Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Analyze brownfield codebase and create initial continuity ledger
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -59% | 0% |
Analyze a brownfield codebase and create an initial continuity ledger.
Spawn the onboard agent:
Use the Task tool with subagent_type: "onboard" and this prompt:
Onboard me to this project at $CLAUDE_PROJECT_DIR.
1. Create required directories if they don't exist:
mkdir -p thoughts/shared/handoffs/<project-name> .claude
2. Explore the codebase using available tools:
- Try: tldr tree . && tldr structure .
- Fallback: find . -type f -name "*.py" -o -name "*.ts" -o -name "*.js" | head -50
3. Detect tech stack (look for package.json, requirements.txt, Cargo.toml, go.mod, etc.)
4. Ask the user about their goals using AskUserQuestion
5. Create a YAML handoff at thoughts/shared/handoffs/<project-name>/onboard-<date>.yaml:
---
date: <ISO date>
type: onboard
status: active
---
goal: <user's stated goal>
now: Start working on <first priority>
tech_stack: [list of detected technologies]
key_files:
- path: <important file>
purpose: <what it does>
architecture: <brief description>
next:
- <suggested first action>The onboard process:
thoughts/shared/handoffs/<project>/, .claude//create_plan instead/create_handoff| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,825 | 10,436 | +79% | 1 | 1 | 0% | 629 | 1,376 | +119% | 0 | 0 | — |
case-02 | fail→fail | 9,825 | 10,116 | +3% | 1 | 1 | 0% | 1,701 | 1,316 | -23% | 0 | 0 | — |
case-03 | fail→fail | 1,474 | 9,536 | +547% | 1 | 1 | 0% | 217 | 1,273 | +487% | 0 | 0 | — |
case-04 | pass→pass | 15,255 | 9,686 | -37% | 1 | 1 | 0% | 2,526 | 2,140 | -15% | 0 | 0 | — |
case-05 | pass→pass | 11,426 | 5,844 | -49% | 1 | 1 | 0% | 1,579 | 1,486 | -6% | 0 | 0 | — |
case-06 | pass→pass | 3,121 | 7,571 | +143% | 1 | 1 | 0% | 485 | 1,123 | +132% | 0 | 0 | — |
case-07 | fail→pass | 13,641 | 5,738 | -58% | 1 | 1 | 0% | 2,246 | 1,594 | -29% | 0 | 0 | — |
case-08 | fail→pass | 12,755 | 2,349 | -82% | 1 | 1 | 0% | 2,182 | 945 | -57% | 0 | 0 | — |
case-09 | fail→fail | 13,204 | 11,079 | -16% | 1 | 1 | 0% | 2,224 | 2,447 | +10% | 0 | 0 | — |
case-10 | fail→pass | 9,297 | 2,613 | -72% | 1 | 1 | 0% | 2,005 | 894 | -55% | 0 | 0 | — |
case-11 | pass→pass | 14,339 | 14,344 | +0% | 1 | 1 | 0% | 3,047 | 2,806 | -8% | 0 | 0 | — |
case-12 | fail→fail | 15,254 | 11,397 | -25% | 1 | 1 | 0% | 2,353 | 2,458 | +4% | 0 | 0 | — |
case-13 | fail→pass | 9,793 | 2,572 | -74% | 1 | 1 | 0% | 1,422 | 942 | -34% | 0 | 0 | — |
case-14 | fail→pass | 10,110 | 1,783 | -82% | 1 | 1 | 0% | 1,885 | 772 | -59% | 0 | 0 | — |
case-15 | fail→pass | 11,460 | 3,224 | -72% | 1 | 1 | 0% | 1,966 | 1,053 | -46% | 0 | 0 | — |
case-16 | fail→pass | 12,229 | 2,531 | -79% | 1 | 1 | 0% | 1,964 | 954 | -51% | 0 | 0 | — |
case-17 | fail→pass | 11,640 | 3,976 | -66% | 1 | 1 | 0% | 1,993 | 1,179 | -41% | 0 | 0 | — |
case-18 | fail→pass | 11,720 | 2,794 | -76% | 1 | 1 | 0% | 1,814 | 932 | -49% | 0 | 0 | — |
case-19 | pass→pass | 13,791 | 4,643 | -66% | 1 | 1 | 0% | 1,990 | 1,211 | -39% | 0 | 0 | — |
case-20 | pass→pass | 7,228 | 2,142 | -70% | 1 | 1 | 0% | 1,224 | 883 | -28% | 0 | 0 | — |
case-21 | pass→pass | 11,576 | 8,296 | -28% | 1 | 1 | 0% | 1,896 | 1,334 | -30% | 0 | 0 | — |
case-22 | pass→pass | 16,344 | 6,693 | -59% | 1 | 1 | 0% | 2,369 | 1,614 | -32% | 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 19 counted toward the lift figure. The other 3 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 +41 percentage points is the difference between those two pass rates over the 19 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/29/2026 | +59% |
Other measured skills in the registry, with their headline benchmark lift.