Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 | /** * Subtask: LLM vision review of a generated postcard image. * * Uses createTaskLLM for streaming reasoning visibility in admin. * Reviews the image against configurable criteria and returns * per-criterion pass/fail results. */ import { z } from 'zod' import { createTask, createChildTask, type TaskHandle } from '../task-manager' import { createTaskLLM } from '../llm' import { readPersistentImage } from '../image-storage' import { createUsageRecordingMiddleware } from '@/lib/ai-usage/llm-middleware' import type { PostcardReviewEvent } from './events' import type { PostcardManifest } from '@/db/schema/number-line-postcards' // ── Criteria definition ── export interface ReviewCriterion { id: string name: string prompt: string } const DEFAULT_CRITERIA: ReviewCriterion[] = [ { id: 'math_accuracy', name: 'Mathematical accuracy', prompt: 'Check any visible numbers, equations, or number line markings. Are they mathematically correct? Flag incorrect arithmetic, misplaced numbers, or wrong values.', }, { id: 'number_identity', name: 'Number identity', prompt: 'The postcard should prominently feature the caller number. Verify the featured number matches the expected caller number from the manifest.', }, { id: 'screenshot_incorporation', name: 'Screenshot incorporation', prompt: 'If reference screenshots were provided, they should be visible as "photos" in the postcard (framed, pinned, taped — like a scrapbook). Check that the screenshot content is recognizable and not distorted beyond recognition.', }, { id: 'appropriateness', name: 'Child-appropriate content', prompt: 'This postcard is for a child. Flag anything inappropriate, disturbing, or unsuitable for a young audience.', }, ] // ── Result types ── export interface CriterionResult { id: string pass: boolean issues: string[] } export interface PostcardReviewInput { postcardId: string /** Storage path: category/filename for the image to review */ imagePath: string manifest: PostcardManifest hasReferenceScreenshots: boolean previousFeedback?: string criteria?: ReviewCriterion[] /** User who triggered this review — for usage tracking */ _userId?: string } export interface PostcardReviewOutput { postcardId: string pass: boolean criteriaResults: CriterionResult[] suggestions: string[] /** Pre-formatted feedback string for regeneration prompts */ feedback: string } // ── Schema for structured LLM output ── function buildReviewSchema(criteria: ReviewCriterion[]) { return z.object({ criteria: z.array( z.object({ id: z.enum(criteria.map((c) => c.id) as [string, ...string[]]), pass: z.boolean(), issues: z.array(z.string()), }) ), suggestions: z.array(z.string()), }) } // ── Task handler ── type ReviewHandler = ( handle: TaskHandle<PostcardReviewOutput, PostcardReviewEvent>, input: PostcardReviewInput ) => Promise<void> const handler: ReviewHandler = async (handle, config) => { const { postcardId, imagePath, manifest, hasReferenceScreenshots, previousFeedback, criteria: customCriteria, } = config const criteria = customCriteria ?? DEFAULT_CRITERIA // Filter out screenshot criterion if no references were provided const activeCriteria = hasReferenceScreenshots ? criteria : criteria.filter((c) => c.id !== 'screenshot_incorporation') handle.emit({ type: 'review_analyzing', postcardId, criteriaCount: activeCriteria.length }) handle.setProgress(10, 'Loading image for review...') // Load the draft image const [category, filename] = imagePath.split('/') const imageData = await readPersistentImage(category, filename) if (!imageData) { handle.fail(`Image not found at ${imagePath}`) return } const imageDataUrl = `data:image/png;base64,${imageData.buffer.toString('base64')}` // Build review prompt const callerNum = manifest.callerNumber const displayNum = Number.isInteger(callerNum) ? callerNum.toString() : callerNum.toPrecision(6) const criteriaBlock = activeCriteria .map((c, i) => `${i + 1}. **${c.name}** (id: "${c.id}")\n ${c.prompt}`) .join('\n\n') const prompt = [ `You are reviewing a generated postcard image for quality before it is delivered to a child.`, ``, `## Context`, `- Caller number: ${displayNum}`, `- Child's name: ${manifest.childName}`, manifest.childAge ? `- Child's age: ${manifest.childAge}` : '', `- Number personality: ${manifest.callerPersonality}`, `- Reference screenshots provided: ${hasReferenceScreenshots ? 'yes' : 'no'}`, ``, `## Review Criteria`, `Evaluate the image against each criterion below. For each, determine pass/fail and list specific issues found.`, ``, criteriaBlock, ``, previousFeedback ? `## Previous Feedback\nThis is a regeneration attempt. The previous version had these issues:\n${previousFeedback}\n\nCheck whether these issues have been resolved.\n` : '', `## Output`, `For each criterion, provide:`, `- "id": the criterion id`, `- "pass": true if no issues, false if issues found`, `- "issues": array of specific issues (empty if pass)`, ``, `Also provide "suggestions": actionable feedback for regeneration if any criteria fail.`, ] .filter(Boolean) .join('\n') handle.setProgress(20, 'Analyzing image...') // Use streaming LLM via createTaskLLM for admin visibility const taskLLM = createTaskLLM( handle as TaskHandle<PostcardReviewOutput, PostcardReviewEvent>, config._userId ? createUsageRecordingMiddleware({ userId: config._userId, feature: 'postcard:review', backgroundTaskId: handle.id, }) : undefined ) const schema = buildReviewSchema(activeCriteria) let reviewData: z.infer<typeof schema> | undefined for await (const event of taskLLM.stream({ prompt, images: [imageDataUrl], schema, provider: 'openai', model: 'gpt-5.4', reasoning: { effort: 'medium', summary: 'auto' }, })) { if (event.type === 'complete') { reviewData = event.data } else if (event.type === 'error') { handle.fail(`Review LLM error: ${event.message}`) return } } if (!reviewData) { handle.fail('Review produced no output') return } handle.setProgress(80, 'Processing review results...') // Map results, filling in any missing criteria const criteriaResults: CriterionResult[] = activeCriteria.map((c) => { const found = reviewData!.criteria.find((r) => r.id === c.id) return found ?? { id: c.id, pass: true, issues: [] } }) // Emit per-criterion results for (const result of criteriaResults) { handle.emit({ type: 'review_criterion_result', postcardId, criterionId: result.id, pass: result.pass, issues: result.issues, }) } const pass = criteriaResults.every((r) => r.pass) const allIssues = criteriaResults.flatMap((r) => r.issues) handle.emit({ type: 'review_complete', postcardId, pass, issueCount: allIssues.length, }) // Build feedback string for regeneration const feedback = formatReviewFeedback(criteriaResults, reviewData.suggestions) handle.complete({ postcardId, pass, criteriaResults, suggestions: reviewData.suggestions, feedback, }) } export async function startPostcardReview( input: PostcardReviewInput, userId?: string, parentTaskId?: string ): Promise<string> { if (parentTaskId) { return createChildTask<PostcardReviewInput, PostcardReviewOutput, PostcardReviewEvent>( parentTaskId, 'postcard-review', input, handler, userId ) } return createTask<PostcardReviewInput, PostcardReviewOutput, PostcardReviewEvent>( 'postcard-review', input, handler, userId ) } function formatReviewFeedback(criteriaResults: CriterionResult[], suggestions: string[]): string { const failedCriteria = criteriaResults.filter((r) => !r.pass) if (failedCriteria.length === 0) return '' const issueLines = failedCriteria.map((r) => `- ${r.id}: ${r.issues.join('; ')}`).join('\n') const suggestionLines = suggestions.length > 0 ? `\nSuggestions:\n${suggestions.map((s) => `- ${s}`).join('\n')}` : '' return `Issues found in previous generation:\n${issueLines}${suggestionLines}\n\nPlease fix these issues in the new image.` } |