Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Consolidate brain memory and mine user sessions since the last consolidation checkpoint (sleep-cycle style). Use to reduce noisy prompt injection while preserving durable high-value memories.
.claude/skills/mikeyobrien-memory-consolidate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -6% | 0% |
Run a "brain sleep cycle":
Use the brain tool for brain changes and the vault tool for reference relocation. Never edit brain.jsonl directly.
Use these symbols while triaging entries:
> For memory retention, high importance means keep longer, not delete.
~/.rho/brain/brain.jsonl)true)last_consolidation) — last_consolidation | <ISO timestamp> | <duration>1) — only used when no checkpoint exists~/.pi/agent/sessions/)10)high) — high | mediummemory_consolidate.last_consolidated_at)List memory by type and count totals.
You MUST report counts for:
Determine the lower bound timestamp:
since is explicit timestamp/duration, use it.since=last_consolidation, read checkpoint from checkpoint_key.days_fallback.Constraints:
Extract durable learnings/preferences from user messages in matched sessions.
Confidence policy:
mediumConstraints:
max_new_entries.source when available (session:<id>).Run brain action=decay.
Constraints:
Identify duplicates/superseded/stale entries and merge candidates.
Apply the importance lens:
Apply the 30-day test to every entry: "Would this change a decision I make 30 days from now?" If no, it's noise — remove it.
Auto-remove categories (these should never have been stored):
Constraints:
"Reference-heavy" means useful knowledge that does not need to be injected every turn and can be searched ad hoc.
Typical candidates:
Constraints:
## Connections with [[wikilinks]].At end of successful consolidation, set/update checkpoint timestamp (now, UTC ISO-8601).
Constraints:
You MUST report:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,065 | 6,584 | -7% | 1 | 1 | 0% | 1,070 | 1,531 | +43% | 0 | 0 | — |
case-02 | fail→fail | 10,372 | 4,851 | -53% | 1 | 1 | 0% | 1,744 | 1,550 | -11% | 0 | 0 | — |
case-03 | fail→fail | 7,425 | 5,451 | -27% | 1 | 1 | 0% | 1,222 | 1,582 | +29% | 0 | 0 | — |
case-04 | fail→pass | 17,385 | 8,458 | -51% | 1 | 1 | 0% | 2,843 | 2,662 | -6% | 0 | 0 | — |
case-05 | pass→pass | 4,989 | 3,168 | -37% | 1 | 1 | 0% | 757 | 1,778 | +135% | 0 | 0 | — |
case-06 | fail→pass | 6,794 | 2,339 | -66% | 1 | 1 | 0% | 1,042 | 1,597 | +53% | 0 | 0 | — |
case-07 | pass→pass | 6,015 | 3,444 | -43% | 1 | 1 | 0% | 950 | 1,808 | +90% | 0 | 0 | — |
case-08 | pass→pass | 6,731 | 2,619 | -61% | 1 | 1 | 0% | 956 | 1,699 | +78% | 0 | 0 | — |
case-09 | pass→pass | 7,649 | 3,205 | -58% | 1 | 1 | 0% | 1,205 | 1,809 | +50% | 0 | 0 | — |
case-10 | pass→pass | 4,350 | 2,712 | -38% | 1 | 1 | 0% | 686 | 1,687 | +146% | 0 | 0 | — |
case-11 | fail→pass | 14,049 | 3,990 | -72% | 1 | 1 | 0% | 2,444 | 1,890 | -23% | 0 | 0 | — |
case-12 | pass→pass | 6,928 | 1,950 | -72% | 1 | 1 | 0% | 1,136 | 1,598 | +41% | 0 | 0 | — |
case-13 | fail→pass | 6,820 | 1,907 | -72% | 1 | 1 | 0% | 1,235 | 1,574 | +27% | 0 | 0 | — |
case-14 | fail→pass | 10,060 | 1,766 | -82% | 1 | 1 | 0% | 1,656 | 1,561 | -6% | 0 | 0 | — |
case-15 | fail→pass | 8,561 | 1,489 | -83% | 1 | 1 | 0% | 1,435 | 1,433 | -0% | 0 | 0 | — |
case-16 | pass→pass | 5,115 | 3,108 | -39% | 1 | 1 | 0% | 704 | 1,749 | +148% | 0 | 0 | — |
case-17 | fail→pass | 9,362 | 2,984 | -68% | 1 | 1 | 0% | 1,528 | 1,716 | +12% | 0 | 0 | — |
case-18 | pass→pass | 8,534 | 3,948 | -54% | 1 | 1 | 0% | 1,360 | 1,940 | +43% | 0 | 0 | — |
case-19 | fail→pass | 4,205 | 1,989 | -53% | 1 | 1 | 0% | 715 | 1,525 | +113% | 0 | 0 | — |
case-20 | fail→fail | 2,292 | 3,749 | +64% | 1 | 1 | 0% | 347 | 1,452 | +318% | 0 | 0 | — |
case-21 | pass→fail | 11,708 | 16,146 | +38% | 1 | 1 | 0% | 2,137 | 1,489 | -30% | 0 | 0 | — |
case-22 | pass→pass | 7,997 | 12,127 | +52% | 1 | 1 | 0% | 1,357 | 2,814 | +107% | 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 17 counted toward the lift figure. The other 5 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 +32 percentage points is the difference between those two pass rates over the 17 comparable cases. 1 case got worse with the skill loaded, and it is 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.