Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Explain how claude-mem captures observations, when memory injection kicks in, and where data lives. Use when the user asks "how does claude-mem work?" or "what is this thing doing?".
.claude/skills/thedotmack-how-it-works/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -75% | 0% |
Every Read, Edit, and Bash that Claude makes turns into a compressed observation. Observations get summarized at session end. Relevant ones get auto-injected into future prompts so the next session starts with context from the last one — no re-explaining the codebase, no re-discovering decisions.
Memory injection starts on your second session in a project.
The first session in a fresh project seeds memory; subsequent sessions receive auto-injected context for relevant past work. Run /learn-codebase if you want to front-load the entire repo into memory in a single pass (~5 minutes, optional).
Everything stays in ~/.claude-mem on this machine.
Nothing leaves your machine except calls to whichever AI provider you configured for compression (Claude / OpenRouter / Gemini). The SQLite database, vector index, logs, and settings all live under that directory and are removed cleanly on npx claude-mem uninstall.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 14,246 | 6,142 | -57% | 1 | 1 | 0% | 2,443 | 1,362 | -44% | 0 | 0 | — |
case-02 | fail→pass | 9,893 | 4,066 | -59% | 1 | 1 | 0% | 1,755 | 974 | -45% | 0 | 0 | — |
case-03 | pass→pass | 9,318 | 6,666 | -28% | 1 | 1 | 0% | 1,794 | 1,360 | -24% | 0 | 0 | — |
case-04 | fail→pass | 7,654 | 5,527 | -28% | 1 | 1 | 0% | 1,295 | 1,212 | -6% | 0 | 0 | — |
case-05 | fail→pass | 8,156 | 5,471 | -33% | 1 | 1 | 0% | 1,393 | 1,212 | -13% | 0 | 0 | — |
case-06 | pass→pass | 7,496 | 3,342 | -55% | 1 | 1 | 0% | 1,281 | 804 | -37% | 0 | 0 | — |
case-07 | pass→pass | 13,484 | 3,436 | -75% | 1 | 1 | 0% | 2,157 | 603 | -72% | 0 | 0 | — |
case-08 | fail→pass | 9,107 | 1,449 | -84% | 1 | 1 | 0% | 1,902 | 472 | -75% | 0 | 0 | — |
case-09 | fail→pass | 15,364 | 2,568 | -83% | 1 | 1 | 0% | 1,109 | 587 | -47% | 0 | 0 | — |
case-10 | pass→pass | 10,436 | 2,155 | -79% | 1 | 1 | 0% | 2,187 | 610 | -72% | 0 | 0 | — |
case-11 | pass→pass | 10,351 | 5,083 | -51% | 1 | 1 | 0% | 2,052 | 1,227 | -40% | 0 | 0 | — |
case-12 | fail→pass | 6,194 | 1,563 | -75% | 1 | 1 | 0% | 1,224 | 461 | -62% | 0 | 0 | — |
case-13 | pass→pass | 9,261 | 1,901 | -79% | 1 | 1 | 0% | 1,627 | 503 | -69% | 0 | 0 | — |
case-14 | pass→pass | 7,912 | 2,733 | -65% | 1 | 1 | 0% | 1,324 | 590 | -55% | 0 | 0 | — |
case-15 | pass→pass | 10,183 | 6,438 | -37% | 1 | 1 | 0% | 1,822 | 1,383 | -24% | 0 | 0 | — |
case-16 | fail→pass | 10,952 | 7,492 | -32% | 1 | 1 | 0% | 2,112 | 1,532 | -27% | 0 | 0 | — |
case-17 | fail→pass | 11,745 | 2,426 | -79% | 1 | 1 | 0% | 1,901 | 677 | -64% | 0 | 0 | — |
case-18 | fail→pass | 13,247 | 6,762 | -49% | 1 | 1 | 0% | 1,955 | 403 | -79% | 0 | 0 | — |
case-19 | fail→pass | 8,318 | 1,355 | -84% | 1 | 1 | 0% | 1,643 | 423 | -74% | 0 | 0 | — |
case-20 | fail→pass | 10,801 | 2,151 | -80% | 1 | 1 | 0% | 1,849 | 598 | -68% | 0 | 0 | — |
case-21 | pass→pass | 9,600 | 3,829 | -60% | 1 | 1 | 0% | 1,594 | 887 | -44% | 0 | 0 | — |
case-22 | fail→pass | 11,454 | 1,906 | -83% | 1 | 1 | 0% | 2,299 | 505 | -78% | 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 +59 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.