ワンクリックで
psychquant-claude-plugins
psychquant-claude-plugins には PsychQuant から収集した 70 個の skills があり、リポジトリ単位の職業カバレッジとサイト内 skill 詳細ページを表示します。
このリポジトリの skills
Turn a journal / reference article PDF into a faithful, compile-verified LaTeX "article record" — metadata frontmatter + every theorem, definition and numbered equation (original numbers preserved) + references + flagged editorial notes — saved next to the PDF as year_author_title.tex. Reach for this whenever the user wants to transcribe, OCR, extract, "make a record of", or pull the theorems / equations out of a paper or PDF into LaTeX, or adds a reference PDF and wants it turned into a structured .tex — even when they literally say "OCR it": this skill checks for a text layer first and deliberately avoids OCR on born-digital PDFs, because re-recognizing rendered math is less accurate than reading the embedded text. Especially valuable for math-heavy papers where equation fidelity matters.
發布或更新 MCP Server 到官方 MCP Registry 及第三方平台(Glama、awesome-mcp-servers)
Unblock Apple notarization / signed releases when blocked by an updated or expired Apple Developer Program License Agreement (the notarytool / Transporter HTTP 403 "A required agreement is missing or has expired"). Accepts the pending agreement on developer.apple.com via Safari + AppleScript, then re-verifies and resumes the release. Use when: `xcrun notarytool` / `make release-signed` fails with 403 "required agreement", or user says "notarization blocked", "公證被擋", "Apple 協議過期", "accept apple agreement", "Program License Agreement", "notary 403", "release 卡在 Apple 協議". Do NOT use for 401 "Invalid credentials" (that is an app-specific-password / key problem — re-run `xcrun notarytool store-credentials`, a user-only action).
macOS native browser automation via Safari + AppleScript. Use when the user needs to automate websites that require login (Plaud, Elementor, banking, social media) — Safari preserves localStorage and cookies permanently. Also use when agent-browser fails due to session/auth issues, or when the user explicitly asks to use Safari. Triggers on: "login to site", "automate with Safari", "use Safari", "session expired with agent-browser", "Plaud upload", or any website automation where persistent auth is needed.
更新 Plugin 到最新版本(marketplace.json 同步 + marketplace update + plugin update + 安裝檢查)。當修改了任何 plugin 原始碼後需要同步、或用戶提到「更新 plugin」、「同步 plugin」、「plugin 沒生效」、「reload plugins」時使用。
LaTeX source 編譯前的 static check — punct_check(半形/全形標點)+ fonts_check(字型缺字,需既有 .log)+ source-level pattern audit。當用戶說「precompile check」「編譯前檢查」「跑 make check」「source audit」「掃半形標點」「掃缺字」時使用。對應暑期班 typesetting-checklist 的
預覽單一 LaTeX 片段(vocabbox / tikz 圖 / 數學公式)不必編整本 PDF — 用 compile_chunk 把片段包進 standalone class 編譯 → 單頁 PNG。當用戶說「預覽這個 vocabbox」「看一下這個 tikz 長怎樣」「試試這個公式」「快速看片段」「不要編整本」時使用。
編譯 LaTeX 專案後跑完整 layout 驗證流程 — detect_layout_issues + find_overlaps + fonts_check + box_warnings。當用戶說「驗證 LaTeX」「檢查排版」「掃 layout 問題」「LaTeX 編完看看有沒有 bug」「latex audit」或在 LaTeX 專案目錄編譯後想自動 audit 時使用。也適用於 commit 前驗證、release 前 audit、暑校稿 batch 修完想知道剩餘 issue。
改動 LaTeX source 後做視覺 diff,知道改動影響哪幾頁 — 用 compile_diff(git_ref) 自動 checkout + compile + compare,或用 compare_pdfs 比兩個既存 PDF。當用戶說「視覺 diff」「比較改動前後」「我改了 X 影響哪幾頁」「跟上一版比」「visual regression」時使用。也適用於 batch 修完想驗證、release 前 ripple 檢查。
Audit-time cross-doc drift detection for math manuscript. Runs R1 (working-file path leak in \texttt{}), R2 (citation key drift / bib orphan), R3 (code symbol drift between analysis/ and manuscript/docs/), R4 (proposition-jsonl iso-bijection check, optional when manuscript has propositions/main.jsonl). Different from `propositions` skill (R1-R8 mechanical gate on JSONL alone) and `proofread` skill (L4 semantic walk per-prop). This skill checks the **cross-artifact consistency** between manuscript/main.tex + propositions/*.jsonl + analysis/*.py + references/*.tex + refs.bib. Use when: 大改稿後 / 投稿前 / 新階段 (initial submission / major revision / camera-ready). NOT for per-PR (use `code-and-manuscript-sync.md` rule + pre-commit hook). v0.1.0 SCAFFOLDING — execution body TODO. Source scripts currently at PsychQuantHsu/psychophysical_representations/scripts/audit-*.py + run-audit.sh.
JSONL-driven 6-layer proofread workflow for math manuscript. Each prop in main.jsonl becomes a [ ] checklist item in manuscript/.proofread/<file>.md; walk through L1-L5 + location-drift per prop; mark [x] (CLEAN) / [~] (finding) / [-] (out of scope). L1 = text-claim match (asserts truly atomic + faithful paraphrase) L2 = claim_type fit (axiom non-derivable / definition has equality / commentary not derived / etc.) L3 = cite completeness (all external refs declared) L4 = cite validity (each cited prop logically implies this prop's asserts) L5 = evidence_class fit (derived needs cites / axiomatic truly axiom / etc.) +location drift (claimed line range matches main.tex actual) Use when: pre-submission final polish, post large rewrite, Hsu-approval-pending area, validating prop-extraction quality. NOT for daily micro-edit (use sync rule + validator). v0.1.0 SCAFFOLDING — execution body TODO. Methodology frozen in PsychQuantHsu/psychophysical_representations/manuscript/.proofread/main_jsonl_l4_walk.md.
Extract atomic propositions from a math manuscript .tex file into a JSONL structural representation, then run R1-R8 mechanical validation. Each prop carries text (verbatim from .tex), location (line range), cites (internal UUID dependency DAG), asserts (atomic claims), claim_type, and evidence_class. Use when: starting propositions side-file for a new manuscript, re-extracting after large rewrite, or running mechanical gate before audit. NOT for L4 semantic walk — that's the /proofread skill. v0.1.0 SCAFFOLDING — execution body TODO. Source-of-truth scripts currently in PsychQuantHsu/psychophysical_representations/scripts/validate-propositions.py.
Semantic synthesis between 2+ document versions (not git-style line merge). Input: N source docs (.tex / .md / .ipynb / .docx). User specifies dimension-level take/leave decisions (e.g., "structure from A, methodology section from B, conclusion from C"). Plugin produces synthesized output with provenance tracking. Use when: combining advantages from multiple drafts, integrating reviewer suggestions with original draft, merging student work + teacher reference into best-of-both. Differs from: git merge (mechanical line-level), revise (single source improved), ensemble (output averaging — this picks dimensions, not averages outputs).
Import a Google Drive file into a GiftHub repo — move to LFS folder, create pointer, and pull. Use when user says "download recording", "import video", "把 Drive 檔案拉下來", "gfh import".
Batch OCR pipeline orchestrator wrapping macdoc — PDF→PNG split, parallel OCR with SSH tunnel health check + auto-retry, per-page → per-PDF markdown merge, idempotent resume from interruption. Trigger when user says "batch OCR", "OCR all the PDFs in this folder", "do OCR on 50+ PDFs", "transcript exam OCR", or invokes /batch-ocr.
部署 Swift CLI 工具(編譯 universal binary、建立 GitHub Release、安裝到 ~/bin)。在 Swift CLI 專案目錄中使用。
建立新的 Swift CLI 專案骨架(Package.swift + ArgumentParser + Version.swift + .gitignore)
部署 MCP Server 專案(編譯、打包 mcpb、建立 GitHub Release)
互動式建立新 MCP Server 專案(Swift/Python/TypeScript)
Apply Developer ID signing + notarization pipeline to a Swift MCP / CLI project. Required for macOS 26 — ad-hoc signed binaries can no longer trigger TCC permission dialogs. Use when creating new MCP/CLI (mcp-new-app / cli-new-app integration) OR upgrading an existing project that needs to ship signed releases. Templates are extracted from che-ical-mcp PR
為單一 plugin 從 plugin.json description 反向 bootstrap CHANGELOG.md(Keep a Changelog 1.1.0 格式)。 Parse description 的 vX.Y.Z 段落(同時支援 colon-style `vX.Y.Z:` 和 space-style `vX.Y.Z `),按 plugin 自己的 major version 過濾掉 dep version 噪音,git log pickaxe 解析每個 version 的引入日期,分類 prose 到 Added/Changed/Deprecated/Removed/Fixed/Security 六個 KAC section。 另一個 mode: normalize 把既有非 KAC CHANGELOG.md(如 PsychQuant 的 em-dash 格式 `## 2.37.0 — 2026-05-02`)改寫成 KAC 標準(`## [2.37.0] - 2026-05-02`)。 Use when: 你想為一個沒 CHANGELOG.md 的 plugin 補一個;或想把舊格式 normalize 成 KAC 標準。 防止的失敗:手寫 KAC entries 太繁瑣 → 索性不寫 CHANGELOG.md;或寫了但格式偏差被 marketplace consumer 解析錯。
跨整個 Claude Code marketplace 批次跑 changelog-init — 為所有缺 CHANGELOG.md 的 plugin 補上 KAC 1.1.0 格式的檔案,並產出 migration report。 Use when: 你的 marketplace 有大量 plugin 沒 CHANGELOG.md(PsychQuant 起初 35/36);想一次性整頓 changelog 健康度;想看哪些 plugin 的 description 結構好/差(segment 數可作為品質代理指標)。 防止的失敗:手動為 N 個 plugin 跑 init 太費時;漏了某些 plugin;migration 結果沒留 audit trail。
驗證 plugin 的 CHANGELOG.md 是否符合 Keep a Changelog 1.1.0 規範,並檢查三處同步(CHANGELOG.md latest entry ↔ plugin.json description ↔ marketplace.json description)。 Exit codes: 0 pass / 1 missing CHANGELOG / 2 KAC violation / 3 sync drift / 4 IO error。CI 友善。 Use when: 發 release 前檢查格式對不對;audit 一個 marketplace 的 CHANGELOG 健康度;改完 description 後驗 sync 沒漏。 防止的失敗:plugin.json 跟 CHANGELOG 版本不同步(marketplace 顯示舊版);用了非 KAC section name(### 改進、### Bug Fix 等);忘記寫日期。
NSQL-derived confirmation workflow for email operations. Use this skill BEFORE executing archive-mail with vague filters, compose_email (sending mail), delete_email/move_email in bulk, or any operation that touches 5+ emails. Show user a structured preview of "what I understood" before taking action, achieving consensus through dialogue. Skip when the input is unambiguous (specific Message-ID, single mark_read, etc.).
Show structured preview of bulk email operations (5+ emails) before execute. Group by thread, flag false-positive candidates, count side-effect scope (files written, attachments downloaded, mailboxes touched). Use after email-search-disambiguation finishes Phase 1, as Phase 2 of the confirmation protocol.
Disambiguate vague email search filters before executing search. Triggers when user uses ambiguous contact references (Chinese name, nickname, role like "老闆"/"老師"), vague time references ("最近"/"上週"), or scope words ("全部"/"所有"). Present 2-3 concrete interpretations and let user choose. Adapted from nsql disambiguation triggers (D1-D5) for email domain.
Guide for using Telegram MCP tools across both servers — telegram-all (personal account, full access) and telegram-bot (Bot API). Covers reading chats, sending messages, search, history, group management, and how to pick the right server. Use when the user asks about Telegram messages, chats, contacts, or bot operations.
發布 plugin 到自己的 marketplace(pre-flight check + version bump + commit + push + sync)。 完整的發布流程,確保 plugin 品質和 marketplace 同步。 當用戶提到「發布 plugin」「deploy plugin」「上架」「release plugin」時使用。
Query OpenAI Codex CLI configuration, features, and documentation. Use this skill proactively when the conversation involves: - Codex CLI installation, setup, or authentication - Codex CLI configuration (config.toml, AGENTS.md, profiles) - Codex models (GPT-5.5, GPT-5.3-Codex, Codex-Spark) - Codex CLI commands, flags, slash commands - Codex features (fast mode, web search, multi-agent, MCP, skills) - Codex speed/reasoning settings - Codex SDK, non-interactive mode, automation - Codex security, sandboxing, approval modes - Codex IDE extension or app configuration - Codex integrations (GitHub, Slack, Linear)
智慧分配 Lean 4 定理證明工作給 Codex。 我(Claude)先分析 sorry/stub,判斷 Codex 適合做什麼, 只把「Codex 擅長」的部分派給他,其餘自己處理。 Use when: 有 sorry 或 True := trivial 需要處理、想利用 Codex 的暴力搜索能力。
Star a GitHub repository in Safari. Use when the user asks to star, favorite, or bookmark a GitHub repo via Safari and wants the action performed inside their own logged-in Safari session.
Upload audio files to Plaud via Safari. Use when the user asks to upload a recording to Plaud, add tracks to a Plaud folder, or trigger Plaud transcription from a file that is already on disk.
Guide for effective use of che-apple-notes-mcp for macOS Apple Notes management. Use when user asks about Apple Notes, 備忘錄, note taking on Mac, creating/searching/organizing notes.
Clone 競品 MCP Server 原始碼到 references/ 進行分析
Clone 參考 MCP Server 到 references/ 並自動分析可升級功能
MCP Server 功能除錯(框架分析、權限問題、錯誤診斷)
MCP Server 連線診斷(檢查連線、binary、基本呼叫)
從 GitHub Release 安裝 MCP Server 到 ~/bin
快速對 MCP Server 的 GitHub repo 開 Issue(bug、feature、改善建議),之後再處理
同步 MCP Binary(.build → mcpb/server → ~/bin 一致性檢查和修復)