Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Project context loading, isolation, and persistent state management across CCPM sessions.
.claude/skills/a5c-ai-context-management/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 28 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -77% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -76% | 0% |
Project context loading, isolation, and persistent state management across CCPM sessions.
.claude/
prds/ # PRD documents
epics/ # Epic and task documents
agents/ # Agent definitions
context/ # Project-wide context
commands/ # Command definitions
rules/ # Reference rulesEach parallel agent receives only the context relevant to its stream, preventing context pollution and keeping agents focused.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 10,987 | 4,914 | -55% | 1 | 1 | 0% | 1,945 | 1,090 | -44% | 0 | 0 | — |
case-02 | fail→pass | 10,890 | 8,110 | -26% | 1 | 1 | 0% | 1,768 | 1,821 | +3% | 0 | 0 | — |
case-03 | fail→fail | 3,504 | 3,363 | -4% | 1 | 1 | 0% | 478 | 685 | +43% | 0 | 0 | — |
case-04 | fail→pass | 8,217 | 2,358 | -71% | 1 | 1 | 0% | 1,186 | 718 | -39% | 0 | 0 | — |
case-05 | fail→fail | 7,995 | 3,630 | -55% | 1 | 1 | 0% | 1,401 | 907 | -35% | 0 | 0 | — |
case-06 | fail→fail | 5,824 | 1,798 | -69% | 1 | 1 | 0% | 854 | 553 | -35% | 0 | 0 | — |
case-07 | fail→fail | 14,115 | 13,545 | -4% | 1 | 1 | 0% | 2,291 | 2,327 | +2% | 0 | 0 | — |
case-08 | fail→pass | 16,125 | 1,953 | -88% | 1 | 1 | 0% | 2,697 | 628 | -77% | 0 | 0 | — |
case-09 | fail→pass | 15,146 | 1,727 | -89% | 1 | 1 | 0% | 2,270 | 553 | -76% | 0 | 0 | — |
case-10 | fail→pass | 14,701 | 2,268 | -85% | 1 | 1 | 0% | 2,187 | 594 | -73% | 0 | 0 | — |
case-11 | fail→pass | 10,594 | 1,870 | -82% | 1 | 1 | 0% | 1,778 | 557 | -69% | 0 | 0 | — |
case-12 | fail→pass | 10,328 | 2,129 | -79% | 1 | 1 | 0% | 1,935 | 601 | -69% | 0 | 0 | — |
case-13 | fail→pass | 13,135 | 2,855 | -78% | 1 | 1 | 0% | 1,859 | 727 | -61% | 0 | 0 | — |
case-14 | fail→pass | 15,487 | 9,054 | -42% | 1 | 1 | 0% | 2,535 | 1,974 | -22% | 0 | 0 | — |
case-15 | fail→pass | 11,357 | 6,273 | -45% | 1 | 1 | 0% | 1,867 | 1,332 | -29% | 0 | 0 | — |
case-16 | fail→pass | 12,588 | 7,208 | -43% | 1 | 1 | 0% | 2,116 | 1,529 | -28% | 0 | 0 | — |
case-17 | fail→fail | 14,970 | 10,622 | -29% | 1 | 1 | 0% | 2,389 | 2,051 | -14% | 0 | 0 | — |
case-18 | fail→pass | 12,162 | 4,081 | -66% | 1 | 1 | 0% | 2,199 | 1,037 | -53% | 0 | 0 | — |
case-19 | fail→fail | 8,252 | 2,783 | -66% | 1 | 1 | 0% | 1,168 | 668 | -43% | 0 | 0 | — |
case-20 | fail→fail | 3,217 | 4,763 | +48% | 1 | 1 | 0% | 517 | 938 | +81% | 0 | 0 | — |
case-21 | fail→fail | 2,602 | 2,698 | +4% | 1 | 1 | 0% | 443 | 719 | +62% | 0 | 0 | — |
case-22 | fail→fail | 11,946 | 9,855 | -18% | 1 | 1 | 0% | 2,022 | 2,034 | +1% | 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. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.