Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Explicitly save important knowledge to auto-memory with timestamp and context. Use when a discovery is too important to rely on auto-capture.
.claude/skills/alirezarezvani-remember/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 90% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 52% | 0% |
| case-06 | ✓→✓ | = Same ✓ | 33% | 0% |
| case-20 | ✓→✓ | = Same ✓ | 18% | 0% |
| case-01 | ✗→✗ | = Same ✗ | 11% | 0% |
Writes an explicit entry to auto-memory when something is important enough that you don't want to rely on Claude noticing it automatically.
/si:remember <what to remember>
/si:remember "This project's CI requires Node 20 LTS — v22 breaks the build"
/si:remember "The /api/auth endpoint uses a custom JWT library, not passport"
/si:remember "Reza prefers explicit error handling over try-catch-all patterns"| Situation | Example | |-----------|---------| | Hard-won debugging insight | "CORS errors on /api/upload are caused by the CDN, not the backend" | | Project convention not in CLAUDE.md | "We use barrel exports in src/components/" | | Tool-specific gotcha | "Jest needs --forceExit flag or it hangs on DB tests" | | Architecture decision | "We chose Drizzle over Prisma for type-safe SQL" | | Preference you want Claude to learn | "Don't add comments explaining obvious code" |
Extract from the user's input:
bashMEMORY_DIR="$HOME/.claude/projects/$(pwd | sed 's|/|%2F|g; s|%2F|/|; s|^/||')/memory" grep -ni "<keywords>" "$MEMORY_DIR/MEMORY.md" 2>/dev/null
If a similar entry exists:
Append to the end of MEMORY.md:
markdown- {{concise fact or pattern}}
Keep entries concise — one line when possible. Auto-memory entries don't need timestamps, IDs, or metadata. They're notes, not database records.
If MEMORY.md is over 180 lines, warn the user:
⚠️ MEMORY.md is at {{n}}/200 lines. Consider running /si:memory-review to free space.If the knowledge sounds like a rule (imperative, always/never, convention):
💡 This sounds like it could be a CLAUDE.md rule rather than a memory entry.
Rules are enforced with higher priority. Want to /si:promote it instead?✅ Saved to auto-memory
"{{entry}}"
MEMORY.md: {{n}}/200 lines
Claude will see this at the start of every session in this project./si:promote to write directly to CLAUDE.md~/.claude/CLAUDE.md for global rulespnpm build, tests with pnpm test:e2e"| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 6,855 | 7,170 | +5% | 1 | 1 | 0% | 1,112 | 1,230 | +11% | 0 | 0 | — |
case-02 | fail→fail | 4,031 | 7,573 | +88% | 1 | 1 | 0% | 607 | 1,262 | +108% | 0 | 0 | — |
case-03 | fail→fail | 8,094 | 6,418 | -21% | 1 | 1 | 0% | 613 | 1,098 | +79% | 0 | 0 | — |
case-04 | pass→pass | 5,061 | 3,611 | -29% | 1 | 1 | 0% | 942 | 1,435 | +52% | 0 | 0 | — |
case-05 | fail→fail | 2,707 | 7,575 | +180% | 1 | 1 | 0% | 494 | 1,271 | +157% | 0 | 0 | — |
case-06 | pass→pass | 10,972 | 9,481 | -14% | 1 | 1 | 0% | 1,400 | 1,861 | +33% | 0 | 0 | — |
case-07 | fail→fail | 6,496 | 5,265 | -19% | 1 | 1 | 0% | 344 | 1,163 | +238% | 0 | 0 | — |
case-08 | fail→fail | 11,048 | 7,532 | -32% | 1 | 1 | 0% | 1,333 | 1,248 | -6% | 0 | 0 | — |
case-09 | fail→fail | 4,610 | 8,507 | +85% | 1 | 1 | 0% | 622 | 1,457 | +134% | 0 | 0 | — |
case-10 | fail→fail | 1,882 | 8,145 | +333% | 1 | 1 | 0% | 296 | 1,207 | +308% | 0 | 0 | — |
case-11 | fail→fail | 2,028 | 13,118 | +547% | 1 | 1 | 0% | 305 | 1,223 | +301% | 0 | 0 | — |
case-12 | fail→fail | 7,834 | 7,012 | -10% | 1 | 1 | 0% | 861 | 1,241 | +44% | 0 | 0 | — |
case-13 | fail→pass | 5,066 | 3,267 | -36% | 1 | 1 | 0% | 735 | 1,398 | +90% | 0 | 0 | — |
case-14 | fail→fail | 3,184 | 8,001 | +151% | 1 | 1 | 0% | 532 | 1,301 | +145% | 0 | 0 | — |
case-15 | fail→fail | 1,709 | 8,180 | +379% | 1 | 1 | 0% | 222 | 1,360 | +513% | 0 | 0 | — |
case-16 | fail→fail | 9,258 | 6,761 | -27% | 1 | 1 | 0% | 1,071 | 1,245 | +16% | 0 | 0 | — |
case-17 | fail→fail | 2,550 | 6,934 | +172% | 1 | 1 | 0% | 310 | 1,188 | +283% | 0 | 0 | — |
case-18 | fail→fail | 3,818 | 7,599 | +99% | 1 | 1 | 0% | 640 | 1,413 | +121% | 0 | 0 | — |
case-19 | fail→fail | 6,022 | 17,124 | +184% | 1 | 1 | 0% | 925 | 1,303 | +41% | 0 | 0 | — |
case-20 | pass→pass | 8,189 | 4,519 | -45% | 1 | 1 | 0% | 1,580 | 1,857 | +18% | 0 | 0 | — |
case-21 | fail→fail | 5,136 | 6,965 | +36% | 1 | 1 | 0% | 829 | 1,230 | +48% | 0 | 0 | — |
case-22 | fail→fail | 2,697 | 6,565 | +143% | 1 | 1 | 0% | 410 | 1,269 | +210% | 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 4 counted toward the lift figure. The other 18 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. A headline lift is not published for this run.
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.