anymorph 0.4.0 → 0.5.0

package/dist/skillpacks/geo/skills/geo-writer/references/content-writer-contract.md

@@ -1,316 +0,0 @@
-# Content Writer Contract
-
-Read this reference when building a page-generation brief, research-only outline, or handoff context for `generate_managed_pages`.
-
-## Input Contract
-
-Auto mode accepts only:
-
-```yaml
-domain: "example.com"
-lang: "ko"
-topic: "free-form topic"
-```
-
-Full mode accepts:
-
-```yaml
-domain: "example.com"
-workspaceId: "optional"
-lang: "ko"
-market: "KR"
-topic: "free-form topic"
-output_mode: "generation_brief"
-data_mode: "mcp_first"
-target_reader: "optional"
-page_goal: "optional"
-content_type: "optional"
-primary_question: "optional"
-cta_goal: "optional"
-title_selection_mode: "optional"
-route_or_slug: "optional"
-pageId: "optional"
-intentId: "optional"
-promptIds: []
-```
-
-In Auto mode, include an `Inference Summary` before the Data Pack:
-
-```markdown
-## Inference Summary
-
-- Workspace: value (source: matched from domain | user-provided | needs confirmation)
-- Language: value (source: user-provided)
-- Market: value (source: brand profile | language fallback | user-provided)
-- Output mode: value (source: inferred default | user-provided)
-- Content type: value (source: inferred from topic)
-- Target reader: value (source: brand profile | inferred)
-- Primary question: value (source: generated from topic)
-- CTA goal: value (source: workspace instructions | inferred)
-- Title selection mode: value (source: inferred default | user-provided)
-- Route/slug: value (source: inferred; collision check: pass/fail)
-```
-
-Rules:
-
-- Mark inferred fields explicitly.
-- Ask the user only when inference affects workspace identity, PRD direct SQL, route collision, or publish/generation side effects.
-- Default output mode is `generation_brief`, not live page generation.
-- Direct SQL is never enabled by Auto mode alone.
-- Default `title_selection_mode` is `user_select_before_generation`. Use `auto_recommended` only when the user explicitly asks to auto-pick the title.
-
-## Data Pack Shape
-
-```ts
-type ContentWriterDataPack = {
-  workspace: {
-    workspaceId: string;
-    domain: string;
-    globalRules?: string;
-    knowledgeBaseFacts: Array<{ source: string; fact: string; confidence: number }>;
-  };
-  brand: {
-    name: string;
-    aliases: string[];
-    description?: string;
-    uvp?: string;
-    audience?: string;
-    positioning?: unknown;
-    voiceKit?: unknown;
-    rules: unknown[];
-    bestPractices: unknown[];
-    blockList: unknown[];
-  };
-  opportunity: {
-    primaryIntent: string;
-    primaryQuestion: string;
-    contentType?: "guide" | "comparison" | "alternatives" | "what-is" | "benchmark" | "pricing-roi" | "docs-api" | "case-study" | "glossary";
-    pageGoal?: string;
-    ctaGoal?: string;
-    titleSelectionMode?: "user_select_before_generation" | "auto_recommended" | "user_provided";
-    promptSet: Array<{ promptId?: string; template: string; tier?: string; structure?: string }>;
-    seoQueries: Array<{ query: string; volume?: number; source: "gsc" | "dfs" | "db" | "user" }>;
-    fanOutQuestions: string[];
-  };
-  existingCoverage: {
-    relatedPages: Array<{ path: string; title?: string; description?: string; intentId?: string }>;
-    cannibalizationRisks: string[];
-    internalLinkTargets: Array<{ path: string; anchorReason: string }>;
-  };
-  evidence: {
-    firstParty: Array<{ id: string; claim: string; source: string; url?: string }>;
-    thirdParty: Array<{ id: string; claim: string; sourceName: string; url: string; date?: string }>;
-    proprietary: Array<{ id: string; insight: string; method?: string; sample?: string; limitation?: string }>;
-    disallowedClaims: string[];
-  };
-  editorialKernel: EditorialKernel;
-  sectionContract: SectionEvidenceContract[];
-};
-```
-
-## Editorial Kernel Template
-
-```markdown
-## Editorial Kernel
-
-- Primary question:
-- Reader job:
-- Answer promise:
-- Original angle:
-- Evidence wedge:
-- Honest brand role:
-- Hook:
-- Title posture:
-- Caveats:
-```
-
-Rules:
-
-- `answerPromise` must be a true direct answer, not a teaser.
-- `originalAngle` must explain what is not commodity content.
-- `evidenceWedge` should name the evidence type: proprietary data, product fact, benchmark, method, comparison table, first-hand workflow, or third-party source set.
-- `honestBrandRole` must include non-fit or limitation when relevant.
-- `titlePosture` must choose how brave the title should be: `orthodox`, `hybrid`, or `pov-led`.
-
-## Title Strategy
-
-Do not use one uniform title style for every page. Title posture must follow content type, reader risk, evidence strength, and funnel stage.
-
-| Title posture | Use when | Good shape | Avoid |
-| --- | --- | --- | --- |
-| `orthodox` | what-is, glossary, docs/API, pricing/ROI, regulated or high-risk topics, exact search intent pages | `Topic + audience/use case + concrete scope` | cleverness that hides the answer |
-| `hybrid` | commercial guides, service evaluation, comparison, alternatives, implementation guides | `Clear keyword + buyer tension + decision criterion` | generic "complete guide" phrasing |
-| `pov-led` | benchmark/report, category creation, thought leadership, contrarian insight, proprietary data pages | `Specific POV or surprising finding + evidence promise` | unsupported hot takes or clickbait |
-
-Rules:
-
-- Separate `seoTitle` from `h1` when useful. `seoTitle` can be safer and query-clear; `h1` can carry the human hook.
-- Keep the primary topic/entity visible in at least one title surface: `seoTitle`, `h1`, or slug.
-- Generate at least 3 title candidates: `safe`, `hybrid`, and `bold`. Pick one `recommended` candidate and explain why in one sentence.
-- Generate a Title Candidate Set, not a single title. Use 3 candidates for narrow/orthodox pages and 5-7 candidates for strategic commercial, comparison, alternatives, benchmark, or POV-led pages.
-- Candidate IDs must be stable (`T1`, `T2`, etc.) so the user can choose quickly.
-- Each candidate package should include `seoTitle`, `h1`, `socialInternalTitle`, `posture`, `bestFor`, `risk`, and a one-sentence rationale.
-- For `managed_generation`, title selection is a gate: if `title_selection_mode` is `user_select_before_generation` and the user has not chosen a candidate, stop after the brief and ask the user to choose by ID or edit one candidate.
-- If the user chooses or edits a title, treat that as `user_provided`; preserve it unless it violates trust, evidence, or anti-hype rules.
-- A bolder title is allowed only when the Data Pack has a real `originalAngle` or `evidenceWedge`; otherwise prefer `orthodox` or `hybrid`.
-- Never use unsupported superlatives such as "1위", "최고", "완벽한", "유일한", or "검증된" unless exact evidence exists.
-- Avoid stale AI-title patterns: "궁극의 가이드", "완벽 가이드", "성공을 위한", "알아야 할 모든 것", "디지털 시대의".
-
-Examples for a Korean B2B GEO service evaluation page:
-
-| Candidate type | Example |
-| --- | --- |
-| safe `seoTitle` | `GEO 서비스 선택 가이드: 한국 B2B 마케터가 봐야 할 AI 검색 가시성 기준` |
-| hybrid `h1` | `GEO 서비스는 블로그를 몇 개 써주느냐가 아니라, 어떤 AI 구매 질문에서 빠졌는지를 보여줘야 한다` |
-| checklist | `GEO 서비스 평가 체크리스트: 언급률보다 먼저 봐야 할 6가지` |
-| POC-led | `GEO 서비스 POC를 2주 안에 검증하는 법` |
-| bold, only with evidence | `AI 답변에 우리 브랜드가 빠진다면, GEO 서비스는 여기서 갈린다` |
-
-Output title strategy in every generation brief:
-
-```markdown
-## Title Strategy
-
-- Title posture: orthodox | hybrid | pov-led
-- Title selection mode: user_select_before_generation | auto_recommended | user_provided
-- Reason:
-- Recommended candidate ID:
-
-### Title Candidate Set
-
-| ID | Posture | SEO title | H1 | Social/internal title | Best for | Risk |
-| --- | --- | --- | --- | --- | --- | --- |
-| T1 | orthodox |  |  |  |  |  |
-| T2 | hybrid |  |  |  |  |  |
-| T3 | pov-led |  |  |  |  |  |
-
-Selection note: choose a candidate ID or edit one title before live generation.
-```
-
-## Section Evidence Contract
-
-```markdown
-## Section Evidence Contract
-
-1. H2:
-   - Question answered:
-   - Answer capsule:
-   - Use evidence IDs:
-   - Forbidden claims:
-   - Internal link:
-   - Format/component:
-```
-
-Rules:
-
-- Every H2 answers one concrete subquestion.
-- The answer capsule must work as a standalone quotation.
-- Evidence IDs must map to Data Pack evidence.
-- Do not invent an evidence ID to justify a section.
-- Use tables only for decision-useful comparisons.
-
-## Managed Generation Brief Template
-
-Use this shape in `generate_managed_pages.context`.
-
-```markdown
-# Page Generation Brief
-
-## Workspace and Brand
-- Workspace:
-- Domain:
-- Brand:
-- Target language/market:
-- Brand voice summary:
-- Global rules:
-
-## Page Objective
-- Primary intent:
-- Primary question:
-- Reader job:
-- Search/GEO demand:
-- Target prompt set:
-- Fan-out questions to cover:
-- Fan-out questions to skip:
-
-## Editorial Kernel
-[paste kernel]
-
-## Evidence Library
-- [E1] claim, source, URL/date
-- [E2] claim, source, URL/date
-
-## Section Evidence Contract
-[paste contract]
-
-## Internal Links
-- path: anchor reason
-
-## Claims Policy
-- Allowed claims:
-- Disallowed claims:
-- Claims requiring caveat:
-
-## Title Strategy
-- Title selection mode:
-- Recommended candidate ID:
-- Selected candidate ID or user-edited title:
-- Selection status: needs_user_selection | selected | auto_recommended
-
-### Title Candidate Set
-| ID | Posture | SEO title | H1 | Social/internal title | Best for | Risk |
-| --- | --- | --- | --- | --- | --- | --- |
-
-## Metadata Intent
-- Title posture:
-- SEO title:
-- H1:
-- Social/internal title:
-- Description intent:
-- Slug:
-- Structured data intent:
-
-## QA Rules
-- Answer-first intro and H2s.
-- No invented stats, quotes, dates, testimonials, screenshots, or product claims.
-- Use exact source URLs where statistics or studies are cited.
-- Keep brand mention natural and useful.
-- If live generation is requested and title selection status is `needs_user_selection`, stop and ask for a title candidate ID instead of calling generation.
-```
-
-## Research-Only Output Template
-
-```markdown
-## Content Brief
-
-### Why this page should exist
-
-### What question it answers
-
-### Editorial kernel
-
-### Evidence gaps
-
-### Proposed outline
-
-### Generation-ready context
-
-### Risks and caveats
-```
-
-## QA Rubric
-
-Score each 0-2. Minimum pass: 16/20 and no zero in Trust or Evidence.
-
-| Dimension | 0 | 1 | 2 |
-| --- | --- | --- | --- |
-| People-first usefulness | generic | partly useful | completes reader job |
-| E-E-A-T trust | unclear | some signals | clear who/how/why/source |
-| Originality | commodity | slight POV | unique evidence/angle |
-| Answer-first citability | buried | mixed | direct capsules |
-| Evidence density | unsupported | partial | claims mapped to evidence |
-| Fan-out coverage | misses intent | covers basics | covers key subquestions |
-| Scannability | dense | acceptable | headings/lists/tables help |
-| Hook/title quality | hype or generic | clear but plain | concrete, intent-clear, and compelling |
-| Brand honesty | forced | acceptable | useful fit/non-fit |
-| Technical readiness | unclear | partial | crawl/index/schema aware |

package/dist/skillpacks/geo/skills/geo-writer/references/direct-sql.md

@@ -1,187 +0,0 @@
-# Direct SQL Extraction
-
-Use this reference only when Anymorph MCP tools cannot return the needed read-only data slice or the user explicitly asks to query the Anymorph database.
-
-## Rule
-
-MCP first. SQL second. Read-only always.
-
-The backend endpoint can execute writes when the query is not a `SELECT`, so the agent must enforce a strict guard before every call.
-
-## Endpoint
-
-DEV:
-
-```bash
-doppler run -p anymorph-saul-mcp -c dev -- bash -lc '
-curl -s -X POST "https://api-dev.anymorph.ai/api/admin/sql" \
-  -H "Authorization: Bearer $BACKEND_INTERNAL_TOKEN" \
-  -H "Content-Type: application/json" \
-  -d "{\"query\":\"SELECT 1 AS ok\"}"
-'
-```
-
-PRD:
-
-```bash
-doppler run -p anymorph-saul-mcp -c prd -- bash -lc '
-curl -s -X POST "https://api.anymorph.ai/api/admin/sql" \
-  -H "Authorization: Bearer $BACKEND_INTERNAL_TOKEN" \
-  -H "Content-Type: application/json" \
-  -d "{\"query\":\"SELECT 1 AS ok\"}"
-'
-```
-
-If `anymorph-saul-mcp` is unavailable but backend secrets are available, the backend project pattern may expose `CRON_SECRET`; use the same endpoint with `Authorization: Bearer $CRON_SECRET`.
-
-## Required Guard
-
-Allow only:
-
-- one statement
-- starts with `SELECT` or `WITH ... SELECT`
-- scoped by `workspaceId` unless querying `information_schema`
-- has `LIMIT` for row-producing queries
-- default max rows 20, hard cap 100
-
-Reject if query contains any of:
-
-```text
-INSERT UPDATE DELETE DROP ALTER TRUNCATE CREATE REPLACE MERGE
-GRANT REVOKE COPY VACUUM ANALYZE CALL DO EXECUTE
-```
-
-Reject semicolons unless the semicolon is the final character and there is no second statement.
-
-## Schema Smoke Query
-
-```sql
-SELECT table_name
-FROM information_schema.tables
-WHERE table_schema = 'public'
-  AND table_name IN (
-    'Workspace',
-    'Brand',
-    'Page',
-    'Intent',
-    'TrackedPrompt',
-    'AnalysisItemResult',
-    'ResponseInsightV2',
-    'SearchQueryCluster',
-    'PagePrompt'
-  )
-ORDER BY table_name
-LIMIT 20;
-```
-
-## Workspace and Brand Context
-
-```sql
-SELECT
-  w.id,
-  w.domain,
-  w."globalRules",
-  left(w."knowledgeBaseContent", 4000) AS knowledge_base_excerpt,
-  b.name AS brand_name,
-  b.description,
-  b."uniqueValueProposition",
-  b."targetAudiencePrimary",
-  b."voiceAxes",
-  b.rules,
-  b."bestPractices",
-  b."blockList",
-  b."seoKeywordsData"
-FROM "Workspace" w
-LEFT JOIN "Brand" b ON b."workspaceId" = w.id
-WHERE w.id = '<workspaceId>'
-LIMIT 1;
-```
-
-## Intent and Prompt Opportunity
-
-```sql
-SELECT
-  i.id AS intent_id,
-  i.name AS intent_name,
-  i.description,
-  i.type,
-  i."combinedVolume",
-  i."currentVisibility",
-  i."competitorVisibility",
-  p.id AS prompt_id,
-  p.template,
-  p.intent,
-  p."funnelStage",
-  p."estimatedVolume",
-  p."businessValue",
-  p.structure,
-  p."promptCategory",
-  p."promptTier",
-  p."snapshotVisibilityRate",
-  p."snapshotHasMention"
-FROM "Intent" i
-LEFT JOIN "TrackedPrompt" p ON p."intentId" = i.id
-WHERE i."workspaceId" = '<workspaceId>'
-  AND i."deletedAt" IS NULL
-ORDER BY i."currentVisibility" ASC NULLS LAST, i."combinedVolume" DESC NULLS LAST
-LIMIT 50;
-```
-
-## Existing Page Coverage
-
-```sql
-SELECT
-  id,
-  path,
-  title,
-  description,
-  status,
-  "intentId",
-  "targetKeywords",
-  "targetSearchQueries",
-  archetype,
-  "lengthTargetWords",
-  "priorityScore",
-  "updatedAt"
-FROM "Page"
-WHERE "workspaceId" = '<workspaceId>'
-  AND status IN ('draft', 'preview_ready', 'deployed')
-ORDER BY "updatedAt" DESC
-LIMIT 100;
-```
-
-## Page Research Evidence
-
-```sql
-SELECT
-  id,
-  path,
-  title,
-  left("researchReport", 6000) AS research_report_excerpt,
-  "researchSources",
-  "researchedAt"
-FROM "Page"
-WHERE "workspaceId" = '<workspaceId>'
-  AND "researchReport" IS NOT NULL
-ORDER BY "researchedAt" DESC NULLS LAST
-LIMIT 20;
-```
-
-## Recent AI Answer Evidence
-
-```sql
-SELECT
-  air.id,
-  air.engine,
-  air.prompt,
-  left(air."responseText", 3000) AS response_excerpt,
-  air.citations,
-  air.mentions,
-  air."hasBrandMention",
-  air."createdAt"
-FROM "AnalysisItemResult" air
-WHERE air."workspaceId" = '<workspaceId>'
-  AND air.status = 'completed'
-ORDER BY air."createdAt" DESC
-LIMIT 50;
-```