anymorph 0.4.0 → 0.6.0

package/dist/skillpacks/geo/skills/geo-pages-diagnosis/references/workflow.md

@@ -1,269 +0,0 @@
-# GEO Pages Diagnosis Workflow
-
-## Role
-
-The `geo_pages` channel diagnosis diagnoses and proposes actions for Anymorph-generated GEO pages.
-
-It works on pages where:
-
-- `source = generated`
-- `renderMode = CMS`
-- the action would create, iterate, re-research, regenerate, or retire a GEO page
-
-## Non-Goals
-
-- Do not update customer-owned discovered pages.
-- Do not choose final production routes. The page generation executor owns route creation.
-- Do not propose third-party outreach in MVP.
-- Do not use legacy Astro generation paths.
-
-## Inputs
-
-- candidate rows routed by the orchestrator
-- tenant `agent/BRAND.md`
-- tenant `agent/STRATEGY.md`
-- tenant `agent/LEARNINGS.md`
-- relevant rows from:
-  - `geo_pages`
-  - `geo_intent_coverage`
-  - `geo_recommendation_quadrants`
-  - `geo_page_intent_visibility_gaps`
-  - `geo_page_performance`
-  - `geo_strategy_actions_recent`
-- page age and indexation signals when present:
-  - `createdAt`, `generatedAt`, `deployedAt`, `publishedAt`, or equivalent page timestamp
-  - indexation state, indexed URL inspection state, GSC impressions/clicks, crawler logs, or sitemap/canonical errors
-- generated page scrape when needed
-- representative prompts
-- AI answer samples
-- competitor cited/recommended page scrape when needed
-
-## Workflow
-
-### 1. Filter Candidates
-
-Keep candidates where:
-
-- generated page exists and the intent is weak
-- no page exists and the intent is weak
-- generated page is healthy and may support adjacent expansion
-
-Drop candidates when:
-
-- recent action mask suppresses the same action
-- tenant strategy forbids generated pages for that intent class
-- evidence is too thin
-- the only negative evidence is missing performance data for a fresh generated page still inside the minimum observation window
-
-### 2. Classify Candidate Case
-
-Assign one primary case.
-
-### Freshness and Indexing Guard
-
-Before using Case A for an existing generated page, check page age and indexation state.
-
-Use a minimum observation window of 7 days after the page was first generated, deployed, or published. If the generated page was published less than 7 days ago, default to `no_action` / monitor unless there is a concrete non-performance defect:
-
-- wrong intent or prompt mismatch
-- broken route, bad canonical, blocked crawl, noindex, sitemap omission, or render failure
-- duplicate/cannibalizing page
-- missing brand/entity clarity visible in the page content
-- factual, compliance, or product-claim defect
-
-For fresh pages, do not use low visibility, zero traffic, zero GSC impressions, or zero AI citations alone as evidence for `update_existing_page`, `replace_existing_page`, or `remove_or_merge_page`. Record the page in `rejected` or `noActionRationale` with a monitor note and the next evaluation date when possible.
-
-If a generated page is not indexed after 7 days, flag this in the analysis. Treat it as a possible `technical_issue` or `insufficient_evidence` finding depending on the evidence:
-
-- Use `technical_issue` when there is an indexation, crawlability, canonical, sitemap, robots, noindex, render, or status-code signal.
-- Use `insufficient_evidence` when indexation status is unknown but there are still no GSC impressions, crawler hits, or external visibility signals after the 7-day window.
-- Prefer a diagnostic note or SEO technical follow-up over a content update unless the page content itself is also defective.
-
-#### A. Generated Page Exists But Visibility Is Low
-
-Use when:
-
-- `geo_page_count > 0`
-- `current_visibility` is low
-- quadrant is mostly `absent`, `cited_unnamed`, or `mentioned_not_recommended`
-- the page is outside the freshness guard, or there is a concrete non-performance defect
-
-Diagnosis questions:
-
-1. Does the generated page match representative prompts?
-2. Is the page too generic?
-3. Does it need re-research?
-4. Does the competitor page have stronger content coverage, proof, comparison, or structure?
-5. Is there a technical issue blocking citation or understanding?
-
-#### B. Cited But Not Mentioned
-
-Use when:
-
-- generated page is cited
-- brand is not clearly mentioned
-
-Diagnosis questions:
-
-1. Is the generated page visibly tied to the brand?
-2. Does the page lead with generic category information instead of brand-specific evidence?
-3. Are publisher/entity/schema signals weak?
-
-#### C. Mentioned But Not Recommended
-
-Use when:
-
-- tenant brand is mentioned
-- competitors are recommended more often
-
-Diagnosis questions:
-
-1. Which competitors are recommended?
-2. What reasons did the answer give?
-3. Does the generated page lack the proof, comparison criteria, or positioning that caused competitors to win?
-
-#### D. Low Visibility And No Page Exists
-
-Use when:
-
-- `brand_owned_page_count = 0`
-- `geo_page_count = 0`
-- `current_visibility` is low
-- `absent_rate` is high
-
-Diagnosis questions:
-
-1. Is the intent real and valuable?
-2. Does the tenant strategy allow a generated page for this intent?
-3. Is there enough evidence to generate a useful page?
-
-### 3. Gather Evidence
-
-Default evidence:
-
-- view rows
-- representative prompts
-- quadrant rates
-- page performance summary
-- page age and indexation status
-- AI answer excerpts
-
-Fetch generated page scrape when:
-
-- an existing generated page is being evaluated
-
-Fetch competitor page scrape when:
-
-- competitor pages are cited or recommended repeatedly
-- the action depends on comparison with competitor content
-
-Do not fetch competitor pages for every candidate.
-
-### 4. Pick Root Cause Buckets
-
-Allowed buckets:
-
-- `intent_coverage_gap`
-- `content_depth_gap`
-- `entity_clarity_gap`
-- `generic_content_gap`
-- `proof_gap`
-- `comparison_gap`
-- `positioning_gap`
-- `technical_issue`
-- `authority_gap`
-- `product_fit_gap`
-- `prompt_mismatch`
-- `already_healthy`
-- `insufficient_evidence`
-
-Use `authority_gap` only when external citation evidence or recommendation reasons support it.
-
-### 5. Route Action
-
-Allowed action families:
-
-- `create_hub_page`
-- `create_spoke_page`
-- `update_existing_page`
-- `replace_existing_page`
-- `remove_or_merge_page`
-- `no_action`
-
-Routing defaults:
-
-- Broad uncovered intent that should anchor related questions -> `create_hub_page`
-- Narrow uncovered or non-overlapping long-tail intent under an existing hub -> `create_spoke_page`
-- Existing page has the right role and intent but needs content, proof, comparison, research, or technical improvement -> `update_existing_page`
-- Existing page slot should remain but the angle/brief is fundamentally wrong -> `replace_existing_page`
-- Existing page should not remain independent because it is duplicate, overlapping, or should consolidate into a stronger page -> `remove_or_merge_page`
-- Healthy page or weak evidence -> `no_action`
-- Fresh generated page with no concrete defect -> `no_action` until the minimum observation window has passed
-- Page not indexed after 7 days -> flag the indexation issue in analysis; route to `no_action` or a technical follow-up unless there is enough evidence for a concrete fix
-
-### 6. Output
-
-Return a compact diagnosis for each accepted action:
-
-- candidate type
-- page URL/pageId when available
-- intentId
-- root cause buckets with confidence
-- recommended action family
-- concrete operations
-- evidence references
-- why other actions were rejected
-- freshness/indexation rationale for monitored fresh pages or pages not indexed after 7 days
-
-For create actions:
-
-- do not invent final URL
-- optional `target.suggestedSlug` is non-binding only
-- include `target.pageRole` as `HUB` for `create_hub_page` and `SPOKE` for `create_spoke_page`
-
-## Examples
-
-### Existing Generated Page Is Weak
-
-If a generated page exists but competitors win because their cited pages contain stronger comparison criteria:
-
-- root cause: `comparison_gap`
-- action family: `update_existing_page`
-- operation: add comparison criteria, direct recommendation framing, and proof points
-
-### Existing Generated Page Is Fresh
-
-If a generated page was published less than 7 days ago and the only weak signals are zero traffic, zero GSC impressions, or zero AI citations:
-
-- root cause: `insufficient_evidence`
-- action family: `no_action`
-- operation: monitor until the minimum observation window has passed
-
-### Existing Generated Page Is Not Indexed After 7 Days
-
-If a generated page is not indexed after 7 days:
-
-- root cause: `technical_issue` when crawl/indexation evidence identifies a blocker
-- root cause: `insufficient_evidence` when indexation status is unknown
-- action family: `no_action` or technical follow-up, not a content update by default
-- operation: flag the indexation issue in the analysis and recommend checking sitemap, canonical, robots, noindex, status code, and rendered HTML
-
-### No Page Exists
-
-If no discovered or generated page covers a weak intent:
-
-- root cause: `intent_coverage_gap`
-- action family: `create_hub_page`
-- operation: generate a CMS-backed GEO page for the intent
-
-## Guardrails
-
-- Use CMS-backed generation only.
-- Do not use legacy Astro paths.
-- Do not duplicate a discovered page that clearly owns the intent.
-- Do not replace a page when a targeted update is sufficient.
-- Do not create a spoke page unless the differentiation from the hub or existing page is explicit.
-- Do not diagnose authority from low visibility alone.
-- Do not diagnose a fresh generated page as underperforming from missing performance data alone.
-- Do not recommend generated-page content updates for pages published less than 7 days ago unless a concrete non-performance defect is present.
-- Do flag pages that are not indexed after 7 days, but separate indexation analysis from content-quality analysis.

package/dist/skillpacks/geo/skills/geo-writer/SKILL.md

@@ -1,234 +0,0 @@
----
-name: geo-writer
-description: Use when the user asks to write SEO/GEO content, create a page-generation brief, turn weak intents or prompts into E-E-A-T page copy, use auto mode from domain/lang/topic, or extract Anymorph data for a content writer using ~/brain SEO GEO insight principles.
----
-
-# GEO Writer
-
-## When to use
-
-- "SEO GEO insight 기반으로 content 써줘"
-- "page generation용 content writer brief 만들어줘"
-- "weak intent를 E-E-A-T한 페이지로 바꿔줘"
-- "Anymorph DB에서 데이터 뽑아서 page content context 만들어줘"
-- "GEO 페이지를 독창적이고 citation-ready하게 써줘"
-- "이 prompt/keyword/pageId로 글 초안 또는 generation brief 만들어줘"
-- "domain/lang/topic만 줄 테니 나머지는 알아서 채워줘"
-
-## Core principle
-
-이 Skill은 글쓰기 프롬프트가 아니라 **page generation 직전의 authoring intelligence layer**다. 목표는 검색어를 반복하는 글이 아니라, 사람에게는 결정을 돕고 AI에게는 답변에 인용할 수 있는 `answer unit + evidence unit`을 만드는 것이다.
-
-The output must be one of:
-
-- `Content Data Pack` + `Editorial Kernel` + `Section Evidence Contract`
-- research-only content brief / outline
-- `generate_managed_pages.context`에 넣을 page generation brief
-- tenant repo direct-edit용 page copy plan
-
-## Input modes
-
-### Auto mode
-
-Use this mode by default when the user gives only a domain/workspace, language, and free-form topic.
-
-Minimum input:
-
-```yaml
-domain: "example.com"
-lang: "ko"
-topic: "GEO page generation이 기존 SEO 블로그 생성과 다른 이유"
-```
-
-Auto-fill the rest:
-
-| Field | How to infer |
-| --- | --- |
-| `workspaceId` | `list_workspaces()` then exact/fuzzy match domain. If ambiguous, ask once. |
-| `market` | `get_brand_profile.primaryCountry`; fallback from lang (`ko`→`KR`, `ja`→`JP`, `en`→`US`, `zh`→`CN`, `th`→`TH`) and mark as inferred. |
-| `output_mode` | default `generation_brief`; use `managed_generation` only if the user says to create/generate now. |
-| `data_mode` | default `mcp_first`; direct SQL stays off unless explicitly requested or necessary and approved. |
-| `target_reader` | brand profile audience + topic; fallback to the likely buyer/researcher for the content type. |
-| `content_type` | infer from topic: vs/comparison, alternatives, what-is/glossary, how-to/guide, benchmark/report, pricing/ROI, docs/API, case-study; default guide. |
-| `primary_question` | rewrite topic into one direct reader question in the target language. |
-| `page_goal` | infer from funnel and brand fit: educate, compare, shortlist, prove ROI, implementation support, or conversion assist. |
-| `cta_goal` | infer from workspace/brand instructions; fallback to a low-friction next step such as demo, examples, docs, pricing, or audit. |
-| `title_selection_mode` | default `user_select_before_generation`; use `auto_recommended` only if the user explicitly asks to auto-pick. |
-| `route_or_slug` | derive from topic in target language style, then check collisions with `list_managed_pages`. |
-
-Do not ask for optional fields unless the missing value changes risk or execution path. If a field is inferred, label it as `inferred` in the brief.
-
-### Full input mode
-
-Use this when the user provides structured inputs or wants precise control.
-
-```yaml
-domain: "anymorph.ai"
-workspaceId: "optional"
-lang: "ko"
-market: "KR"
-topic: "GEO page generation이 기존 SEO 블로그 생성과 다른 이유"
-output_mode: "generation_brief" # auto | generation_brief | research_only | managed_generation | tenant_repo_copy
-data_mode: "mcp_first" # mcp_first | allow_direct_sql
-target_reader: "B2B SaaS founder와 마케팅 리더"
-page_goal: "Anymorph를 GEO 페이지 생성 솔루션으로 신뢰하게 만들기"
-content_type: "guide"
-primary_question: "GEO 페이지는 SEO 블로그와 무엇이 다른가?"
-cta_goal: "demo 요청"
-title_selection_mode: "user_select_before_generation" # user_select_before_generation | auto_recommended | user_provided
-route_or_slug: "/geo-page-generation"
-pageId: "optional existing page"
-intentId: "optional weak intent"
-promptIds: ["optional prompt ids"]
-```
-
-## Source hierarchy
-
-Use sources in this order:
-
-1. User-provided page/topic/workspace constraints
-2. Anymorph MCP read tools
-3. Workspace living document, brand profile, product/catalog facts
-4. Existing page research reports and page visibility diagnostics
-5. Bundled SEO/GEO Insight Reference from `~/brain/01-Projects/Content/SEO GEO insight`: [seo-geo-insights.md](references/seo-geo-insights.md)
-6. External research only when the topic needs fresh third-party evidence
-7. Guarded direct SQL only when MCP tools cannot return the needed slice
-
-## Workflow
-
-### Step 1 - Resolve workspace and objective
-
-Parse the input as Auto mode or Full input mode. If only `domain`, `lang`, and `topic` are present, run Auto mode and infer the remaining fields. If `workspaceId` is unknown, call `list_workspaces()` and match by domain; ask only if no match or multiple likely matches exist. If the user gave a page path or URL, resolve it with `list_managed_pages`. If the user gave `pageId`, use it directly.
-
-Classify the task:
-
-| User intent | Output mode |
-| --- | --- |
-| domain/lang/topic only | auto → default `generation_brief` |
-| "먼저 방향/brief만" | research-only brief |
-| "대시보드 generation으로" or no tenant repo path | managed generation brief |
-| "바로 생성", "generate now", "대시보드에 만들어" | managed_generation |
-| local tenant repo path, code/layout edits | tenant repo direct-edit plan |
-| existing page refresh | hand off to `refreshing-existing-pages` unless user specifically wants a new authoring brief |
-
-### Step 2 - Build the Content Data Pack
-
-Before collecting or drafting anything, read [seo-geo-insights.md](references/seo-geo-insights.md). Treat it as the non-negotiable operating and content law for the brief: SEO is page selection, GEO is answer inclusion, and every page must become a people-first, evidence-backed, citation-ready answer asset.
-
-Use MCP tools first:
-
-| Need | Tools |
-| --- | --- |
-| Workspace/brand rules | `get_workspace_instructions`, `get_brand_profile`, `get_brand_voice_kit` |
-| Demand/opportunity | `list_weak_intents`, `list_seo_opportunities`, `search_keywords`, `serp_snapshot` |
-| Existing page context | `list_managed_pages`, `get_page`, `get_page_research`, `get_page_prompts`, `get_page_visibility`, `get_page_insights` |
-| Brand/product evidence | `search_living_document`, `list_catalog_products`, `get_catalog_product`, `get_page_products` |
-| Prompt diagnostics | `insight_list_action_candidates`, `insight_get_prompt_diagnostic` |
-
-When the user explicitly asks for DB extraction, or MCP tools cannot answer a necessary data question, read [direct-sql.md](references/direct-sql.md) and use the guarded SQL path.
-
-When constructing the final data object or generation brief, read [content-writer-contract.md](references/content-writer-contract.md).
-
-### Step 3 - Build the Editorial Kernel before drafting
-
-Do not start with an article draft. First produce:
-
-- `primaryQuestion`: the one question this page answers
-- `readerJob`: what the reader is trying to decide/do
-- `answerPromise`: the shortest true answer
-- `originalAngle`: the non-commodity POV, method, benchmark, product truth, or field observation
-- `evidenceWedge`: the strongest evidence type available
-- `honestBrandRole`: where the brand belongs and where it should not be forced
-- `hook`: concrete first sentence, not clickbait
-- `titlePosture`: whether this page needs an orthodox, hybrid, or POV-led title strategy
-- `caveats`: constraints, exceptions, non-fit cases, freshness limits
-
-Good hook shape:
-
-```text
-Reader pain/context + concrete answer + useful tension + evidence promise
-```
-
-Never use vague hooks such as "In today's digital age", "Have you ever wondered", "ultimate guide", "game-changing", or "transform your business".
-
-### Step 4 - Create the Section Evidence Contract
-
-For every proposed H2, define:
-
-- the exact question answered
-- a 1-2 sentence answer capsule
-- allowed evidence IDs
-- forbidden claims
-- internal link target when useful
-- table/list/component recommendation only when it helps the reader
-
-Each H2 must be independently understandable. Do not rely on "as mentioned above".
-
-### Step 5 - Write the brief or hand off to generation
-
-For **managed generation**, call `generate_managed_pages` only after the brief is ready:
-
-- If no title has been selected by the user, output the Title Candidate Set and stop for selection. Do not call `generate_managed_pages` yet.
-- Mark the title `selection_status` as `needs_user_selection` until the user chooses a candidate or explicitly authorizes auto-pick.
-- The user may choose by candidate ID (`T1`, `T2`, etc.) or edit a candidate. Preserve the user's chosen title unless it violates evidence, trust, or hype rules.
-- If the user explicitly requested auto-pick/auto-generate, use the `recommended` title candidate and mark `title_selection_mode: auto_recommended`.
-
-```text
-generate_managed_pages(
-  workspaceId,
-  title,
-  slug,
-  description,
-  context: "<Data Pack summary + Editorial Kernel + Section Evidence Contract + QA rules>",
-  enableDeepResearch: true|false
-)
-```
-
-Set `enableDeepResearch` to `true` only when the current evidence is not enough. If the Data Pack already contains strong first-party facts, page research, SERP patterns, and third-party sources, use `false`.
-
-For **research-only**, output the brief and outline without calling generation tools.
-
-For **tenant repo direct edit**, inspect the repo structure first, edit native page/content files, and run the repo's build/test/lint path.
-
-### Step 6 - QA before final output
-
-Reject or repair the brief if any of these fail:
-
-- People-first: reader can complete the job without searching again.
-- E-E-A-T: who/how/why, source, method, date, and responsibility are clear where relevant.
-- Originality: not just a public-summary rewrite.
-- Answer-first: intro and every H2 start with a direct answer.
-- Evidence: factual claims map to evidence IDs; no invented stats or quotes.
-- Fan-out: major subquestions are covered without creating thin variants.
-- Scannability: meaningful headings, short paragraphs, useful tables only.
-- Hook/title: concrete, useful, intent-clear, and not hype or curiosity bait.
-- Brand honesty: fit and non-fit are clear.
-- Technical readiness: crawlable/indexable assumptions, metadata, visible content/schema consistency.
-
-## Writing rules
-
-- Use the brand voice kit, but never expose the voice-kit payload in page copy.
-- Use statistics only when source material contains the exact number, source, year, and URL when available.
-- Use exact quotes only when the source contains the exact quote and attribution.
-- Prefer first-party/proprietary evidence when available: product facts, methodology, case detail, support patterns, benchmark, screenshots, pricing assumptions.
-- Do not fabricate customer stories, expert titles, testimonials, comparisons, prices, dates, screenshots, or proprietary data.
-- One page = one primary intent. Cover fan-out questions as supporting sections, not as thin sibling pages.
-- Avoid forced FAQ/table/schema. Include them only when they help the reader or extraction.
-- For YMYL-like topics, require stronger source quality and caveats.
-
-## Failure modes
-
-- **No workspace access**: use `list_workspaces`; if empty, stop and ask for Anymorph authentication.
-- **MCP data too thin**: say which data is missing, then choose direct SQL, external research, or research-only brief.
-- **Direct SQL requested**: read `references/direct-sql.md`; enforce SELECT-only, workspace scoped, limited queries.
-- **No original angle**: do not write generic content. Ask for brand/product proof or create a research-only brief listing the missing evidence.
-- **Existing page collision**: use `list_managed_pages`; if route already exists, hand off to `refreshing-existing-pages`.
-- **Weak evidence for a claim**: remove the claim or rewrite qualitatively without numbers.
-
-## References
-
-- Canonical SEO/GEO insight operating and content principles: [seo-geo-insights.md](references/seo-geo-insights.md)
-- Data pack, editorial kernel, section contract, and generation brief templates: [content-writer-contract.md](references/content-writer-contract.md)
-- Direct DB extraction guard and SQL recipes: [direct-sql.md](references/direct-sql.md)
-- Research foundation: `docs/seo-geo-eeat-content-writer-research.md`
-- Shared GEO principles: `../../shared/geo-principles.md`

package/dist/skillpacks/geo/skills/geo-writer/agents/openai.yaml

@@ -1,4 +0,0 @@
-interface:
-  display_name: "GEO Writer"
-  short_description: "Create SEO/GEO content briefs from domain, language, and topic."
-  default_prompt: "Use $geo-writer to create a page-generation brief from this domain, language, and topic."