12 Agentic Harness Patterns derived from Claude Code leak
_jaydeepkarale · x · 2026-08-18
Based on the Claude Code source leak earlier this year, this article summarizes 12 design patterns for building production-grade coding agents. The patterns are categorized into four groups:
Memory & Context
- Persistent Instruction File Pattern: Automatically loads build/test commands and architecture rules via a project-level config.
- Directory Scoped Rules Pattern: Scopes instructions by directory structure.
- Tiered Memory Pattern: Organizes memory into different levels.
- Background Cleanup Pattern: Automatically cleans stale memory.
- Conversation Compression Pattern: Compresses history when the context window fills.
Workflow & Orchestration
- Tool Orchestration.
- Flow Control.
Tools & Permissions
- Permission Sandbox.
- Dynamic Tool Loading.
Automation
- Autonomous Task Decomposition.
- Self-Correction Loop.
- Progressive Delivery.
The article details the use cases and trade-offs for each pattern, providing actionable architectural references for developers.
More from coding & agent
- Sentence Transformers v6.0 Ships Late Interaction Models, Its Largest Update Yet — tomaarsen · 2026-08-18
- sentence-transformers v6.0 adds ColBERT-style multi-vector late interaction retrieval — tomaarsen · 2026-08-18
- Sentence Transformers v6.0: fp16 + FlashAttention Hits 3.87x Over fp32 — tomaarsen · 2026-08-18
- Sentence Transformers v6.0 Arrives: Late Interaction Lands in Project's Biggest Release — tomaarsen · 2026-08-18
- Breaking Changes in Sentence Transformers v6.0: transformers v5 Floor, API Shifts — tomaarsen · 2026-08-18
- bfloat16 Sigmoid Saturates CrossEncoder Rankings; Fix Lifts nDCG From 0.18 to 0.68 — tomaarsen · 2026-08-18