Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a deliberate shortcut is taken during a fix or feature, when investigating a bug that might already be logged, or when closing out an entry after a proper fix lands.
.claude/skills/ccplugins-logging-tradeoffs/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 133% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 311% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -2% | 0% |
A shortcut without a record is a landmine for whoever touches the code next — often future-you. This skill defines one convention, KNOWN_ISSUES.md, kept at the root of each project, so deliberate tradeoffs are findable instead of rediscovered the hard way.
Core principle: if you knowingly ship something imperfect, write down what, why, and when to revisit it — in the same motion as writing the code, not as a follow-up you'll forget.
KNOWN_ISSUES.md already exists at the project root. If an equivalent fileexists under a different name (TODO.md, ISSUES.md, BACKLOG.md), don't create a duplicate — ask the user whether to adopt the existing file for this convention instead.
graphify-out/graph.json, use the graphify-recurring-bugs skill's dedupcheck first — a new bug may be the same root cause as an existing open entry, not a new one.
explicitly request — surface the proposed content, get a yes, then write it.
Append one block per issue, most recent first:
## <date> — <short title>
- What changed / what shortcut was taken:
- Ceiling (when this breaks):
- Upgrade trigger (what to do when it's hit):
- Status: open | resolved (<date>, <how>)file:line reference where relevant.testable fact ("more than 10k rows", "concurrent writers", "amounts with fractional cents").
it properly later."
resolved with a date and one line on how, when addressed. Don't deleteresolved entries — they're evidence the log is actually maintained, not decoration.
If reviewer or coder touches a file referenced by an open KNOWN_ISSUES.md entry, say explicitly whether the current change resolves that entry, is unrelated to it, or makes it worse. Silence here is how the same shortcut gets shipped around three times.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,215 | 14,304 | +28% | 1 | 1 | 0% | 1,672 | 2,540 | +52% | 0 | 0 | — |
case-02 | fail→pass | 7,218 | 15,401 | +113% | 1 | 1 | 0% | 1,118 | 2,601 | +133% | 0 | 0 | — |
case-03 | fail→fail | 9,994 | 6,236 | -38% | 1 | 1 | 0% | 1,357 | 1,036 | -24% | 0 | 0 | — |
case-04 | fail→pass | 3,001 | 6,216 | +107% | 1 | 1 | 0% | 367 | 1,508 | +311% | 0 | 0 | — |
case-05 | fail→pass | 11,662 | 7,158 | -39% | 1 | 1 | 0% | 1,735 | 1,561 | -10% | 0 | 0 | — |
case-06 | fail→fail | 4,460 | 13,473 | +202% | 1 | 1 | 0% | 441 | 1,907 | +332% | 0 | 0 | — |
case-07 | fail→pass | 11,298 | 7,044 | -38% | 1 | 1 | 0% | 1,580 | 1,680 | +6% | 0 | 0 | — |
case-08 | pass→fail | 7,719 | 5,391 | -30% | 1 | 1 | 0% | 1,075 | 1,098 | +2% | 0 | 0 | — |
case-09 | pass→pass | 11,864 | 9,494 | -20% | 1 | 1 | 0% | 1,704 | 1,355 | -20% | 0 | 0 | — |
case-10 | pass→pass | 8,655 | 4,846 | -44% | 1 | 1 | 0% | 1,350 | 1,337 | -1% | 0 | 0 | — |
case-11 | pass→pass | 12,627 | 6,846 | -46% | 1 | 1 | 0% | 1,869 | 1,633 | -13% | 0 | 0 | — |
case-12 | pass→pass | 9,110 | 3,871 | -58% | 1 | 1 | 0% | 1,278 | 1,081 | -15% | 0 | 0 | — |
case-13 | pass→pass | 19,626 | 6,192 | -68% | 1 | 1 | 0% | 1,576 | 1,404 | -11% | 0 | 0 | — |
case-14 | pass→pass | 13,093 | 3,941 | -70% | 1 | 1 | 0% | 1,817 | 1,151 | -37% | 0 | 0 | — |
case-15 | fail→pass | 12,062 | 6,585 | -45% | 1 | 1 | 0% | 1,612 | 1,585 | -2% | 0 | 0 | — |
case-16 | pass→pass | 5,324 | 5,862 | +10% | 1 | 1 | 0% | 807 | 1,405 | +74% | 0 | 0 | — |
case-17 | pass→pass | 5,537 | 16,327 | +195% | 1 | 1 | 0% | 779 | 1,725 | +121% | 0 | 0 | — |
case-18 | pass→pass | 11,159 | 9,724 | -13% | 1 | 1 | 0% | 1,742 | 2,216 | +27% | 0 | 0 | — |
case-19 | pass→pass | 12,934 | 12,443 | -4% | 1 | 1 | 0% | 2,228 | 2,967 | +33% | 0 | 0 | — |
case-20 | fail→fail | 10,129 | 4,702 | -54% | 1 | 1 | 0% | 1,578 | 1,283 | -19% | 0 | 0 | — |
case-21 | pass→pass | 8,064 | 5,785 | -28% | 1 | 1 | 0% | 1,371 | 1,409 | +3% | 0 | 0 | — |
case-22 | pass→pass | 8,342 | 3,650 | -56% | 1 | 1 | 0% | 1,312 | 1,170 | -11% | 0 | 0 | — |
case-23 | fail→pass | 13,655 | 6,480 | -53% | 1 | 1 | 0% | 1,866 | 1,694 | -9% | 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. 23 cases were attempted, and 22 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 +22 percentage points is the difference between those two pass rates over the 22 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.