Enforces a dbt + Redshift model **writing-style & structure** contract — CTE roles, the zero-logic `final` CTE, naming, the two-block YAML header, column comments/tags, and Redshift syntax. Use when authoring, editing, or reviewing a dbt model (`.sql`), or when asked whether one 「符合規範嗎」. Adoptable template: project-specific items are tagged `(adapt)`; model comments & frontmatter values stay in the user's working language. Do NOT use for calculation logic, business rules, metric formulas, or layer-dependency design — style & structure ONLY. dbt 撰寫風格・CTE 結構・命名・註解・排版。dbt スタイル・命名規則。
Behavioral black-box dogfood of a skill-IN-DEVELOPMENT — a raw SKILL.md in the working tree that is NOT yet installed. Use when the user wants to gut-check how a drafted/edited skill actually behaves before trusting it: does it FIRE when it should and NOT over-fire, and does its workflow produce output that meets its own declared contract on real input. A fresh blind subagent that does NOT know the author's intent probes the triggers and the workflow and reports what breaks with reproducible transcript evidence. Tests BOTH dimensions co-equally: triggering (trigger-miss / over-trigger) AND output quality (workflow drift / gate bypass / valid-but-wrong), on working-tree files — no install, emitting a fix-actionable report. Triggers — zh-TW:「dogfood 這個 skill」「測試我的 skill 會不會觸發」; ja:「スキルをドッグフード」「発火するか試す」; en: "dogfood this skill", "behavioral / blind-test my skill", "will this skill fire before I ship". Do NOT use for: static design scoring of a SKILL.md (use dev-workflow:skill-judge — 8-dimension rubric, reads t
Deeply understand ONE large document or book — build a structured understanding (sections, claims, methodology, caveats, argument-structure) of a single source, depth-on-one-source vs deep-research's breadth-across-many. Use when the user wants to thoroughly comprehend one long document, paper, or book, run inside any coding agent host using the host's own tools (zero API-key setup).
把散在 Gmail / Slack / Notion / Asana / Drive / Calendar / GitHub 的事整理成一份可信、每列可點處理的晨報,外加一張零省略行動表;每天累積、跨日延續(已結/仍在等你/新發生)。 每天開工、想知道今天有什麼要處理、怕漏掉待回覆時用——before starting your workday。 觸發詞:每日簡報、daily brief、morning brief、晨報、今天要做什麼、待回覆、要回的訊息、standup、今日まとめ、朝のブリーフィング。 Do NOT use for 績效回顧/自評/專案盤點(那是 performance-evidence-audit,同機制反方向時間軸);Do NOT use 來寫回官方系統、代送或自動回覆(本 skill 唯讀 + 只寫本機草稿)。
Audit an existing document's cited claims — fetch each cited source and check it actually supports the claim; flag unsupported / misattributed / dead-link citations. Use when the user wants to verify that a document's citations hold up, run inside any coding agent host using the host's own LLM + web tools (zero API-key setup).
Lightweight single-claim adversarial verdict — supported / refuted / inconclusive with cited evidence. Use when one factual claim needs checking mid-conversation — the host agent gathers a little evidence and runs the same adversarial quorum as deep-research, returning a verdict (not a report), using the host's own WebSearch/WebFetch + LLM with zero API-key setup.
First-time setup for dbt-wiki: scaffold .dbt-wiki/ knowledge base from target/manifest.json (model / source / macro / seed / snapshot / test / exposure metadata, ref/source dependencies, schema.yml columns and tests), plus target/compiled/<project>/**/*.sql parsed via sqlglot for column-level lineage, plus dbt/models/**/*.sql raw files parsed via regex for inline SQL/jinja comments. Generates one markdown page per resource, plus index.md (grouped by tier / materialization / tag / group), lineage.md (ASCII DAG + adjacency list), log.md, SCHEMA.md, and an idempotent CLAUDE.md drop-in. Re-runnable: refreshes manifest-derived fields, archives orphans, preserves user-owned body sections. Pre-condition: dbt parse && dbt compile must be run first (init checks for target/manifest.json and target/compiled/), and sqlglot must be installed (pip install sqlglot). Triggers on "init dbt-wiki", "set up dbt-wiki", "scaffold dbt knowledge base", "seed dbt model wiki", "build dbt-wiki from manifest", "first-time dbt knowledge"
Package the dbt-wiki knowledge base into a portable, warehouse-tool-agnostic analytics skill. Run by the PROJECT OWNER (someone who has a distilled `.dbt-wiki/`) to freeze the curated semantic knowledge into a self-contained Agent Skill folder `<project>-analytics/` (SKILL.md + knowledge/ + references/ + examples/) that a teammate or agent can drop into `~/.claude/skills/` and use with their OWN warehouse-connect tool — no dbt project, no DB driver required. The emitted bundle carries knowledge + SQL-generation guidance, NOT an executor. Triggers: "package the dbt-wiki knowledge", "build a portable analytics skill", "freeze the knowledge base into a bundle", "export dbt-wiki for a teammate", "make a shareable analytics skill", "/dbt-wiki:pack", "把 dbt-wiki 知識打包", "封裝成可攜知識 skill", "做一個分析 skill 給同事", "dbt-wiki をパッケージ化", "知識 skill を書き出す". Do NOT trigger for: querying the knowledge base (use /dbt-wiki:query), building or refreshing `.dbt-wiki/` itself (use /dbt-wiki:init or /dbt-wiki:refresh), adding tribal knowl