Token Saving Assistant

Problem it solves

ShengToken Assistant targets 'too many turns, context bloat, and fast credit burn' in AI chat, content creation, and Skill debugging. It is not a generic tool caller; it is a set of constraints for prompting, templating, task archiving, and model selection. It turns vague requests into executable prompts, compresses repeated trial-and-error into one output pass, and clears context after a task ends.

How it works

The skill activates on triggers such as 'how to save tokens,' 'help me optimize my prompt,' 'batch write,' or 'archive,' then gives concise guidance:

• Precise targeting: for code or file tasks, provide path, line number, and target function so the model spends less time searching.
• One-shot clarity: use a 'goal + style + format + constraints' template to reduce revision loops.
• Output control: require Markdown, lists, tables, or no explanatory filler to avoid oversized responses.
• Batch creation: group multiple copywriting, short posts, or Q&A items into one task.
• Timely archiving: write finished state to a memory file or working directory, then reload it when resuming instead of carrying an ever-growing context.

It also includes seven task templates and model-selection guidance, such as using inference-heavy models for debugging and lower-cost models for simple Q&A.

Boundaries

It fits engineers, creators, and Skill developers who know the desired output and want to control conversation cost. It is not ideal for autonomous exploration of unknown systems, long unattended runs, or complex multi-tool orchestration. Archiving requires a writable working directory, and actual token usage still varies with task complexity, model, and context length.