Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Automatically evaluates OmA sessions to extract reusable patterns (error resolutions, workarounds, conventions) and save them to `.omg/rules/learned/`.
.claude/skills/bilal140202-learn-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -11% | 0% |
Automatically evaluates OmA sessions to extract reusable patterns (error resolutions, workarounds, conventions) and save them to .omg/rules/learned/.
SessionEnd hook for session evaluation..omg/rules/learned/.This skill runs as a SessionEnd hook at the end of each session:
.omg/rules/learned/.When /oma:learn is run, the agent will:
The learn skill focuses on reusable patterns rather than simple chat history:
Edit .omg/rules/learn.json to customize:
json{ "min_session_length": 10, "extraction_threshold": "medium", "auto_approve": false, "learned_skills_path": ".omg/rules/learned/", "patterns_to_detect": [ "error_resolution", "user_corrections", "workarounds", "debugging_techniques", "project_specific" ], "ignore_patterns": [ "simple_typos", "one_time_fixes", "external_api_issues" ] }
/oma:memory - Project-level knowledge./oma:rules - Context-aware rule application./oma:learn - Manual pattern extraction command.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 23,355 | 6,755 | -71% | 1 | 1 | 0% | 2,450 | 1,864 | -24% | 0 | 0 | — |
case-02 | fail→pass | 6,356 | 9,506 | +50% | 1 | 1 | 0% | 1,134 | 1,953 | +72% | 0 | 0 | — |
case-03 | fail→pass | 6,961 | 5,061 | -27% | 1 | 1 | 0% | 1,307 | 1,533 | +17% | 0 | 0 | — |
case-04 | pass→pass | 11,139 | 5,557 | -50% | 1 | 1 | 0% | 1,889 | 1,652 | -13% | 0 | 0 | — |
case-17 | pass→pass | 9,288 | 1,872 | -80% | 1 | 1 | 0% | 1,662 | 939 | -44% | 0 | 0 | — |
case-05 | pass→pass | 11,894 | 2,953 | -75% | 1 | 1 | 0% | 1,898 | 1,048 | -45% | 0 | 0 | — |
case-06 | pass→pass | 13,112 | 4,799 | -63% | 1 | 1 | 0% | 2,183 | 1,430 | -34% | 0 | 0 | — |
case-07 | fail→pass | 8,706 | 3,801 | -56% | 1 | 1 | 0% | 1,384 | 1,203 | -13% | 0 | 0 | — |
case-08 | fail→pass | 6,666 | 2,241 | -66% | 1 | 1 | 0% | 1,018 | 907 | -11% | 0 | 0 | — |
case-09 | fail→fail | 8,970 | 4,846 | -46% | 1 | 1 | 0% | 1,446 | 1,379 | -5% | 0 | 0 | — |
case-10 | pass→pass | 4,403 | 3,436 | -22% | 1 | 1 | 0% | 784 | 1,195 | +52% | 0 | 0 | — |
case-11 | pass→pass | 2,259 | 2,597 | +15% | 1 | 1 | 0% | 407 | 938 | +130% | 0 | 0 | — |
case-12 | pass→pass | 10,194 | 2,555 | -75% | 1 | 1 | 0% | 1,662 | 1,012 | -39% | 0 | 0 | — |
case-13 | pass→pass | 11,303 | 3,845 | -66% | 1 | 1 | 0% | 1,727 | 1,167 | -32% | 0 | 0 | — |
case-14 | fail→pass | 9,334 | 3,235 | -65% | 1 | 1 | 0% | 1,555 | 1,046 | -33% | 0 | 0 | — |
case-15 | fail→pass | 12,460 | 3,083 | -75% | 1 | 1 | 0% | 1,979 | 1,194 | -40% | 0 | 0 | — |
case-16 | pass→pass | 12,689 | 3,542 | -72% | 1 | 1 | 0% | 2,123 | 1,173 | -45% | 0 | 0 | — |
case-18 | pass→pass | 11,121 | 6,397 | -42% | 1 | 1 | 0% | 2,136 | 1,683 | -21% | 0 | 0 | — |
case-19 | fail→pass | 6,853 | 1,815 | -74% | 1 | 1 | 0% | 1,135 | 847 | -25% | 0 | 0 | — |
case-20 | fail→pass | 9,301 | 3,049 | -67% | 1 | 1 | 0% | 1,619 | 1,051 | -35% | 0 | 0 | — |
case-21 | pass→pass | 6,898 | 2,082 | -70% | 1 | 1 | 0% | 1,102 | 906 | -18% | 0 | 0 | — |
case-22 | fail→fail | 9,699 | 1,816 | -81% | 1 | 1 | 0% | 1,669 | 853 | -49% | 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 +41 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.