Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Consolidate recent logs, sessions, and existing memory files into durable topic memories, normalize dates, prune stale entries, and keep MEMORY.md short enough for prompt use.
.claude/skills/kentunderage549-dream-memory/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 177% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -64% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 129% | 0% |
Use this skill when you want a deliberate memory-consolidation pass instead of storing more raw notes.
This bundle is intentionally portable. It borrows the workflow shape from Claude Code's dream system, but it does not depend on Claude Code internals.
MEMORY.md small and index-likeInspect a memory directory before running the consolidation prompt:
bashpython3 {baseDir}/scripts/dream_memory.py --memory-root /path/to/memory --transcripts-dir /path/to/transcripts
Then use the portable prompt in references/prompt-template.md.
MEMORY.md, topic files, and recent logs or transcripts.MEMORY.md as a concise index.MEMORY.md is an index, not a content dump.python3 {baseDir}/scripts/dream_memory.py ...| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-08 | pass→pass | 10,312 | 6,398 | -38% | 1 | 1 | 0% | 1,890 | 1,548 | -18% | 0 | 0 | — |
case-01 | fail→fail | 2,333 | 5,692 | +144% | 1 | 1 | 0% | 418 | 680 | +63% | 0 | 0 | — |
case-02 | fail→fail | 6,969 | 4,521 | -35% | 1 | 1 | 0% | 1,211 | 701 | -42% | 0 | 0 | — |
case-03 | fail→fail | 9,641 | 11,322 | +17% | 1 | 1 | 0% | 1,846 | 880 | -52% | 0 | 0 | — |
case-04 | fail→pass | 4,303 | 4,182 | -3% | 1 | 1 | 0% | 716 | 1,017 | +42% | 0 | 0 | — |
case-05 | pass→pass | 3,772 | 3,223 | -15% | 1 | 1 | 0% | 690 | 920 | +33% | 0 | 0 | — |
case-06 | fail→pass | 2,240 | 4,416 | +97% | 1 | 1 | 0% | 395 | 1,093 | +177% | 0 | 0 | — |
case-07 | pass→fail | 8,617 | 3,605 | -58% | 1 | 1 | 0% | 1,474 | 1,059 | -28% | 0 | 0 | — |
case-09 | pass→pass | 6,927 | 2,854 | -59% | 1 | 1 | 0% | 1,240 | 869 | -30% | 0 | 0 | — |
case-10 | fail→pass | 8,280 | 2,876 | -65% | 1 | 1 | 0% | 1,260 | 792 | -37% | 0 | 0 | — |
case-11 | pass→pass | 5,896 | 3,961 | -33% | 1 | 1 | 0% | 1,012 | 1,019 | +1% | 0 | 0 | — |
case-12 | fail→pass | 10,143 | 1,605 | -84% | 1 | 1 | 0% | 1,771 | 630 | -64% | 0 | 0 | — |
case-13 | fail→pass | 2,091 | 1,348 | -36% | 1 | 1 | 0% | 246 | 564 | +129% | 0 | 0 | — |
case-14 | pass→pass | 11,096 | 3,675 | -67% | 1 | 1 | 0% | 1,779 | 961 | -46% | 0 | 0 | — |
case-15 | fail→pass | 14,106 | 9,376 | -34% | 1 | 1 | 0% | 2,394 | 2,056 | -14% | 0 | 0 | — |
case-16 | pass→pass | 10,421 | 7,299 | -30% | 1 | 1 | 0% | 1,681 | 1,621 | -4% | 0 | 0 | — |
case-17 | pass→pass | 10,345 | 2,801 | -73% | 1 | 1 | 0% | 1,668 | 856 | -49% | 0 | 0 | — |
case-18 | fail→pass | 7,754 | 2,169 | -72% | 1 | 1 | 0% | 1,271 | 795 | -37% | 0 | 0 | — |
case-19 | fail→fail | 6,111 | 3,111 | -49% | 1 | 1 | 0% | 944 | 816 | -14% | 0 | 0 | — |
case-20 | fail→pass | 11,293 | 6,043 | -46% | 1 | 1 | 0% | 1,911 | 1,463 | -23% | 0 | 0 | — |
case-21 | pass→pass | 9,693 | 1,485 | -85% | 1 | 1 | 0% | 1,673 | 617 | -63% | 0 | 0 | — |
case-22 | fail→pass | 11,118 | 3,342 | -70% | 1 | 1 | 0% | 1,526 | 931 | -39% | 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 19 counted toward the lift figure. The other 3 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 +36 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 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.