Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Mine high-signal transcript deltas, update `AGENTS.md`, and keep the incremental transcript index in sync.
.claude/skills/kunanonj-cursor-plugin-continual-agent-agents-memory-updater/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -60% | 0% |
Own the full memory update flow for continual learning.
Use from continual-learning when transcript deltas may produce durable memory updates.
AGENTS.md first. If it does not exist, create it with only:## Learned User Preferences## Learned Workspace Facts~/.cursor/projects/<workspace-slug>/agent-transcripts/ that are new or have newer mtimes than the index.AGENTS.md carefully:AGENTS.md changes, leave AGENTS.md unchanged but still refresh the index.No high-signal memory updates.## Learned User Preferences## Learned Workspace FactsAGENTS.md and .cursor/hooks/state/continual-learning-index.json when neededNo high-signal memory updates.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 2,114 | 4,038 | +91% | 1 | 1 | 0% | 258 | 739 | +186% | 0 | 0 | — |
case-02 | fail→fail | 3,722 | 4,915 | +32% | 1 | 1 | 0% | 255 | 637 | +150% | 0 | 0 | — |
case-08 | pass→pass | 2,214 | 1,650 | -25% | 1 | 1 | 0% | 486 | 733 | +51% | 0 | 0 | — |
case-13 | pass→pass | 10,560 | 2,052 | -81% | 1 | 1 | 0% | 2,058 | 826 | -60% | 0 | 0 | — |
case-03 | fail→fail | 5,355 | 8,475 | +58% | 1 | 1 | 0% | 749 | 980 | +31% | 0 | 0 | — |
case-04 | pass→pass | 9,268 | 3,118 | -66% | 1 | 1 | 0% | 1,788 | 1,050 | -41% | 0 | 0 | — |
case-05 | pass→pass | 6,480 | 3,499 | -46% | 1 | 1 | 0% | 1,426 | 1,190 | -17% | 0 | 0 | — |
case-06 | fail→pass | 8,169 | 4,141 | -49% | 1 | 1 | 0% | 1,547 | 1,336 | -14% | 0 | 0 | — |
case-07 | pass→pass | 8,613 | 1,955 | -77% | 1 | 1 | 0% | 1,645 | 753 | -54% | 0 | 0 | — |
case-09 | pass→fail | 5,370 | 2,440 | -55% | 1 | 1 | 0% | 1,094 | 843 | -23% | 0 | 0 | — |
case-10 | pass→pass | 7,420 | 1,407 | -81% | 1 | 1 | 0% | 1,441 | 678 | -53% | 0 | 0 | — |
case-11 | pass→pass | 6,392 | 5,092 | -20% | 1 | 1 | 0% | 1,332 | 1,595 | +20% | 0 | 0 | — |
case-12 | fail→pass | 7,904 | 1,859 | -76% | 1 | 1 | 0% | 1,557 | 806 | -48% | 0 | 0 | — |
case-19 | fail→fail | 1,755 | 2,843 | +62% | 1 | 1 | 0% | 246 | 585 | +138% | 0 | 0 | — |
case-14 | pass→pass | 6,999 | 2,857 | -59% | 1 | 1 | 0% | 1,440 | 1,018 | -29% | 0 | 0 | — |
case-15 | pass→pass | 6,417 | 3,592 | -44% | 1 | 1 | 0% | 1,296 | 1,136 | -12% | 0 | 0 | — |
case-16 | fail→pass | 9,788 | 1,750 | -82% | 1 | 1 | 0% | 1,849 | 686 | -63% | 0 | 0 | — |
case-17 | fail→pass | 8,292 | 1,601 | -81% | 1 | 1 | 0% | 1,572 | 701 | -55% | 0 | 0 | — |
case-18 | fail→pass | 8,230 | 1,404 | -83% | 1 | 1 | 0% | 1,652 | 654 | -60% | 0 | 0 | — |
case-20 | pass→fail | 4,548 | 7,444 | +64% | 1 | 1 | 0% | 856 | 1,885 | +120% | 0 | 0 | — |
case-21 | pass→fail | 7,574 | 10,292 | +36% | 1 | 1 | 0% | 1,683 | 1,698 | +1% | 0 | 0 | — |
case-22 | pass→fail | 6,215 | 7,114 | +14% | 1 | 1 | 0% | 1,176 | 1,815 | +54% | 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 18 counted toward the lift figure. The other 4 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 +5 percentage points is the difference between those two pass rates over the 18 comparable cases. 4 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.