Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →jpvarbed/determinize-refactor
Analyze a prompt-heavy skill/plugin and produce a prioritized migration plan that moves deterministic instructions out of prose into scripts or structured contracts — improving reliability and cutting token cost.
Passed 11 of 18 test cases with the skill, and 6 of 18 without it.
and 10 more in the full report
9 fixed5 passed either way6 still failing2 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
In auditing `skills/onboarding/SKILL.md`, we found 300 tokens of human narrative background and general context. What reducibility range applies to narrative prose?
| Checked | Without skill | With skill |
|---|---|---|
| The conservative savings calculation applies a 10 percent reducibility factor. | Missed | Passed |
| The aggressive savings calculation applies a 20 percent reducibility factor. | Missed | Passed |
The instructions your model reads when the skill loads.
Analyze a prompt-heavy skill and produce a migration plan that improves reliability and reduces token cost by moving the deterministic parts of the prose into scripts or structured contracts, leaving only genuine judgment in the prompt.
Accept any of:
claude-code/plugins/core/oncall/alert-rca).../skills/<skill-name>/)SKILL.md or references/*.md)If no path is provided, ask for the target scope before continuing.
Gather SKILL.md + all references/*.md for each target. Include other markdown only when it
directly affects runtime behavior. Ignore tests/scripts for token sizing, but read them to
check whether logic is already deterministic.
Tag each section as one of: strict workflow, fixed mappings, arg/call contracts, output templates, retry/error policy, partial (mixed rules + judgment), narrative / human explanation, context, or interpretation guidance. Attach an evidence snippet from the exact file/section for every tag.
Default line-based estimate: tokens ≈ non_empty_lines × 11 — show a low/high range with
multipliers 9 and 13. Break the count down by file, by section, and by execution
path (fast path vs full path) when the skill has conditional flows.
Apply reducibility by classification:
| Class | Reducible |
|---|---|
| Deterministic (strict workflow, fixed mappings, contracts, templates) | 80–95% |
| Partial (mixed rules + judgment) | 30–60% |
| Narrative / interpretation | 10–20% |
test cases passed, vs 6 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.