Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Memory is the cornerstone of intelligent agents. Without it, every interaction starts from zero. This skill covers the architecture of agent memory: short-term (context window), long-term (vector stores), and the cognitive architectures that organize them. Key insight: Memory isn't just storage - it's retrieval. A million stored facts mean nothing if you can't find the right one. Chunking, embedding, and retrieval strategies determine whether your agent remembers or forgets. The field is fragm
.claude/skills/dokhacgiakhoa-agent-memory-systems/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flashlowest | 92% | 265 |
| gemini-3.1-pro-preview | 100% | 1 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 0% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 2% | 0% |
| case-08 | ✓→✓ | ＝ Same ✓ | 7% | 0% |
You are a cognitive architect who understands that memory makes agents intelligent. You've built memory systems for agents handling millions of interactions. You know that the hard part isn't storing - it's retrieving the right memory at the right time.
Your core insight: Memory failures look like intelligence failures. When an agent "forgets" or gives inconsistent answers, it's almost always a retrieval problem, not a storage problem. You obsess over chunking strategies, embedding quality, and
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 13,819 | 17,382 | +26% | 1 | 1 | 0% | 2,010 | 2,938 | +46% | 0 | 0 | — |
case-02 | pass→pass | 20,808 | 19,084 | -8% | 1 | 1 | 0% | 3,250 | 3,250 | 0% | 0 | 0 | — |
case-03 | pass→pass | 20,065 | 17,814 | -11% | 1 | 1 | 0% | 2,939 | 3,009 | +2% | 0 | 0 | — |
case-08 | pass→pass | 15,948 | 13,466 | -16% | 1 | 1 | 0% | 2,410 | 2,575 | +7% | 0 | 0 | — |
case-04 | pass→pass | 10,724 | 8,086 | -25% | 1 | 1 | 0% | 1,756 | 1,721 | -2% | 0 | 0 | — |
case-05 | pass→pass | 7,286 | 8,315 | +14% | 1 | 1 | 0% | 1,053 | 1,805 | +71% | 0 | 0 | — |
case-06 | pass→pass | 5,816 | 7,563 | +30% | 1 | 1 | 0% | 951 | 1,580 | +66% | 0 | 0 | — |
case-07 | pass→pass | 16,206 | 20,385 | +26% | 1 | 1 | 0% | 2,607 | 3,698 | +42% | 0 | 0 | — |
case-09 | pass→pass | 9,400 | 13,028 | +39% | 1 | 1 | 0% | 1,648 | 2,761 | +68% | 0 | 0 | — |
case-10 | pass→pass | 7,987 | 14,532 | +82% | 1 | 1 | 0% | 1,615 | 2,570 | +59% | 0 | 0 | — |
case-11 | pass→pass | 15,560 | 19,298 | +24% | 1 | 1 | 0% | 2,647 | 3,596 | +36% | 0 | 0 | — |
case-12 | pass→pass | 15,325 | 13,251 | -14% | 1 | 1 | 0% | 2,091 | 2,594 | +24% | 0 | 0 | — |
case-13 | pass→pass | 10,721 | 9,510 | -11% | 1 | 1 | 0% | 1,738 | 1,915 | +10% | 0 | 0 | — |
case-14 | fail→pass | 12,562 | 16,386 | +30% | 1 | 1 | 0% | 2,301 | 3,125 | +36% | 0 | 0 | — |
case-15 | pass→pass | 9,399 | 9,914 | +5% | 1 | 1 | 0% | 1,474 | 1,957 | +33% | 0 | 0 | — |
case-16 | pass→pass | 12,436 | 16,311 | +31% | 1 | 1 | 0% | 2,114 | 3,126 | +48% | 0 | 0 | — |
case-17 | pass→pass | 15,729 | 17,258 | +10% | 1 | 1 | 0% | 2,641 | 3,452 | +31% | 0 | 0 | — |
case-18 | pass→pass | 9,793 | 12,176 | +24% | 1 | 1 | 0% | 1,537 | 2,343 | +52% | 0 | 0 | — |
case-19 | pass→pass | 15,882 | 17,273 | +9% | 1 | 1 | 0% | 2,714 | 3,497 | +29% | 0 | 0 | — |
case-20 | pass→pass | 13,424 | 10,151 | -24% | 1 | 1 | 0% | 2,164 | 2,346 | +8% | 0 | 0 | — |
case-21 | pass→pass | 8,412 | 9,275 | +10% | 1 | 1 | 0% | 1,660 | 2,247 | +35% | 0 | 0 | — |
case-22 | pass→pass | 13,943 | 15,145 | +9% | 1 | 1 | 0% | 2,609 | 3,341 | +28% | 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 +9 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.