skill-zipper

To reduce token usage losslessly, restructure skills by moving always-loaded content to on-demand files and removing model-known redundancy. This compression trims size while preserving every instruction, verified by a diff script.

Skills fail to trigger due to poor frontmatter descriptions. You can fix trigger accuracy by rewriting the description using a specific rubric, ensuring the skill loads on relevant turns and skips irrelevant ones without altering behavior.

Lossless skill restructuring is the process of compressing and reorganizing skill content to reduce token costs without changing behavior. It applies operations like encapsulate, enrich, and harden, producing a token delta report and a diff proof.

Yes, you can audit cross-runtime portability by applying a checklist that aligns your skills with the open standard's portable core fields. This ensures compatibility across different runtimes while restructuring content for reliability.

Yes, tiktoken is required as a dependency to measure token efficiency accurately. It calculates the token delta report, enabling you to verify that content compression and progressive disclosure effectively reduce always-loaded token cost.

The best way to fix vague instructions is to apply enrichment and hardening operations during skill restructuring. This tightens rules to ensure consistent model behavior while generating a lossless diff proof to verify no instructions were lost.