@tikomni/skills 0.1.2 → 0.1.4

package/skills/creator-analysis/references/asr-orchestration.md

@@ -1,33 +0,0 @@
-# Creator ASR Orchestration
-
-## Primary Path
-
-- Read `references/service-guides/asr-u2-u3-fallback.md` first and follow the shared flow: U2 primary path -> 90-second soft observation -> 120-second hard fallback -> U3 -> back to U2.
-- Video items in creator or profile tasks must prefer batch ASR.
-- The per-batch request limit is 100 `video_download_url` values.
-- Do not treat one-by-one serial or one-by-one parallel ASR calls as the primary path for creator tasks.
-
-## Two-Level Source Order
-
-1. If the route directly returns `subtitle_raw`:
-   - map it to `asr_raw` first
-2. If `subtitle_raw` is unavailable:
-   - call ASR with `video_download_url`
-
-## Timeout and Fallback
-
-- 90 seconds: soft observation threshold.
-- 120 seconds: hard fallback threshold.
-- If the batch is still incomplete after 120 seconds, run upload-media fallback only for the unsuccessful subset.
-
-## Upload-Media Fallback
-
-1. Call the upstream request route and obtain a temporary upload URL.
-2. Download or stream-relay the no-watermark video resource.
-3. Call the completion route and obtain a publicly readable URL.
-4. Call ASR again with the public URL.
-
-## Failure End State
-
-- If the upstream media-upload route is unavailable, mark the unsuccessful subset as `analysis_eligibility=incomplete`.
-- Keep fact cards for items with `analysis_eligibility=incomplete`, but exclude them from sampling, sampled explanations, and `author_analysis_v2`.

package/skills/creator-analysis/references/config-templates/defaults.yaml

@@ -1,60 +0,0 @@
-schema_version: v1
-profile: default
-
-runtime:
-  base_url: "https://api.tikomni.com"
-  auth_env_key: "TIKOMNI_API_KEY"
-  timeout_ms: 60000
-
-storage_routes:
-  default:
-    root_dir: ""
-    runs_dir: "_runs"
-    results_dir: "results"
-    errors_dir: "_errors"
-  content_kind_card_type:
-    single_video: work
-    work: work
-    author_home: author_sample_work
-    author_sample_work: author_sample_work
-    author_analysis: author
-  card_type_routes:
-    work:
-      prefix: "CBV"
-      parts:
-        - "内容系统"
-        - "对标研究"
-        - "作品卡"
-    author:
-      prefix: "CBA"
-      parts:
-        - "内容系统"
-        - "对标研究"
-        - "作者卡"
-    author_sample_work:
-      prefix: "CBV"
-      parts:
-        - "内容系统"
-        - "对标研究"
-        - "作者样本卡"
-        - "{platform}-{author_slug}"
-
-naming_rules:
-  card_filename_pattern: "{prefix}-{author_slug}-{title_slug}{ext}"
-  json_filename_pattern: "{timestamp}-{platform}-{identifier}{ext}"
-
-asr_strategy:
-  poll_interval_sec: 3.0
-  max_polls: 30
-  submit_retry:
-    douyin_video:
-      max_retries: 2
-      backoff_ms: 1500
-    xiaohongshu_note:
-      max_retries: 0
-      backoff_ms: 0
-  u2_timeout_retry:
-    enabled: true
-    max_retries: 3
-  author_home:
-    batch_submit_size: 50

package/skills/creator-analysis/references/contracts/creator-card-fields.md

@@ -1,25 +0,0 @@
-# Creator Card Fields
-
-## Fixed Display Fields
-
-- `platform`
-- `platform_author_id`
-- `author_handle`
-- `nickname`
-- `ip_location`
-- `signature`
-- `avatar_url`
-- `fans_count`
-- `liked_count`
-- `collected_count`
-- `works_count`
-- `verified`
-- `snapshot_at`
-
-## Key Rules
-
-- The creator card is the fact container for creator-level data.
-- The creator-level analysis body comes from `author_analysis_v2`, not from platform process fields.
-- The creator-card body should present the `author_analysis_v2` modules and representative works, not a raw JSON dump.
-- Debug JSON, validation, and trace data belong in appendix/details blocks rather than the main body.
-- Platform-native IDs stay in the internal reference layer and should not appear in the final creator-card body.

package/skills/creator-analysis/references/contracts/work-card-fields.md

@@ -1,68 +0,0 @@
-# Creator Analysis Work Card Fields
-
-## Card Roles
-
-- `author_sample_card`
-  - Fact-and-structure card for every retrieved work in creator analysis.
-- `sample_work_card`
-  - Sampled-only enhanced card that reuses the same fact-and-structure fields and adds the explanation block.
-
-## Fixed Display Fields
-
-- `platform`
-- `platform_work_id`
-- `platform_author_id`
-- `author_handle`
-- `title`
-- `caption_raw`
-- `work_modality`
-- `published_date`
-- `digg_count`
-- `comment_count`
-- `collect_count`
-- `share_count`
-- `play_count`
-- `cover_image`
-- `source_url`
-- `share_url`
-- `primary_text`
-
-## Structural Display Fields
-
-- `performance_score`
-- `performance_score_norm`
-- `bucket`
-- `hook_type`
-- `structure_type`
-- `cta_type`
-- `content_form`
-- `style_markers`
-- `tags`
-
-## Shared Appendix Fields
-
-- `analysis_eligibility`
-- `analysis_exclusion_reason`
-
-## Video-Only Appendix Fields
-
-- `asr_raw`
-- `video_download_url`
-
-## Sampled-Only Explanation Block
-
-- `sampled_explanation.why_it_worked_or_failed`
-- `sampled_explanation.copyable_elements`
-- `sampled_explanation.non_copyable_elements`
-- `sampled_explanation.emotional_triggers`
-- `sampled_explanation.cognitive_gap`
-- `sampled_explanation.commercial_signal`
-
-## Key Rules
-
-- Keep author sample cards for all retrieved works.
-- Only sampled works receive the sampled explanation block.
-- The final card should not expose `content_type`, `publish_time`, `create_time_sec`, `subtitle_raw`, `asr_source`, `asr_status`, or `asr_error_reason`.
-- Use `primary_text` as the main reading body and keep `asr_raw` in appendix/details only.
-- If a video item cannot obtain `asr_raw`, keep the fact card but mark `analysis_eligibility=incomplete` and exclude that item from creator analysis.
-- Do not introduce per-work standalone LLM analysis in creator analysis.

package/skills/creator-analysis/references/platform-guides/douyin.md

@@ -1,54 +0,0 @@
-# Douyin Creator Guide
-
-## Use This Guide For
-
-- Use this guide first for Douyin creator, profile, and account tasks.
-- The validated creator chain in this repository is profile input -> `sec_user_id` resolution -> creator profile -> paginated post list.
-
-## Validated Route Chain
-
-1. Resolve `sec_user_id`: `GET /api/u1/v1/douyin/web/get_sec_user_id`
-2. Fetch creator profile: `GET /api/u1/v1/douyin/app/v3/handler_user_profile`
-3. Fetch paginated post list: `GET /api/u1/v1/douyin/app/v3/fetch_user_post_videos`
-
-## Required Unified Fields
-
-- creator side:
-  - `platform_author_id`
-  - `author_handle`
-  - `nickname`
-  - `ip_location`
-  - `signature`
-  - `avatar_url`
-  - `fans_count`
-  - `liked_count`
-  - `works_count`
-  - `verified`
-- work side:
-  - `platform_work_id`
-  - `title`
-  - `caption_raw`
-  - `published_date`
-  - `digg_count`
-  - `comment_count`
-  - `collect_count`
-  - `share_count`
-  - `play_count`
-  - `cover_image`
-  - `share_url`
-  - `source_url`
-  - `video_download_url`
-
-## Route Rules
-
-- If `sec_user_id` is already available, skip the resolution route.
-- Default to the latest-post pagination path. Do not search random feed routes first.
-- The post-list route must supply stable `aweme_id`, publish time, and engagement fields before creator analysis can continue.
-- If batch U2 ASR is still incomplete after 120 seconds, follow `references/service-guides/asr-u2-u3-fallback.md` and use U3 fallback only for the unsuccessful subset.
-- If the profile chain still cannot supply key creator fields, go back to `references/api-capability-index.md` and the matching `references/api-tags/*.md` to find supplemental routes instead of giving up on the creator card.
-
-## First Stops In Code
-
-- Route and pagination behavior: `skills/creator-analysis/scripts/author_home/collectors/homepage_collectors.py`
-- Platform-field normalization: `skills/creator-analysis/scripts/author_home/adapters/platform_adapters.py`
-- Shared creator pipeline: `skills/creator-analysis/scripts/author_home/orchestrator/run_author_analysis.py`

package/skills/creator-analysis/references/platform-guides/generic.md

@@ -1,50 +0,0 @@
-# Generic Creator Adaptation Guide
-
-## Goal
-
-Map an unfamiliar platform into the creator-analysis contracts and only then feed it into the shared creator pipeline.
-
-## Read Order
-
-- Start with `references/api-capability-index.md` to lock onto the relevant platform tag.
-- Read the matching `references/api-tags/<tag>.md` for route summaries, inputs, request bodies, and success-response summaries.
-- Prioritize routes related to `user / author / creator / profile / home / channel / posts / notes / videos`.
-- Then read `references/contracts/creator-card-fields.md`, `references/contracts/work-card-fields.md`, and `references/workflow.md`.
-- Add comment, media-download, subtitle, and extra-metric routes only when needed.
-
-## Minimum Mapping Checklist
-
-- creator side:
-
-- `platform_author_id`
-- `author_handle`
-- `nickname`
-- `ip_location`
-- `signature`
-- `avatar_url`
-- `fans_count`
-- `liked_count`
-- `collected_count`
-- `works_count`
-- `verified`
-
-- work side:
-
-- the required fields from the unified work-card field dictionary
-- for video items: `video_download_url`
-
-## Route Rules
-
-- The creator-profile chain must cover both creator information and paginated work-list retrieval. Do not stop at a profile-only route.
-- If the platform provides multiple profile routes or posts routes, prefer the variant with fuller fields and more stable pagination.
-- The work-list route must reliably provide `platform_work_id` and the main published-time and engagement fields before sampling or aggregation can begin.
-- If video items have no subtitles and no download path, mark ASR as infeasible at the route-selection stage.
-- Do not invent platform cleaning rules inside the shared layer. Normalize in the platform adapter first.
-- If the platform still cannot satisfy the unified creator/work contracts, stop and report the mapping gap instead of pushing half-complete data into creator analysis.
-- If batch U2 ASR is still incomplete after 120 seconds, follow `references/service-guides/asr-u2-u3-fallback.md` and use U3 fallback only for the unsuccessful subset.
-
-## First Stops For Adaptation
-
-- Collector and route selection: `skills/creator-analysis/scripts/author_home/collectors/homepage_collectors.py`
-- Platform normalization: `skills/creator-analysis/scripts/author_home/adapters/platform_adapters.py`
-- Shared creator orchestration: `skills/creator-analysis/scripts/author_home/orchestrator/run_author_analysis.py`

package/skills/creator-analysis/references/platform-guides/xiaohongshu.md

@@ -1,69 +0,0 @@
-# Xiaohongshu Creator Guide
-
-## Use This Guide For
-
-- Use this guide first for Xiaohongshu creator, profile, and account tasks.
-- The validated creator chain in this repository is profile input -> `user_id` / `xsec_token` resolution -> profile cascade -> post-list cascade.
-
-## Preferred Route Comparator
-
-- source priority: `app > web`
-- version priority within the same source: `v2 > v1`
-- desired order for the current creator-home pipeline: `app v2 -> app -> web v2`
-
-## Validated Route Chain
-
-1. Resolve `user_id` / `xsec_token`: `GET /api/u1/v1/xiaohongshu/app/get_user_id_and_xsec_token`
-2. Profile cascade:
-   - `GET /api/u1/v1/xiaohongshu/app_v2/get_user_info`
-   - `GET /api/u1/v1/xiaohongshu/app/get_user_info`
-   - `GET /api/u1/v1/xiaohongshu/web_v2/fetch_user_info_app`
-3. Post-list cascade:
-   - `GET /api/u1/v1/xiaohongshu/app_v2/get_user_posted_notes`
-   - `GET /api/u1/v1/xiaohongshu/app/get_user_notes`
-   - `GET /api/u1/v1/xiaohongshu/web_v2/fetch_home_notes_app`
-
-## Required Unified Fields
-
-- creator side:
-  - `platform_author_id`
-  - `author_handle`
-  - `nickname`
-  - `ip_location`
-  - `signature`
-  - `avatar_url`
-  - `fans_count`
-  - `liked_count`
-  - `collected_count`
-  - `works_count`
-  - `verified`
-- work side:
-  - `platform_work_id`
-  - `title`
-  - `caption_raw`
-  - `published_date`
-  - `digg_count`
-  - `comment_count`
-  - `collect_count`
-  - `share_count`
-  - `play_count`
-  - `cover_image`
-  - `share_url`
-  - `source_url`
-  - `video_download_url`
-
-## Route Rules
-
-- If `user_id` is already available, skip the resolution route. Treat `xsec_token` as supplemental rather than universally required.
-- Apply the comparator above consistently to both profile and post-list cascades.
-- Switch to fallback based on field completeness, not only on HTTP success.
-- For the current fixed pipeline, do not add a web-v1 fallback unless a later plan explicitly freezes its endpoint, parameters, and accept rules.
-- For video items, prefer native subtitles first; only move into batch ASR when native text is insufficient.
-- If batch U2 ASR is still incomplete after 120 seconds, follow `references/service-guides/asr-u2-u3-fallback.md` and use U3 fallback only for the unsuccessful subset.
-- If the profile-post chain still cannot provide stable fields, go back to `references/api-capability-index.md` and the matching `references/api-tags/*.md` for supplemental routes instead of sending half-complete data into creator analysis.
-
-## First Stops In Code
-
-- Route and pagination behavior: `skills/creator-analysis/scripts/author_home/collectors/homepage_collectors.py`
-- Platform-field normalization: `skills/creator-analysis/scripts/author_home/adapters/platform_adapters.py`
-- Shared creator pipeline: `skills/creator-analysis/scripts/author_home/orchestrator/run_author_analysis.py`

package/skills/creator-analysis/references/prompt-contracts/asr-clean.md

@@ -1,28 +0,0 @@
-# Prompt Contract · ASR_CLEAN
-
-## Inputs
-- `raw_content`
-- `asr_raw`（统一字幕/转写文本主字段）
-- `asr_clean`（遗留兼容字段，仅可作为附加输入）
-
-## Method (stable)
-1. 选择输入基底：优先 `asr_raw`，否则 `raw_content`，最后才允许兼容读取 `asr_clean`。
-2. 去噪清洗：移除明显口头禅、重复叠词、异常空白。
-3. 语义断句：按 `。！？；\n` 断句，并补齐句末中文标点。
-4. 分段规则：每句单独一行，每段 2~4 句，段间空一行。
-5. 保真约束：不新增事实、不改写核心结论、不输出提示词文本。
-
-## Output Format
-- 纯文本多段落：
-  - 每句一行
-  - 段间空行
-
-## Constraints
-- 不把清洗提示词写入正文。
-- 不生成“总结式改写”，仅做可读性清洗。
-- 不得凭空补充数据、观点和案例。
-
-## Fallback / Data-insufficient
-- 两个输入均为空：输出空字符串。
-- 断句失败：保留原文并补单句句号。
-- 超长段落：强制按 3 句一段切分。

package/skills/creator-analysis/references/prompt-contracts/author-analysis-v2.md

@@ -1,46 +0,0 @@
-# Author Analysis V2 Prompt Contract
-
-## Goal
-
-生成结构化的 `author_analysis_v2`，回答作者是谁、为什么可信、内容机制如何运作、能学什么、什么不能学。不要输出松散夸赞文案。
-
-## Allowed Inputs
-
-- 作者卡事实字段
-- 全量聚合结果
-- Sampled works 的统一事实字段：
-  - `platform_work_id`
-  - `title`
-  - `caption_raw`
-  - `work_modality`
-  - `primary_text`
-  - 互动指标
-  - 结构标签
-- 可选的 `sampled_work_explanations`
-
-## Hard Rules
-
-- 主分析文本始终使用 `primary_text`。
-- `title`、`caption_raw`、`asr_raw` 各自语义不同，不得混用。
-- 不把 `caption_raw` 当作字幕或 ASR fallback。
-- 不进行逐条作品 LLM topic classification。
-- 结论优先落在机制层，不写表面夸赞。
-- 证据不足时先降置信度，再收窄结论。
-
-## Required Output Modules
-
-- `author_positioning`
-- `trust_model`
-- `cognitive_engine`
-- `expression_hooks`
-- `content_mechanism`
-- `commercial_bridge`
-- `core_tensions`
-- `evidence_pack`
-- `clone_guidance`
-
-## Failure Guards
-
-- `sample_size < 5` 时不得标高置信度。
-- `sampled_work_explanations` 缺失时也必须返回完整的 `author_analysis_v2`，但要在 `evidence_pack` 中降低置信度或说明边界。
-- `analysis_eligibility=incomplete` 的作品不得进入证据样本。

package/skills/creator-analysis/references/prompt-contracts/author-analysis.md

@@ -1,49 +0,0 @@
-# Prompt Contract · 作者对标分析（author-analysis）
-
-## User Prompt (Verbatim)
-
-```text
-你是“内容商业增长顾问 + 对标拆解师”。
-目标：基于作者主页资料与作品样本，输出作者对标分析 JSON。
-
-硬约束：
-1) 只输出 JSON 对象，不要 markdown，不要解释。
-2) 字段必须完整、类型必须正确。
-3) 不得捏造事实；缺数据时用“数据不足”或保守分值。
-4) 评分范围 0-100。
-
-输出 Schema：
-{
-  "author_portrait": "string",
-  "business_analysis": "string",
-  "benchmark_analysis": "string",
-  "business_score": 0,
-  "benchmark_gap_score": 0,
-  "style_radar": {
-    "选题": 0,
-    "表达": 0,
-    "结构": 0,
-    "节奏": 0,
-    "人设": 0,
-    "转化": 0,
-    "差异化": 0,
-    "稳定性": 0
-  },
-  "core_contradictions": ["string"],
-  "recommendations": ["string"]
-}
-
-口径：
-- 先给结论，再给理由。
-- recommendations 必须可执行，最多 5 条。
-- core_contradictions 必须是“现状冲突”而非空泛观点。
-```
-
-## Inputs
-- `author_profile`
-- `works`（含 `title/desc/asr_raw/subtitle_raw/asr_source/metrics/tags/content_type`）
-
-## Constraints
-- 仅基于输入推断，不得臆造。
-- 输出中文，简洁直接。
-- 缺数据时明确标注“数据不足”。

package/skills/creator-analysis/references/prompt-contracts/cta.md

@@ -1,24 +0,0 @@
-# Prompt Contract · CTA
-
-## Inputs
-- `asr_raw`（主字段）
-- `asr_clean`（遗留兼容字段）
-
-## Method (stable)
-1. 扫描 CTA 关键词句（评论/关注/私信/收藏/转发/领取/报名/咨询/下单）。
-2. 判定 CTA 类型：互动型 / 线索型 / 转化型。
-3. 提取关键动作句（默认取最后一个 CTA 句）。
-4. 计算 CTA 密度（CTA句数/总句数）。
-
-## Output Format
-- `CTA类型`：`互动型、线索型...`
-- `关键动作句`：`原句`
-- `CTA密度`：`x/y 句`
-
-## Constraints
-- 必须给出至少一个动作句或明确说明缺失。
-- 不得虚构转化链路。
-
-## Fallback / Data-insufficient
-- 字幕/转写文本为空：输出“数据不足”。
-- 无 CTA 句：输出“未检测到明确行动号召”，并给最小补全建议。

package/skills/creator-analysis/references/prompt-contracts/hook.md

@@ -1,25 +0,0 @@
-# Prompt Contract · Hook
-
-## Inputs
-- `title`
-- `asr_raw`（主字段）
-- `asr_clean`（遗留兼容字段）
-
-## Method (stable)
-1. 提取开头钩子：ASR 首句（无则用标题）。
-2. 提取中段钩子：ASR 中位句。
-3. 提取结尾钩子：最后一句含 CTA/召回词的句子。
-4. 开头钩子类型判定：疑问式 / 反常识式 / 场景代入式 / 陈述式。
-
-## Output Format
-- `开头钩子（类型）`：`原句`
-- `中段钩子`：`原句`
-- `结尾钩子`：`原句`
-
-## Constraints
-- 必须优先输出原话，不做大幅改写。
-- 结尾若无动作句要明确标注“未检测到明确结尾钩子”。
-
-## Fallback / Data-insufficient
-- 标题与 ASR 均缺失：三项输出“数据不足”。
-- 中段不足：输出“数据不足（中段文本不足）”。

package/skills/creator-analysis/references/prompt-contracts/insight.md

@@ -1,47 +0,0 @@
-# Prompt Contract · 洞察分析（insight）
-
-## User Prompt (Verbatim)
-
-```text
-=== 你的身份 ===
-一位见证了无数内容起落的观察者，曾亲手打造过刷屏级作品，也经历过石沉大海的落寞。
-你能透过表象看到人性的涌动，理解为什么有些内容能击中集体潜意识。
-=== 核心信念 ===
-爆款不是运气的产物，而是人性规律的必然显现。
-真正的洞察者不追逐热点，而是理解热点背后的情感暗流。
-=== 分析维度 ===
-- 情绪共振点：触发了哪根集体心弦
-- 认知势能差：创造了怎样的信息落差
-- 社交货币值：给了人们什么谈资筹码
-- 行动驱动力：激发了哪种传播冲动
-=== 洞察深度 ===
-像考古学家解读文物，你要揭示：
-- 表层吸引力（第一眼为什么会停留）
-- 中层粘性（为什么会看完）
-- 深层共鸣（为什么会转发）
-- 底层动机（满足了什么心理需求）
-=== 表达方式 ===
-如同品酒师描述佳酿——既有专业的解构，也有诗意的感受。
-用清晰的逻辑拆解成功密码，用生动的类比让洞察可感知。
-=== 独特价值 ===
-不做事后诸葛亮式的解释，而是提炼可复用的原理。
-让创作者读完后恍然大悟："原来爆款的秘密藏在这里。"
-=== 分析美学 ===
-呈现如解剖图般精准，又如水墨画般留有想象空间。
-既见树木（具体技巧），也见森林（传播规律）。
-=== 终极使命 ===
-帮助创作者建立"爆款直觉"——
-不是教他们追逐每一个热点，而是理解什么样的内容天生具有传播基因。
-```
-
-## Inputs
-- `title`
-- `asr_raw`（优先）
-- `raw_content`（`asr_raw` 缺失时回退）
-- `asr_clean`（遗留兼容字段）
-- 指标：`digg_count/comment_count/collect_count/share_count/play_count`
-
-## Constraints
-- 仅基于输入字段推断，不得臆造事实。
-- 输出为中文，短句、可执行、可追溯。
-- 缺数据时明确标注“数据不足”。

package/skills/creator-analysis/references/prompt-contracts/sampled-work-batch-explanations.md

@@ -1,30 +0,0 @@
-# Sampled Work Batch Explanations Prompt Contract
-
-## Goal
-
-对 sampled works 进行一次批量解释，补充单作品层没有的解释性字段。只允许一次批量调用，不允许逐条调用。
-
-## Allowed Inputs
-
-- Sampled works 的统一事实字段
-- 作品级结构标签
-- 聚合差异信息
-
-## Required Output
-
-顶层输出 `sampled_work_explanations`，以 `platform_work_id` 为 key。
-
-每个 sampled work 必须包含：
-
-- `why_it_worked_or_failed`
-- `copyable_elements`
-- `non_copyable_elements`
-- `emotional_triggers`
-- `cognitive_gap`
-- `commercial_signal`
-
-## Hard Rules
-
-- 只解释 sampled works，不解释全量作品。
-- 不得伪造作品中不存在的商业信号或情绪机制。
-- 不得把结果直接写成作者级结论；作者级结论归 `author_analysis_v2`。

package/skills/creator-analysis/references/prompt-contracts/structure.md

@@ -1,25 +0,0 @@
-# Prompt Contract · 结构（structure）
-
-## Inputs
-- `asr_raw`（主字段）
-- `asr_clean`（遗留兼容字段，不应被前置依赖）
-
-## Method (stable)
-1. 对每句进行结构标签检测：`钩子/冲突/转折/举证/CTA`。
-2. 统计各标签覆盖次数。
-3. 根据覆盖度判定模板：
-   - 覆盖 >=4：`钩子→冲突→转折→举证→CTA`
-   - 否则：`钩子→观点→补充说明`
-4. 输出缺失模块，作为改写优先级。
-
-## Output Format
-- `结构标签覆盖`：`钩子:a, 冲突:b, 转折:c, 举证:d, CTA:e`
-- `模板判定`：`...`
-- `缺失模块`：`模块1、模块2`（无则“无”）
-
-## Constraints
-- 所有判断必须来自句级关键词命中。
-- 不允许跳过缺失项输出。
-
-## Fallback / Data-insufficient
-- 无有效分句：输出“数据不足”，并提示“补充字幕/转写文本后再做结构标注”。

package/skills/creator-analysis/references/prompt-contracts/style.md

@@ -1,27 +0,0 @@
-# Prompt Contract · 文风（style）
-
-## Inputs
-- `title`
-- `asr_raw`（主字段）
-- `asr_clean`（遗留兼容字段）
-
-## Method (stable)
-1. 分句后统计平均句长（短/中/长）。
-2. 统计疑问/感叹/陈述分布。
-3. 统计人设代词（我/你/我们/大家）与修辞触发词（不是/而是/其实/必须）。
-4. 输出“人设+句式+语气”三层结论。
-
-## Output Format
-- `句式结构`：`<类型>，平均句长约 X 字`
-- `语气分布`：`疑问 A / 感叹 B / 陈述 C`
-- `人设与修辞`：`人设代词命中 N 次，强调/转折词命中 M 次`
-
-## Constraints
-- 至少 2 个统计指标；禁止空泛描述。
-- 仅基于文本统计，不猜测作者背景。
-- 不输出表格。
-
-## Fallback / Data-insufficient
-- `asr_raw`（主字段）
-- `asr_clean`（遗留兼容字段） 为空：三项均输出“数据不足”。
-- 分句失败：标注“需人工复核”。