Research Pack 2026-07-07

Every entry below was fetched and verified on the retrieved date; claims derived from them were adversarially checked (see the claim ledger). Grouped by source type. Refresh: official 30 days, practitioner 90 days.

Official sources

• Claude Fable (retrieved 2026-07-07) [official] [evidence-based]
• Making Claude Code more secure and autonomous with sandboxing (retrieved 2026-07-07, published 2025-10-20) [official] [evidence-based]
• More details on Fable 5’s cyber safeguards and our jailbreak framework (retrieved 2026-07-07, published 2026-07-02) [official] [evidence-based]
• Project Glasswing (retrieved 2026-07-07, published 2026-04-07) [official] [evidence-based]
• Claude Fable 5 and Claude Mythos 5 (retrieved 2026-07-07, published 2026-06-09) [official] [evidence-based]
• Redeploying Claude Fable 5 (retrieved 2026-07-07, published 2026-06-30) [official] [evidence-based]
• Anthropic’s Responsible Scaling Policy (retrieved 2026-07-07, published 2026-05-26) [official] [evidence-based]
• Anthropic Usage Policy (retrieved 2026-07-07, published 2025-09-15) [official] [evidence-based]
• Building Effective Agents (Anthropic Engineering) (retrieved 2026-07-07, published 2024-12-19) [official] [evidence-based]
• Best practices for Claude Code (official docs, formerly anthropic.com/engineering/claude-code-best-practices) (retrieved 2026-07-07) [official] [evidence-based]
• Run Claude Code programmatically (non-interactive mode, official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Hooks reference (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• How Claude remembers your project: CLAUDE.md and auto memory (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Claude Code Overview (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Choose a permission mode (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Configure the sandboxed Bash tool (Claude Code Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Security (Claude Code Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Claude Code settings (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Extend Claude with skills (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Create custom subagents (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Use Claude Code in VS Code (official docs) (retrieved 2026-07-07) [official] [evidence-based]
• Building a C compiler with a team of parallel Claudes (Anthropic Engineering) (retrieved 2026-07-07, published 2026-02-05) [official] [evidence-based]
• Compaction (Claude Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Context editing (Claude Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Managing context on the Claude Developer Platform (announcement) (retrieved 2026-07-07, published 2025-09-29) [official] [evidence-based]
• Context windows (Claude Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Effective context engineering for AI agents (Anthropic Engineering) (retrieved 2026-07-07, published 2025-09-29) [official] [evidence-based]
• Demystifying evals for AI agents (Anthropic Engineering) (retrieved 2026-07-07, published 2026-01-09) [official] [evidence-based]
• Effective harnesses for long-running agents (Anthropic Engineering) (retrieved 2026-07-07, published 2025-11-26) [official] [evidence-based]
• Harness design for long-running application development (Anthropic Engineering) (retrieved 2026-07-07, published 2026-03-24) [official] [evidence-based]
• Scaling Managed Agents: Decoupling the brain from the hands (Anthropic Engineering) (retrieved 2026-07-07, published 2026-04-08) [official] [evidence-based]
• How we built our multi-agent research system (Anthropic Engineering) (retrieved 2026-07-07, published 2025-06-13) [official] [evidence-based]
• Memory tool (Claude Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Prompt caching (Claude Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Prompting best practices, including long context and multi-window workflows (Claude Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Agent Skills - Claude Platform Docs (overview) (retrieved 2026-07-07) [official] [evidence-based]
• Prompting best practices (retrieved 2026-07-07) [official] [evidence-based]
• Customize Claude Code with plugins - Claude blog announcement (retrieved 2026-07-07, published 2025-10-09) [official] [evidence-based]
• Prompt engineering overview (retrieved 2026-07-07) [official] [evidence-based]
• Create plugins - Claude Code Docs (retrieved 2026-07-07) [official] [evidence-based]
• Prompting Claude Fable 5 (retrieved 2026-07-07) [official] [evidence-based]
• Connect Claude Code to tools via MCP - Claude Code Docs (retrieved 2026-07-07) [official] [evidence-based]
• Building agents with the Claude Agent SDK (retrieved 2026-07-07, published 2025-09-29) [official] [evidence-based]
• Equipping agents for the real world with Agent Skills (Anthropic Engineering) (retrieved 2026-07-07, published 2025-10-16) [official] [evidence-based]
• Writing effective tools for agents, with agents (Anthropic Engineering) (retrieved 2026-07-07, published 2025-09-11) [official] [evidence-based]
• Claude Mythos (retrieved 2026-07-07) [official] [evidence-based]
• System Card: Claude Fable 5 and Claude Mythos 5 (retrieved 2026-07-07, published 2026-06-09) [official] [evidence-based]
• System Card: Claude Mythos Preview (retrieved 2026-07-07, published 2026-04-07) [official] [evidence-based]
• Project Glasswing initial update (retrieved 2026-07-07, published 2026-05-22) [official] [evidence-based]
• Expanding Project Glasswing (retrieved 2026-07-07, published 2026-06-02) [official] [evidence-based]
• Claude Mythos Preview (retrieved 2026-07-07, published 2026-04-07) [official] [evidence-based]
• Why Claude switched models in your conversation with Fable 5 (retrieved 2026-07-07) [official] [evidence-based]
• Real-time cyber safeguards on Claude Opus and Sonnet (retrieved 2026-07-07) [official] [evidence-based]

Primary sources

• anthropics/claude-code GitHub repository (retrieved 2026-07-07) [primary] [evidence-based]
• Code execution with MCP - Anthropic Engineering (retrieved 2026-07-07, published 2025-11-04) [primary] [evidence-based]
• Claude Fable 5 system prompt export (claude.ai session) (retrieved 2026-07-07, published 2026-07-07) [primary] [evidence-based]
• Anthropic coordinated vulnerability disclosure dashboard (retrieved 2026-07-07, published 2026-05-22) [primary] [evidence-based]

Vendor sources

• Anthropic Claude Fable 5 on AWS: Mythos-class capabilities with built-in safeguards now available (retrieved 2026-07-07, published 2026-06-09) [vendor] [practitioner]

Api-Docs sources

• Introducing Claude Fable 5 and Claude Mythos 5 - Claude Platform Docs (retrieved 2026-07-07) [api-docs] [practitioner]
• Models overview - Claude Platform Docs (retrieved 2026-07-07) [api-docs] [practitioner]
• Refusals and fallback - Claude Platform Docs (retrieved 2026-07-07) [api-docs] [practitioner]
• Extended thinking (retrieved 2026-07-07) [api-docs] [practitioner]
• Adaptive thinking (retrieved 2026-07-07) [api-docs] [practitioner]
• Effort (retrieved 2026-07-07) [api-docs] [practitioner]
• Tool use with Claude (retrieved 2026-07-07) [api-docs] [practitioner]
• Define tools (retrieved 2026-07-07) [api-docs] [practitioner]
• Agent SDK overview (Claude Code docs) (retrieved 2026-07-07) [api-docs] [practitioner]
• Claude model migration guide (retrieved 2026-07-07) [api-docs] [evidence-based]

Authority sources

• US lifts restrictions on Anthropic’s powerful AI models Fable and Mythos (retrieved 2026-07-07, published 2026-07-01) [authority] [practitioner]
• Model Context Protocol Specification (revision 2025-11-25) (retrieved 2026-07-07, published 2025-11-25) [authority] [practitioner]
• modelcontextprotocol/servers - MCP reference server implementations (retrieved 2026-07-07) [authority] [practitioner]
• Agent Skills Overview - agentskills.io (open standard) (retrieved 2026-07-07) [authority] [practitioner]
• MCP Security Best Practices (spec 2025-11-25) (retrieved 2026-07-07, published 2025-11-25) [authority] [practitioner]

Practitioner sources

• Claude Code Hooks Explained: The Deterministic Layer Around Your Agent (Blake Crosley) (retrieved 2026-07-07, published 2026-07-01) [practitioner] [practitioner]
• Context Engineering for Agents (LangChain blog) (retrieved 2026-07-07, published 2025-07-02) [practitioner] [practitioner]
• Claude Fable 5 Prompting Guide: Migration Notes for Long Tasks, Agents, and High Effort (retrieved 2026-07-07, published 2026-06-10) [practitioner] [practitioner]
• Claude Code Advanced Best Practices: 11 Practical Techniques for Hooks, Subagents and Context Management (SmartScope) (retrieved 2026-07-07, published 2026-07-06) [practitioner] [practitioner]
• Initial impressions of Claude Fable 5 (retrieved 2026-07-07, published 2026-06-09) [practitioner] [practitioner]
• Claude Fable 5 and Mythos 5: The System Card (Zvi Mowshowitz) (retrieved 2026-07-07, published 2026-06-12) [practitioner] [practitioner]

Coverage

• 96 sources total; 81 official, primary, vendor, authority, or API-doc sources.
• 187 claim rows kept, including 5 contested rows, 20 Mythos or Glasswing rows added in the deep research wave, and 7 GitHub prompt-leak sweep rows.
• Corpus: 40 line-referenced claims from System Prompt Export 2026-07.

Behavioral and thinking sources (added 2026-07-07, wave 2)

• Adaptive thinking (Claude Platform Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Effort parameter (Claude Platform Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Model configuration (Claude Code Docs) (retrieved 2026-07-07) [official] [evidence-based]
• Claude’s Character (retrieved 2026-07-07, published 2024-06-08) [official] [evidence-based]
• Claude’s Constitution (retrieved 2026-07-07) [official] [evidence-based]
• Building safeguards for Claude (retrieved 2026-07-07, published 2025-08-12) [official] [evidence-based]
• Protecting the wellbeing of our users (retrieved 2026-07-07, published 2025-12-18) [official] [evidence-based]
• How people use Claude for support, advice, and companionship (retrieved 2026-07-07, published 2025-06-27) [official] [evidence-based]
• Measuring political bias in Claude (retrieved 2026-07-07, published 2025-11-13) [official] [evidence-based]
• Aligning on child safety principles (retrieved 2026-07-07, published 2024-04-23) [official] [evidence-based]
• System Card: Claude Fable 5 and Claude Mythos 5 (PDF) (retrieved 2026-07-07, published 2026-06-09) [official] [evidence-based]

System prompt corroboration sources (added 2026-07-07, wave 3)

• Anthropic release notes: system prompts (official Fable 5 claude.ai prompt) (retrieved 2026-07-07, published 2026-06-09) [official] [evidence-based]
• Piebald-AI claude-code-system-prompts (mechanical extraction), link omitted for public release (retrieved 2026-07-07, published 2026-07-06) [practitioner] [practitioner]
• Community Claude Fable 5 capture lineage, link omitted for public release (retrieved 2026-07-07, published 2026-06-10) [practitioner] [practitioner]
• Community Claude Fable 5 mirror, link omitted for public release (retrieved 2026-07-07, published 2026-06-09) [practitioner] [practitioner]
• Community fable5-system-prompt behavioral verification, link omitted for public release (retrieved 2026-07-07, published 2026-06-15) [practitioner] [practitioner]
• AlphaSignal: Fable 5 prompt leak analysis, link omitted for public release (retrieved 2026-07-07) [practitioner] [practitioner]

GitHub prompt-leak sweep (added 2026-07-07, wave 5)

• GitHub repository search: claude-fable-5 system prompt, URL omitted for public release (retrieved 2026-07-07) [primary] [practitioner]
• GitHub repository search: claude-mythos-5 system prompt, URL omitted for public release (retrieved 2026-07-07) [primary] [practitioner]
• GitHub repository search: Claude Mythos leak, URL omitted for public release (retrieved 2026-07-07) [primary] [practitioner]
• Mislabeled Mythos system prompt community copy, link omitted for public release (retrieved 2026-07-07, published 2026-06-11) [practitioner] [practitioner]
• Community Claude Fable 5 system-prompt archive, link omitted for public release (retrieved 2026-07-07) [practitioner] [practitioner]
• HalalifyMusic/fable-mode, link omitted for public release (retrieved 2026-07-07) [practitioner] [practitioner]
• kyegomez/OpenMythos, link omitted for public release (retrieved 2026-07-07) [practitioner] [practitioner]
• alexey-max-fedorov/claude-mythos-preview, link omitted for public release (retrieved 2026-07-07) [practitioner] [practitioner]