Skip to content

Latest commit

 

History

History
55 lines (42 loc) · 3.1 KB

File metadata and controls

55 lines (42 loc) · 3.1 KB

Pattern Catalog

This catalog tracks candidate runtime patterns that appear to recur across coding agents. Each pattern page should distinguish observed convergence from proposed standardization.

Current Opinionated Summary

If this repository had to summarize the emerging pattern in one sentence, it would say this:

The most reusable coding-agent runtime looks like a guarded tool-result loop with explicit completion, recoverable failures, explicit runtime state, and policy boundaries around tool use and stopping.

The patterns that currently look closest to standardization are:

  • tool-result loop,
  • explicit completion action,
  • recoverable tool errors,
  • command-output streaming,
  • a minimal state model that the UI can read without transcript parsing.

For a prose overview rather than a matrix, read synthesis/standardization-view.md.

Comparative Matrix

Pattern Or Capability Roo Code Cline Claude Code Copilot Cursor OpenHands Confidence
Tool-result loop Strong Strong Strong Medium Medium Strong High
Explicit completion action Strong Strong Medium Medium Medium Mixed High
Guarded completion Strong Strong Medium Medium Medium Mixed Medium
Recoverable tool errors Strong Strong Strong Medium Medium Strong High
Separate runtime state machine Strong Medium Medium Strong Strong Medium Medium
Context compaction Medium Medium Strong Medium Medium Medium Medium
Resumability Medium Medium Strong Medium Medium Medium Emerging
Bounded subagents Medium Strong Strong Medium Medium Mixed Medium
Hooks and policy injection Medium Medium Strong Medium Medium Medium Medium
Command-output streaming Strong Strong Strong Strong Strong Medium High
Journal and telemetry events Medium Medium Strong Strong Strong Medium Emerging

Prose Synthesis

  • High-confidence convergence appears strongest around the loop itself, explicit completion semantics, recoverable failure handling, and streaming of tool activity.
  • Medium-confidence convergence appears around state modeling, subagent bounds, hook insertion, and compaction behavior, where implementations differ but the need keeps recurring.
  • Emerging patterns include explicit journaling contracts, resumability guarantees, and stop-hook validation as a reusable primitive.
  • The repo treats these patterns as hypotheses to be refined through system reviews rather than as fixed doctrine.

Pattern Pages