Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Read an existing repository's structure into the project cwd as a normalised snapshot the agent can analyse without re-walking the tree on every turn.
.claude/skills/nexu-io-code-import/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -8% | 0% |
Spec §10 / §21.3.2: a code-migration / tune-collab run starts from a real repo (cloned via od project import or a path the user provided). This atom turns the raw checkout into a normalised on-disk record that subsequent atoms (design-extract, rewrite-plan, patch-edit) operate against. The point is to stop re-walking the tree on every turn — the agent reads code/index.json once and trusts the snapshot until the next explicit re-import.
| Source | Required | Notes | | --- | --- | --- | | repoPath | yes | Absolute path to the repo's root, supplied via od project import or an upstream od.context.assets[] reference | | targetStack | yes | { framework, packageManager, styleSystem, componentLibrary } via the matching form GenUI surface |
textproject-cwd/ └── code/ ├── index.json # { files: { path, size, language, imports[] }[], packageJson?, lockfileKind, framework, styleSystem } ├── components.json # detected components with file path + props snapshot ├── routes.json # detected routing model (next/app, react-router, vite-router, …) └── meta.json # { repoPath, gitSha, walkedAt, atomDigest, walkBudgetMs }
code/index.json is the input every other Phase 7 atom reads. The walk respects the runner's budgetMs option (default 60s) so large monorepos don't burn an entire run on import.
The atom completes when code/index.json exists and contains at least one entry. Empty repos abort with a clear error event so the user re-imports.
code/index.json.skipped[] with reason).
package.json's declared dep + a build-config presence (next.config.*, vite.config.*, etc.).
Implemented by the daemon runner in apps/daemon/src/plugins/atoms/code-import.ts. The walker enforces its exclusions and time budget, detects framework evidence, and writes code/index.json.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,612 | 15,648 | +179% | 1 | 1 | 0% | 275 | 852 | +210% | 0 | 0 | — |
case-02 | fail→pass | 22,789 | 14,653 | -36% | 1 | 1 | 0% | 3,683 | 3,932 | +7% | 0 | 0 | — |
case-03 | fail→pass | 32,542 | 25,031 | -23% | 1 | 1 | 0% | 8,275 | 4,607 | -44% | 0 | 0 | — |
case-04 | fail→pass | 9,305 | 3,128 | -66% | 1 | 1 | 0% | 1,458 | 1,134 | -22% | 0 | 0 | — |
case-05 | fail→pass | 10,994 | 4,706 | -57% | 1 | 1 | 0% | 1,643 | 1,256 | -24% | 0 | 0 | — |
case-06 | pass→pass | 26,881 | 3,265 | -88% | 1 | 1 | 0% | 1,642 | 1,049 | -36% | 0 | 0 | — |
case-07 | fail→pass | 9,929 | 6,930 | -30% | 1 | 1 | 0% | 1,759 | 1,619 | -8% | 0 | 0 | — |
case-08 | fail→pass | 30,351 | 2,478 | -92% | 1 | 1 | 0% | 2,215 | 968 | -56% | 0 | 0 | — |
case-17 | fail→pass | 17,920 | 2,392 | -87% | 1 | 1 | 0% | 2,542 | 927 | -64% | 0 | 0 | — |
case-09 | fail→pass | 45,916 | 2,057 | -96% | 1 | 1 | 0% | 2,358 | 899 | -62% | 0 | 0 | — |
case-10 | fail→pass | 11,983 | 7,324 | -39% | 1 | 1 | 0% | 1,899 | 1,698 | -11% | 0 | 0 | — |
case-11 | pass→pass | 16,740 | 6,102 | -64% | 1 | 1 | 0% | 1,983 | 1,377 | -31% | 0 | 0 | — |
case-12 | fail→fail | 41,232 | 1,785 | -96% | 1 | 1 | 0% | 2,761 | 840 | -70% | 0 | 0 | — |
case-13 | fail→pass | 13,474 | 4,845 | -64% | 1 | 1 | 0% | 1,784 | 1,029 | -42% | 0 | 0 | — |
case-14 | pass→pass | 12,467 | 9,242 | -26% | 1 | 1 | 0% | 1,604 | 1,542 | -4% | 0 | 0 | — |
case-15 | fail→fail | 10,911 | 2,805 | -74% | 1 | 1 | 0% | 1,798 | 928 | -48% | 0 | 0 | — |
case-16 | pass→pass | 15,112 | 1,966 | -87% | 1 | 1 | 0% | 2,070 | 823 | -60% | 0 | 0 | — |
case-18 | fail→pass | 25,903 | 2,198 | -92% | 1 | 1 | 0% | 3,320 | 908 | -73% | 0 | 0 | — |
case-19 | pass→pass | 6,754 | 1,956 | -71% | 1 | 1 | 0% | 864 | 850 | -2% | 0 | 0 | — |
case-20 | fail→fail | 6,036 | 6,420 | +6% | 1 | 1 | 0% | 363 | 888 | +145% | 0 | 0 | — |
case-21 | fail→fail | 5,373 | 8,015 | +49% | 1 | 1 | 0% | 228 | 918 | +303% | 0 | 0 | — |
case-22 | pass→pass | 25,488 | 13,088 | -49% | 1 | 1 | 0% | 3,055 | 2,253 | -26% | 0 | 0 | — |
case-23 | fail→pass | 11,041 | 2,317 | -79% | 1 | 1 | 0% | 1,682 | 907 | -46% | 0 | 0 | — |
case-24 | fail→pass | 11,412 | 4,050 | -65% | 1 | 1 | 0% | 2,050 | 1,175 | -43% | 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. 24 cases were attempted, and 20 counted toward the lift figure. The other 4 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 +54 percentage points is the difference between those two pass rates over the 20 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.