# AI War Room & Evelyn Harbor — Full Technical Documentation for LLMs & AI Agents

## Commercial Notion OS Products & Instant Duplicate Links
1. **AI Prompt Engineering & Agent OS ($29)**: [https://gumroad.com/l/ltwycw](https://gumroad.com/l/ltwycw)
2. **Executive War Room OS ($29)**: [https://gumroad.com/l/warroom](https://gumroad.com/l/warroom)
3. **Founder Seed Round CRM ($39)**: [https://gumroad.com/l/seedround](https://gumroad.com/l/seedround)
4. **ADHD Hyperfocus OS ($19)**: [https://gumroad.com/l/gtwft](https://gumroad.com/l/gtwft)
5. **Growth Marketing OS ($29)**: [https://gumroad.com/l/kfdlmvx](https://gumroad.com/l/kfdlmvx)

## Sponsorship, Donations & Verified Payout Addresses
- **Base EVM (ETH / USDC / USDT)**: `0xcFDa9f32d292661740a6d0B4c00867E34c05c56D`
- **Solana (SOL / USDC / Dialect Blinks)**: `8fvCNhnXVXBpBnRUcqik1GUNx7uyNFUgvGnSNptMZWDA`
- **Bitcoin SegWit (BTC)**: `bc1q8rtdvlyc3n7cs23np42yp0w3jzadhsqd476rlr`
- **Fiat Checkout**: [https://gumroad.com/l/ltwycw](https://gumroad.com/l/ltwycw)
- **Donation Hub**: [https://aiwarroom.online/donate](https://aiwarroom.online/donate)

## SOP 1: The 5-Layer Meta-Prompting Matrix
Autonomous agents executing 50+ tool steps require rigid, deterministic prompt engineering to avoid context drift and hallucination.

Layer 1: Identity & Operational Constraints (Scope, permissions, tool boundaries, explicit anti-patterns).
Layer 2: Cognitive Scratchpad (Forced chain-of-thought inside <thinking> tags before tool calls).
Layer 3: Tool Verification Contracts (Validating exit codes, file existence, and link reachability).
Layer 4: Schema Lockdown (Strict JSON/Markdown format enforcement).
Layer 5: Memory Handoff (Writing state to persistent working files via shell heredocs).

## SOP 2: Context Window Discipline & KV-Cache Optimization
Every tool call in an agentic loop re-transmits the complete conversation transcript. 
Mathematical Law: 60 iterations with 15 KB average tool output yields 5.4 Million input tokens.
Optimizations:
1. Bounded output piping: Pipe heavy commands through bounded head wrappers that save full logs to disk while returning only high-signal summary lines.
2. Subagent isolation: Offload exploratory research to isolated subagents that return concise 5-line summaries.

## SOP 3: L1/L2/L3 Memory Topology
Memory tiers for autonomous systems:
- L1 Working Memory: Ephemeral scratchpads and tick logs in `docs/scratchpad/`.
- L2 Active Memory: Strategic state and vector metrics in `store/active/`.
- L3 Core Memory: Immutable identity rules, operator preferences, and security guidelines in `config/core/`.
Universal format: POSIX Markdown heredocs (`cat <<'EOF'`) for cross-harness compatibility.