Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate process JS files following Babysitter SDK patterns including task definitions, quality gates, breakpoints, and proper io configuration.
.claude/skills/a5c-ai-process-generator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 141% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 87% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 61% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 190% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 62% | 0% |
You are process-generator - a specialized skill for generating Babysitter SDK process files with proper structure, task definitions, and quality gates.
This skill generates complete process JS files including:
Generate complete process files:
javascript/** * @process specialization/process-name * @description Process description * @inputs { param1: type, param2: type } * @outputs { result: type, artifacts: array } */ import { defineTask } from '@a5c-ai/babysitter-sdk'; export async function process(inputs, ctx) { const { param1, param2 = 'default' } = inputs; const artifacts = []; // Phase 1 ctx.log('info', 'Phase 1: Description'); const result1 = await ctx.task(task1, { param1 }); artifacts.push(...result1.artifacts); // Breakpoint await ctx.breakpoint({ question: 'Review phase 1?', title: 'Phase 1 Review', context: { runId: ctx.runId, files: artifacts } }); return { success: true, artifacts }; } export const task1 = defineTask('task-name', (args, taskCtx) => ({ kind: 'agent', title: 'Task title', skill: { name: 'skill-name' }, agent: { name: 'agent-name', prompt: { role: 'Role description', task: 'Task description', context: args, instructions: ['instruction1', 'instruction2'], outputFormat: 'JSON with fields...' }, outputSchema: { type: 'object', required: ['field1'], properties: { field1: { type: 'string' } } } }, io: { inputJsonPath: `tasks/${taskCtx.effectId}/input.json`, outputJsonPath: `tasks/${taskCtx.effectId}/result.json` }, labels: ['agent', 'category'] }));
Support multiple task kinds:
agent - LLM agent tasksskill - Claude Code skill tasksnode - Node.js script tasksshell - Shell script tasksbreakpoint - Human approval gatesAdd quality gates at decision points:
javascript// Quality scoring task const qualityScore = await ctx.task(qualityScoringTask, { artifact: result.artifact, criteria: ['completeness', 'accuracy'] }); const qualityMet = qualityScore.score >= 80;
Place breakpoints strategically:
javascriptawait ctx.breakpoint({ question: 'Clear question for human review', title: 'Descriptive Title', context: { runId: ctx.runId, files: artifacts.map(a => ({ path: a.path, format: a.format, label: a.label })), summary: { key: 'value' } } });
json{ "processFile": "path/to/process.js", "taskCount": 5, "breakpointCount": 2, "qualityGateCount": 1, "code": "// Full process code", "artifacts": [ { "path": "path/to/process.js", "type": "javascript", "label": "Process file" } ] }
This skill integrates with:
process-creation.js - Primary process generationphase3-implement-processes.js - Batch process creationspecialization-creation.js - Full specialization workflow| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 16,570 | 28,739 | +73% | 1 | 1 | 0% | 2,638 | 6,353 | +141% | 0 | 0 | — |
case-02 | fail→pass | 19,748 | 28,475 | +44% | 1 | 1 | 0% | 3,253 | 6,074 | +87% | 0 | 0 | — |
case-03 | fail→pass | 22,491 | 28,760 | +28% | 1 | 1 | 0% | 3,731 | 6,005 | +61% | 0 | 0 | — |
case-04 | fail→pass | 14,273 | 30,273 | +112% | 1 | 1 | 0% | 2,750 | 7,982 | +190% | 0 | 0 | — |
case-05 | fail→pass | 25,431 | 41,386 | +63% | 1 | 1 | 0% | 5,117 | 8,266 | +62% | 0 | 0 | — |
case-06 | fail→pass | 18,269 | 17,766 | -3% | 1 | 1 | 0% | 2,863 | 4,902 | +71% | 0 | 0 | — |
case-07 | fail→pass | 13,746 | 13,194 | -4% | 1 | 1 | 0% | 1,995 | 3,099 | +55% | 0 | 0 | — |
case-08 | fail→pass | 14,505 | 12,446 | -14% | 1 | 1 | 0% | 2,513 | 2,971 | +18% | 0 | 0 | — |
case-09 | fail→pass | 12,167 | 11,588 | -5% | 1 | 1 | 0% | 2,139 | 3,331 | +56% | 0 | 0 | — |
case-10 | fail→pass | 14,758 | 17,093 | +16% | 1 | 1 | 0% | 2,853 | 2,413 | -15% | 0 | 0 | — |
case-11 | fail→pass | 15,511 | 22,395 | +44% | 1 | 1 | 0% | 2,706 | 4,785 | +77% | 0 | 0 | — |
case-12 | pass→pass | 18,081 | 20,686 | +14% | 1 | 1 | 0% | 2,913 | 4,559 | +57% | 0 | 0 | — |
case-13 | fail→pass | 44,094 | 27,202 | -38% | 1 | 1 | 0% | 3,826 | 7,249 | +89% | 0 | 0 | — |
case-14 | fail→pass | 20,755 | 13,824 | -33% | 1 | 1 | 0% | 3,263 | 3,840 | +18% | 0 | 0 | — |
case-15 | fail→pass | 15,189 | 12,012 | -21% | 1 | 1 | 0% | 2,563 | 3,723 | +45% | 0 | 0 | — |
case-16 | fail→pass | 17,540 | 31,220 | +78% | 1 | 1 | 0% | 3,676 | 6,272 | +71% | 0 | 0 | — |
case-17 | fail→pass | 20,150 | 19,151 | -5% | 1 | 1 | 0% | 3,326 | 5,662 | +70% | 0 | 0 | — |
case-18 | pass→pass | 13,629 | 13,068 | -4% | 1 | 1 | 0% | 2,687 | 3,185 | +19% | 0 | 0 | — |
case-19 | pass→pass | 21,771 | 34,111 | +57% | 1 | 1 | 0% | 4,480 | 6,902 | +54% | 0 | 0 | — |
case-20 | fail→fail | 8,525 | 13,662 | +60% | 1 | 1 | 0% | 1,506 | 3,118 | +107% | 0 | 0 | — |
case-21 | pass→pass | 7,875 | 7,899 | +0% | 1 | 1 | 0% | 1,506 | 2,240 | +49% | 0 | 0 | — |
case-22 | pass→fail | 15,761 | 24,397 | +55% | 1 | 1 | 0% | 2,636 | 5,527 | +110% | 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 +68 percentage points is the difference between those two pass rates over the 22 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.