Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Break a plan, spec, or PRD into independently-grabbable task files using tracer-bullet vertical slices.
.claude/skills/serejaris-to-issues/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -29% | 0% |
Break a plan into independently-grabbable task files using vertical slices (tracer bullets).
Work from whatever is already in the conversation context. If PRD.md exists in the project folder root, read it in full. If the user points at another spec file, read that instead.
If you have not already explored the codebase, do so to understand the current state of the code. Task titles and descriptions should use the project's domain glossary vocabulary, and respect ADRs in the area you're touching.
Look for opportunities to prefactor the code to make the implementation easier. "Make the change easy, then make the easy change."
Break the plan into tracer bullet tasks. Each task is a thin vertical slice that cuts through ALL integration layers end-to-end, NOT a horizontal slice of one layer.
<vertical-slice-rules>
</vertical-slice-rules>
Present the proposed breakdown as a numbered list. For each slice, show:
Ask the user:
Iterate until the user approves the breakdown.
For each approved slice, create one markdown file under tasks/ in the project folder.
NN-slug.md where NN is 01, 02, … assigned in dependency order (blockers first, then dependents).Use the task file template below.
<task-template>
Short descriptive title (same as the slice name).
A concise description of this vertical slice. Describe the end-to-end behavior, not layer-by-layer implementation.
Avoid specific file paths or code snippets — they go stale fast. Exception: if a prototype produced a snippet that encodes a decision more precisely than prose can (state machine, reducer, schema, type shape), inline it here and note briefly that it came from a prototype. Trim to the decision-rich parts — not a working demo, just the important bits.
tasks/NN-slug.md references for prerequisite tasks, one per line, or none if the task can start immediately.
</task-template>
When listing dependencies in Depends on, use paths relative to the project folder (e.g. tasks/01-roster-ingest.md). Do not modify PRD.md when writing tasks.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | fail→pass | 21,512 | 16,989 | -21% | 1 | 1 | 0% | 2,417 | 2,668 | +10% | 0 | 0 | — |
case-01 | fail→fail | 11,176 | 9,896 | -11% | 1 | 1 | 0% | 273 | 920 | +237% | 0 | 0 | — |
case-02 | fail→fail | 3,839 | 6,947 | +81% | 1 | 1 | 0% | 279 | 988 | +254% | 0 | 0 | — |
case-03 | fail→fail | 18,969 | 9,414 | -50% | 1 | 1 | 0% | 3,335 | 915 | -73% | 0 | 0 | — |
case-04 | pass→pass | 16,614 | 18,448 | +11% | 1 | 1 | 0% | 2,812 | 2,921 | +4% | 0 | 0 | — |
case-05 | pass→pass | 32,259 | 28,675 | -11% | 1 | 1 | 0% | 4,398 | 4,838 | +10% | 0 | 0 | — |
case-06 | pass→pass | 15,337 | 23,704 | +55% | 1 | 1 | 0% | 2,424 | 3,361 | +39% | 0 | 0 | — |
case-07 | fail→pass | 16,360 | 22,172 | +36% | 1 | 1 | 0% | 2,015 | 2,063 | +2% | 0 | 0 | — |
case-08 | fail→pass | 21,669 | 26,011 | +20% | 1 | 1 | 0% | 2,776 | 3,789 | +36% | 0 | 0 | — |
case-10 | fail→fail | 9,740 | 3,660 | -62% | 1 | 1 | 0% | 1,643 | 1,040 | -37% | 0 | 0 | — |
case-11 | fail→pass | 11,303 | 12,764 | +13% | 1 | 1 | 0% | 1,862 | 2,034 | +9% | 0 | 0 | — |
case-12 | fail→pass | 11,852 | 3,936 | -67% | 1 | 1 | 0% | 1,811 | 1,278 | -29% | 0 | 0 | — |
case-13 | pass→pass | 18,467 | 7,198 | -61% | 1 | 1 | 0% | 2,526 | 1,881 | -26% | 0 | 0 | — |
case-14 | pass→pass | 8,391 | 2,065 | -75% | 1 | 1 | 0% | 1,269 | 1,024 | -19% | 0 | 0 | — |
case-15 | pass→pass | 8,188 | 2,651 | -68% | 1 | 1 | 0% | 1,417 | 1,177 | -17% | 0 | 0 | — |
case-16 | fail→pass | 15,771 | 11,135 | -29% | 1 | 1 | 0% | 2,630 | 2,704 | +3% | 0 | 0 | — |
case-17 | pass→pass | 15,674 | 6,022 | -62% | 1 | 1 | 0% | 2,570 | 1,526 | -41% | 0 | 0 | — |
case-18 | pass→pass | 10,935 | 4,083 | -63% | 1 | 1 | 0% | 1,664 | 1,377 | -17% | 0 | 0 | — |
case-19 | pass→fail | 7,552 | 4,290 | -43% | 1 | 1 | 0% | 1,411 | 1,475 | +5% | 0 | 0 | — |
case-20 | fail→pass | 10,930 | 7,920 | -28% | 1 | 1 | 0% | 1,889 | 2,112 | +12% | 0 | 0 | — |
case-21 | fail→fail | 2,248 | 4,637 | +106% | 1 | 1 | 0% | 253 | 972 | +284% | 0 | 0 | — |
case-22 | pass→pass | 13,829 | 9,000 | -35% | 1 | 1 | 0% | 2,388 | 1,909 | -20% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +27 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.