Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →name: harness-fetch
.claude/skills/bilal140202-jasoncheroske-lab-os/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 11% | 0% |
name: harness-fetch description: >- Guides a user through fetching and setting up their chosen AI harness config from .ai/.<harness>/ into their local working environment. Use when the user says /harness-fetch, asks how to set up Cursor, Claude, or another harness for this lab, or wants to pull pre-built harness config without committing personal artifacts. created: 2026-04-05 updated: 2026-04-05 status: phase-2-planned
/harness-fetch)> Status: Planned (phase 2). This skill is a design stub. The workflow below describes the intended behavior when implemented. For now, follow the manual steps.
Purpose: Pull the pre-built AI harness configuration from .ai/.<harness>/ into the correct location in your local working environment—without committing personal artifacts to the repo.
.ai/.<harness>/ skills or rules| Harness | Config source | Where it should live locally | |---------|--------------|------------------------------| | Cursor | .ai/.cursor/ | .cursor/ at repo root (or workspace root) | | Claude Code | .ai/.claude/ | .claude/ at repo root | | (your harness) | .ai/.<harness>/ | wherever .<harness>/ is expected |
bash ls .ai/ Look for .<your-harness>/ alongside .cursor/ and other entries.
Option A — Copy (safest): bash cp -r .ai/.cursor/ .cursor/
Option B — Symlink (stays in sync automatically): bash ln -s .ai/.cursor .cursor On Windows: mklink /D .cursor .ai\.cursor or configure your harness to look inside .ai/.
.cursor/settings.json, .cursor/chat/, .claude/ at the repo root are gitignored..ai/ is committed—never the root-level harness config that contains your personal data.If .ai/.<your-harness>/ does not exist yet:
.ai/.<your-harness>/skills/ and .ai/.<your-harness>/rules/.ai/.cursor/ as a starting point.README.md inside .ai/.<your-harness>/ documenting what the config does.When fully implemented, this skill will:
.ai/.<harness>/ to the appropriate location automaticallyThis skill is harness-agnostic by design: written in plain instructions any agent or human can follow, not tied to any single tool's invocation format.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→fail | 11,585 | 1,482 | -87% | 1 | 1 | 0% | 1,921 | 1,058 | -45% | 0 | 0 | — |
case-03 | fail→pass | 8,824 | 3,846 | -56% | 1 | 1 | 0% | 1,665 | 1,569 | -6% | 0 | 0 | — |
case-04 | fail→fail | 7,472 | 4,919 | -34% | 1 | 1 | 0% | 1,500 | 1,851 | +23% | 0 | 0 | — |
case-16 | pass→pass | 9,194 | 5,427 | -41% | 1 | 1 | 0% | 1,659 | 1,740 | +5% | 0 | 0 | — |
case-09 | fail→fail | 6,714 | 2,141 | -68% | 1 | 1 | 0% | 1,611 | 1,235 | -23% | 0 | 0 | — |
case-01 | fail→fail | 7,603 | 4,133 | -46% | 1 | 1 | 0% | 1,461 | 1,684 | +15% | 0 | 0 | — |
case-02 | fail→fail | 8,554 | 2,271 | -73% | 1 | 1 | 0% | 1,460 | 1,100 | -25% | 0 | 0 | — |
case-05 | pass→pass | 4,487 | 4,195 | -7% | 1 | 1 | 0% | 901 | 1,608 | +78% | 0 | 0 | — |
case-06 | fail→pass | 8,255 | 1,752 | -79% | 1 | 1 | 0% | 1,464 | 1,126 | -23% | 0 | 0 | — |
case-07 | pass→pass | 10,488 | 6,851 | -35% | 1 | 1 | 0% | 1,786 | 2,014 | +13% | 0 | 0 | — |
case-08 | pass→pass | 10,609 | 4,358 | -59% | 1 | 1 | 0% | 2,079 | 1,775 | -15% | 0 | 0 | — |
case-11 | pass→pass | 7,906 | 4,233 | -46% | 1 | 1 | 0% | 1,471 | 1,640 | +11% | 0 | 0 | — |
case-12 | fail→pass | 10,187 | 4,541 | -55% | 1 | 1 | 0% | 1,989 | 1,738 | -13% | 0 | 0 | — |
case-13 | fail→pass | 11,306 | 4,631 | -59% | 1 | 1 | 0% | 2,018 | 1,712 | -15% | 0 | 0 | — |
case-14 | fail→pass | 8,981 | 6,231 | -31% | 1 | 1 | 0% | 1,788 | 1,977 | +11% | 0 | 0 | — |
case-15 | fail→pass | 9,865 | 3,504 | -64% | 1 | 1 | 0% | 1,913 | 1,465 | -23% | 0 | 0 | — |
case-17 | pass→pass | 8,574 | 2,703 | -68% | 1 | 1 | 0% | 1,520 | 1,303 | -14% | 0 | 0 | — |
case-18 | fail→pass | 9,961 | 4,733 | -52% | 1 | 1 | 0% | 1,696 | 1,801 | +6% | 0 | 0 | — |
case-19 | pass→pass | 9,315 | 3,075 | -67% | 1 | 1 | 0% | 1,624 | 1,420 | -13% | 0 | 0 | — |
case-20 | pass→pass | 9,468 | 6,634 | -30% | 1 | 1 | 0% | 1,924 | 2,256 | +17% | 0 | 0 | — |
case-21 | pass→pass | 11,467 | 9,964 | -13% | 1 | 1 | 0% | 2,390 | 3,019 | +26% | 0 | 0 | — |
case-22 | pass→pass | 6,112 | 5,164 | -16% | 1 | 1 | 0% | 1,163 | 1,780 | +53% | 0 | 0 | — |
case-23 | pass→pass | 7,678 | 4,338 | -44% | 1 | 1 | 0% | 1,410 | 1,715 | +22% | 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. 23 cases were attempted. The headline lift of +30 percentage points is the difference between those two pass rates over the 23 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.