agent fan-out.
• Public prompt-copy corroboration and GitHub leak-sweep findings.

Quick Start

python -m pip install -e .
python scripts/lint_vault.py --vault .
python scripts/audit_brain.py --require market-ready

Demo vault:

fable-5-brain demo
fable-5-brain lint --vault examples/sample-vault
fable-5-brain visuals --vault examples/sample-vault
fable-5-brain report --vault examples/sample-vault --html-only

Create a client vault:

fable-5-brain new acme --client-name "Acme Co" --owner "Your Name" --out-dir ~/fable-5-brain-vaults
fable-5-brain ingest --vault ~/fable-5-brain-vaults/acme --file tests/fixtures/sample-source.md
fable-5-brain synthesize --vault ~/fable-5-brain-vaults/acme
fable-5-brain visuals --vault ~/fable-5-brain-vaults/acme
fable-5-brain report --vault ~/fable-5-brain-vaults/acme --html-only
fable-5-brain next --vault ~/fable-5-brain-vaults/acme

Verification

python -m compileall scripts fable_5_brain tests
python tests/test_import_export.py
python scripts/lint_vault.py --vault .
python scripts/audit_brain.py --json
python scripts/audit_brain.py --require market-ready

Release packaging is intentionally stricter and requires a clean reproducible worktree:

python scripts/package_release.py --version 0.1.0
python scripts/package_release.py --version 1.0.0 --release-type market-ready

GitHub Pages

A browsable vault site can be built from site/ and deployed with the manual workflow at .github/workflows/pages.yml. Enable GitHub Pages on the repository first, then set baseUrl in site/quartz.config.ts to match.

The site excludes .obsidian, hot.md, and log.md; the accounts section contains only public-safe tier information. It must never expose raw captures, full prompt material, ledgers, credentials, tokens, or local machine paths.

GitHub's own docs warn that Pages visibility is separate from repository visibility and can make a site public even when the repository is private. Confirm the desired Pages visibility in repository settings before running the manual workflow.

License

Code in this repository is licensed under the MIT License. Vault content is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0), as described in LICENSE-CONTENT.md.

Third-party source material, product names, documentation excerpts, and cited public artifacts remain the property of their respective owners and are not relicensed by this repository.

GitHub Issues

Use GitHub Issues for bugs and repository questions. See SUPPORT.md for support details, SECURITY.md for sensitive reports, and CONTRIBUTING.md for contribution workflow.