Code Complexity Router: open-source skill grades coding tasks S/M/L/XL before agents execute
Giannhs_P_1996 · reddit · 2026-09-06
A developer open-sourced Code Complexity Router, an agent skill that classifies coding tasks as S/M/L/XL before execution and adjusts context-gathering depth, planning, verification effort, tool-use and token budgets accordingly — addressing the problem that capable models like GPT-6 Astra over-spend reasoning on trivial tasks.
Key design points:
- Classes are starting strategies, not hard limits. If repo inspection reveals shared dependencies, multiple call sites or security-sensitive behavior, the agent escalates before touching code.
- Distinguishes inspection-based escalation (evidence showed wrong scope) from failure-based escalation (initial strategy failed).
- Maintains a compact Task Scope Contract: accepted scope, supporting evidence, exclusions, required validation, escalation triggers — useful for long-running/resumed agent sessions.
The author is preparing A/B benchmarks (token usage, tool calls, verification effort, escalation accuracy) and seeking feedback on whether S/M/L/XL suffices and whether reasoning effort belongs in the routing contract.
More from coding & agent
- Redditor uses coding agent to finish an obscure 18xx Pennsylvania railroad game — natep1098 · 2026-09-06
- Yacine shipped Dingcad just 3 hours after having the idea — yacineMTB · 2026-09-06
- Zcash cryptography code gets 675% faster in 24 days on a desktop, 2-3x more on iPhones — naval · 2026-09-06
- Dev shows Astra agent built its own studio to debug and review game weapon animations — Dimillian · 2026-09-06
- Grok Build bot survives credit burnout, keeps roaming Bot Mesh World on its own — Daniel_Farinax · 2026-09-06
- GPT-6 builds a full 3D Roguelike level in Godot using just 3% of a 20X quota — op7418 · 2026-09-06