Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Apply when you catch yourself writing the same instruction a second time, or notice a recurring correction. Encode the rule as a lint, metadata flag, runtime check, or script instead of more text.
.claude/skills/sediman-agent-principle-encode-lessons-in-structure/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-22 | ✓→✗ | ▼ Worse | -90% | 0% |
| case-12 | ✓→✓ | ＝ Same ✓ | -20% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | -25% | 0% |
Encode recurring fixes in mechanisms (tools, code, metadata, automation) instead of textual instructions. Every error, human correction, and unexpected outcome is a learning signal. Capture it, route it, and close the loop.
Why: Textual instructions are easy to miss. They require the reader to notice, remember, and comply. Structural mechanisms (lint rules, metadata flags, runtime checks, automation scripts) enforce the rule without cooperation.
Pattern: When you catch yourself writing the same instruction a second time:
Corollary: Don't paper over symptoms. If the fix is structural, ONLY use the structural fix. The instruction IS the symptom.
Feedback loop:
Anti-patterns:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→fail | 11,935 | 7,916 | -34% | 1 | 1 | 0% | 2,020 | 1,505 | -25% | 0 | 0 | — |
case-11 | fail→fail | 10,414 | 6,659 | -36% | 1 | 1 | 0% | 1,486 | 1,382 | -7% | 0 | 0 | — |
case-12 | pass→pass | 11,652 | 6,469 | -44% | 1 | 1 | 0% | 1,614 | 1,284 | -20% | 0 | 0 | — |
case-01 | fail→pass | 9,130 | 5,923 | -35% | 1 | 1 | 0% | 1,397 | 1,271 | -9% | 0 | 0 | — |
case-02 | pass→pass | 10,285 | 5,983 | -42% | 1 | 1 | 0% | 1,635 | 1,226 | -25% | 0 | 0 | — |
case-03 | pass→pass | 13,345 | 9,082 | -32% | 1 | 1 | 0% | 2,132 | 1,740 | -18% | 0 | 0 | — |
case-04 | pass→pass | 13,252 | 7,362 | -44% | 1 | 1 | 0% | 1,914 | 1,410 | -26% | 0 | 0 | — |
case-06 | pass→pass | 9,724 | 5,190 | -47% | 1 | 1 | 0% | 1,494 | 1,185 | -21% | 0 | 0 | — |
case-07 | pass→pass | 10,837 | 5,904 | -46% | 1 | 1 | 0% | 1,684 | 1,367 | -19% | 0 | 0 | — |
case-08 | pass→pass | 16,050 | 13,377 | -17% | 1 | 1 | 0% | 2,365 | 2,349 | -1% | 0 | 0 | — |
case-09 | pass→pass | 16,206 | 11,148 | -31% | 1 | 1 | 0% | 2,308 | 2,023 | -12% | 0 | 0 | — |
case-10 | fail→fail | 11,047 | 7,842 | -29% | 1 | 1 | 0% | 1,653 | 1,596 | -3% | 0 | 0 | — |
case-13 | pass→pass | 13,804 | 8,647 | -37% | 1 | 1 | 0% | 2,058 | 1,630 | -21% | 0 | 0 | — |
case-14 | pass→pass | 11,939 | 9,836 | -18% | 1 | 1 | 0% | 1,791 | 1,724 | -4% | 0 | 0 | — |
case-15 | pass→pass | 10,956 | 6,271 | -43% | 1 | 1 | 0% | 1,553 | 1,231 | -21% | 0 | 0 | — |
case-16 | pass→pass | 17,581 | 14,399 | -18% | 1 | 1 | 0% | 2,442 | 2,334 | -4% | 0 | 0 | — |
case-17 | pass→pass | 9,999 | 8,127 | -19% | 1 | 1 | 0% | 1,448 | 1,482 | +2% | 0 | 0 | — |
case-18 | fail→fail | 11,876 | 6,936 | -42% | 1 | 1 | 0% | 1,679 | 1,303 | -22% | 0 | 0 | — |
case-19 | pass→pass | 12,949 | 6,665 | -49% | 1 | 1 | 0% | 1,855 | 1,289 | -31% | 0 | 0 | — |
case-20 | fail→pass | 8,474 | 6,617 | -22% | 1 | 1 | 0% | 1,304 | 1,356 | +4% | 0 | 0 | — |
case-21 | pass→pass | 13,923 | 6,532 | -53% | 1 | 1 | 0% | 2,149 | 1,317 | -39% | 0 | 0 | — |
case-22 | pass→fail | 33,828 | 4,232 | -87% | 1 | 1 | 0% | 6,182 | 615 | -90% | 0 | 0 | — |
case-23 | pass→pass | 8,183 | 7,519 | -8% | 1 | 1 | 0% | 1,401 | 1,633 | +17% | 0 | 0 | — |
case-24 | pass→pass | 6,525 | 6,911 | +6% | 1 | 1 | 0% | 1,116 | 1,464 | +31% | 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. 24 cases were attempted, and 23 counted toward the lift figure. The other 1 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 +4 percentage points is the difference between those two pass rates over the 23 comparable cases. 2 cases got worse with the skill loaded, and they are 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.