Open Issues Need Help
View All on GitHub No controlled comparison yet: same tasks with and without the skill about 1 month ago
help wanted
humeicw/dirty-work humeicw/dirty-work Stop context rot. Your main agent hands off the digging - searches, logs, whole files - and gets back only the conclusions its current task needs. Each job goes to the smallest model tier that can do it well, so you burn fewer top-model tokens and the job still gets done right. For Claude Code, Codex and any Agent Skills tool.
Open project →
0
Stop context rot. Your main agent hands off the digging - searches, logs, whole files - and gets back only the conclusions its current task needs. Each job goes to the smallest model tier that can do it well, so you burn fewer top-model tokens and the job still gets done right. For Claude Code, Codex and any Agent Skills tool.
#agent-skills#agentic-workflow#ai-agents#claude#claude-code#claude-skills#codex#context-engineering#context-management#context-rot#context-window#delegation#developer-tools#llm-agents#model-routing#multi-agent#prompt-engineering#sub-agents#subagents#token-efficiency