Every structured-output call now uses an Anthropic tool instead of
parsing JSON out of free-form prose, and stable system prompts are
sent as cacheable blocks. Behaviour-equivalent to phase 1 from the
caller's point of view; the savings show up in token usage and in the
absence of "AI returned non-JSON response" failure modes.
* src/lib/llmTools.js — single source of truth for tool definitions:
emit_knowledge_graph, emit_handbook_delta, emit_learning_article /
_slides / _infographic / _all, emit_custom_topic, emit_quiz_questions,
emit_graph_actions, plus five article-patch tools (set_intro,
set_section, add_section, remove_section, replace_takeaways).
* src/lib/articlePatches.js — pure applyArticlePatches +
applyAndValidate; rebuilds the article from a sequence of patch tool
calls and re-validates against learningArticleSchema. set_section
falls back to appending when no matching heading exists so the
model's intent is preserved rather than silently dropped.
* src/lib/llmSchemas.js — Zod schemas for the five patch ops,
registered in toolSchemaRegistry so callLLM validates them
automatically.
* src/lib/llm.js — simulation mode now returns a tool_use stub matching
toolChoice.name, so the UI keeps working with Simulation Mode on
after the structured-output migration.
* src/lib/extractionPipeline.js — processSourceText and
analyzeHandbookDelta migrated to callLLM + tool use. System prompts
sent as { cache_control: ephemeral } blocks. Handbook results pass
through normalizeHandbookResult to collapse legacy "executes"
relations into executed_by with swapped source/target.
* src/lib/learningService.js — generateLearningContent picks the right
tool per selectedType; generateCustomTopic uses emit_custom_topic;
refineLearningContent now drives the five patch tools with
toolChoice 'any' and rejects the whole turn if the patched article
fails validation. Article-only refinement is intentional for phase 2;
refining a topic without an article surfaces a clear error.
* src/lib/testService.js — quiz generation via emit_quiz_questions.
* src/components/admin/KnowledgeGraph.jsx — analyzeGraph routed through
the reasoning tier (Opus) since graph-wide consolidation benefits
from a stronger reasoner.
* src/components/chat/prompts.js — buildSystemPrompt now returns three
text blocks: stable preamble (cached), KB context (cached, hash-bust
deferred to phase 5), per-turn user/admin tail (uncached).
* src/lib/__tests__/ — 13 new tests covering each patch op, multi-op
sequencing, post-patch validation failure, and tool/registry shape.
Acceptance: lint and 45/45 tests green; build succeeds; no
`match(/\{[\s\S]*\}/)` JSON extraction left in src/. Live verification
of cache hits on a second extraction within 5 minutes is deferred to
manual smoke testing — needs real `/api/anthropic` traffic.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
233 lines
6.5 KiB
JavaScript
233 lines
6.5 KiB
JavaScript
/**
|
|
* Zod schemas for every structured LLM output the platform consumes.
|
|
*
|
|
* Field names mirror what callers already produce — do not rename them
|
|
* without migrating the corresponding service module.
|
|
*/
|
|
|
|
import { z } from 'zod';
|
|
|
|
const topicTypeEnum = z.enum(['concept', 'role', 'process']);
|
|
const relationTypeStrict = z.enum(['related_to', 'depends_on', 'part_of', 'executed_by']);
|
|
const relationTypeLoose = z.enum(['related_to', 'depends_on', 'part_of', 'executed_by', 'executes']);
|
|
const learningRelevanceEnum = z.enum(['core', 'standard', 'peripheral', 'exclude']);
|
|
|
|
const extractionTopicSchema = z.object({
|
|
id: z.string().min(1),
|
|
label: z.string().min(1),
|
|
type: topicTypeEnum,
|
|
description: z.string().min(1),
|
|
learning_relevance: learningRelevanceEnum,
|
|
});
|
|
|
|
const extractionRelationSchema = z.object({
|
|
source: z.string().min(1),
|
|
target: z.string().min(1),
|
|
type: relationTypeStrict,
|
|
});
|
|
|
|
export const extractionResultSchema = z.object({
|
|
topics: z.array(extractionTopicSchema),
|
|
relations: z.array(extractionRelationSchema),
|
|
});
|
|
|
|
const handbookTopicSchema = extractionTopicSchema.extend({
|
|
metadata: z.object({ source: z.string() }).optional(),
|
|
});
|
|
|
|
const handbookRelationSchema = z.object({
|
|
source: z.string().min(1),
|
|
target: z.string().min(1),
|
|
type: relationTypeLoose,
|
|
description: z.string().optional(),
|
|
});
|
|
|
|
export const handbookResultSchema = z.object({
|
|
topics: z.array(handbookTopicSchema),
|
|
relations: z.array(handbookRelationSchema),
|
|
});
|
|
|
|
/**
|
|
* Normalise legacy `executes` relations into the canonical `executed_by`
|
|
* vocabulary by swapping source and target. The handbook prompt previously
|
|
* emitted `role → executes → process`; the canonical form is
|
|
* `process → executed_by → role`.
|
|
*/
|
|
export function normalizeHandbookResult(parsed) {
|
|
return {
|
|
...parsed,
|
|
relations: parsed.relations.map((r) =>
|
|
r.type === 'executes'
|
|
? { ...r, type: 'executed_by', source: r.target, target: r.source }
|
|
: r,
|
|
),
|
|
};
|
|
}
|
|
|
|
const articleSectionSchema = z.object({
|
|
heading: z.string().min(1),
|
|
body: z.string().min(1),
|
|
});
|
|
|
|
const articleBodySchema = z.object({
|
|
title: z.string().min(1),
|
|
intro: z.string().min(1),
|
|
sections: z.array(articleSectionSchema).min(1),
|
|
keyTakeaways: z.array(z.string().min(1)).min(1),
|
|
});
|
|
|
|
export const learningArticleSchema = z.object({
|
|
article: articleBodySchema,
|
|
});
|
|
|
|
const slideSchema = z.object({
|
|
title: z.string().min(1),
|
|
bullets: z.array(z.string().min(1)).min(1),
|
|
speakerNote: z.string().min(1),
|
|
});
|
|
|
|
export const learningSlidesSchema = z.object({
|
|
slides: z.array(slideSchema).min(1),
|
|
});
|
|
|
|
const infographicStatSchema = z.object({
|
|
value: z.string().min(1),
|
|
label: z.string().min(1),
|
|
icon: z.string().min(1),
|
|
});
|
|
|
|
const infographicStepSchema = z.object({
|
|
number: z.number().int().min(1),
|
|
title: z.string().min(1),
|
|
description: z.string().min(1),
|
|
icon: z.string().min(1),
|
|
});
|
|
|
|
const infographicBodySchema = z.object({
|
|
headline: z.string().min(1),
|
|
tagline: z.string().min(1),
|
|
stats: z.array(infographicStatSchema).min(1),
|
|
steps: z.array(infographicStepSchema).min(1),
|
|
quote: z.string().min(1),
|
|
colorTheme: z.string().min(1),
|
|
});
|
|
|
|
export const learningInfographicSchema = z.object({
|
|
infographic: infographicBodySchema,
|
|
});
|
|
|
|
export const learningAllSchema = z.object({
|
|
article: articleBodySchema,
|
|
slides: z.array(slideSchema).min(1),
|
|
infographic: infographicBodySchema,
|
|
});
|
|
|
|
const quizQuestionSchema = z.object({
|
|
id: z.string().min(1),
|
|
question: z.string().min(1),
|
|
topicLabel: z.string().min(1),
|
|
options: z.array(z.string().min(1)).length(4),
|
|
correctIndex: z.number().int().min(0).max(3),
|
|
explanation: z.string().min(1),
|
|
});
|
|
|
|
export const quizQuestionsSchema = z.object({
|
|
questions: z.array(quizQuestionSchema).min(1),
|
|
});
|
|
|
|
export const customTopicSchema = z.object({
|
|
label: z.string().min(1),
|
|
type: topicTypeEnum,
|
|
description: z.string().min(1),
|
|
});
|
|
|
|
const mergeActionSchema = z.object({
|
|
keepId: z.string().min(1),
|
|
deleteId: z.string().min(1),
|
|
});
|
|
|
|
const newRelationSchema = z.object({
|
|
source: z.string().min(1),
|
|
target: z.string().min(1),
|
|
type: relationTypeStrict,
|
|
});
|
|
|
|
const relevanceUpdateSchema = z.object({
|
|
id: z.string().min(1),
|
|
learning_relevance: learningRelevanceEnum,
|
|
});
|
|
|
|
export const graphActionsSchema = z.object({
|
|
merges: z.array(mergeActionSchema).optional().default([]),
|
|
deletions: z.array(z.string().min(1)).optional().default([]),
|
|
newRelations: z.array(newRelationSchema).optional().default([]),
|
|
relevanceUpdates: z.array(relevanceUpdateSchema).optional().default([]),
|
|
});
|
|
|
|
const deltaTopicSchema = z.object({
|
|
id: z.string().min(1),
|
|
label: z.string().min(1),
|
|
type: topicTypeEnum,
|
|
description: z.string().min(1),
|
|
});
|
|
|
|
const deltaRelationSchema = z.object({
|
|
source: z.string().min(1),
|
|
target: z.string().min(1),
|
|
type: relationTypeStrict,
|
|
});
|
|
|
|
export const proposeGraphDeltaSchema = z.object({
|
|
reason: z.string().min(1),
|
|
topics: z.array(deltaTopicSchema).max(3).optional(),
|
|
relations: z.array(deltaRelationSchema).max(5).optional(),
|
|
});
|
|
|
|
// ── Article patch operation schemas (Phase 2.4) ──────────────────────────────
|
|
|
|
export const setIntroPatchSchema = z.object({
|
|
intro: z.string().min(1),
|
|
});
|
|
|
|
export const setSectionPatchSchema = z.object({
|
|
heading: z.string().min(1),
|
|
body: z.string().min(1),
|
|
});
|
|
|
|
export const addSectionPatchSchema = z.object({
|
|
heading: z.string().min(1),
|
|
body: z.string().min(1),
|
|
position: z.enum(['start', 'end']),
|
|
});
|
|
|
|
export const removeSectionPatchSchema = z.object({
|
|
heading: z.string().min(1),
|
|
});
|
|
|
|
export const replaceTakeawaysPatchSchema = z.object({
|
|
items: z.array(z.string().min(1)).min(1),
|
|
});
|
|
|
|
/**
|
|
* Registry mapping known tool names to their input schemas. `callLLM`
|
|
* consults this when the caller does not pass an explicit `toolSchemas`
|
|
* override.
|
|
*/
|
|
export const toolSchemaRegistry = {
|
|
emit_knowledge_graph: extractionResultSchema,
|
|
emit_handbook_delta: handbookResultSchema,
|
|
emit_learning_article: learningArticleSchema,
|
|
emit_learning_slides: learningSlidesSchema,
|
|
emit_learning_infographic: learningInfographicSchema,
|
|
emit_learning_all: learningAllSchema,
|
|
emit_quiz_questions: quizQuestionsSchema,
|
|
emit_custom_topic: customTopicSchema,
|
|
emit_graph_actions: graphActionsSchema,
|
|
propose_graph_delta: proposeGraphDeltaSchema,
|
|
set_intro: setIntroPatchSchema,
|
|
set_section: setSectionPatchSchema,
|
|
add_section: addSectionPatchSchema,
|
|
remove_section: removeSectionPatchSchema,
|
|
replace_takeaways: replaceTakeawaysPatchSchema,
|
|
};
|