devinat1/skills

記録された最新のソース活動: 2026年11月6日 16:50

SkillsMP カタログ更新: 2026年11月12日 12:52

収集済み skills: 48

GitHub スター: 1
GitHub フォーク: 0

AI アシスタントでインストール

このプロンプトをコピーして、使用中の AI アシスタントに送ってください。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/devinat1/skills.

リポジトリの README

README の元ファイル：README.md

README の元ファイルを読む

README 全文を表示

# Skills Agent skills for learning, productivity, and writing workflows. Engineering, hands-on lab, and interview-practice skills live in [`devinat1/engineering-skills`](https://github.com/devinat1/engineering-skills). ## Quickstart (30-second setup) 1. Run the skills.sh installer: ```bash npx skills@latest add devinat1/skills ``` 2. Pick the skills you want and which coding agents to install them on. **Include `setup-devinat1-skills`.** 3. Run `/setup-devinat1-skills` in your agent. It will ask for: - Agent memory project ID (for the learning loop) - Blog content directory (for `/blog` and `/update-blog-refs`) - MCP integrations (agentmemory, Granola, Todoist) 4. For Claude Code users who clone this repo directly: ```bash git clone git@github.com:devinat1/skills.git ~/.agentic/repos/skills ./scripts/link-skills.sh ``` ## Shared agent home The local checkouts live at `~/.agentic/repos/skills` and the sibling `engineering-skills`. Harness skill directories are symlinks to the union catalog at `~/.agentic/skills`. Owned catalog entries link to their repository sources. Global instructions are maintained in `config/agentic/AGENTS.md`, exposed through `~/.agentic/AGENTS.md`. Project maintenance instructions live in each repository's `AGENTS.md`; `CLAUDE.md` links to it. Native credentials, sessions, caches, and plugin installations remain harness-owned. The root `settings.json` is an installation example, not the live Claude configuration. Run `agentic upkeep --dry-run` to preview new skills from any harness and `agentic upkeep --apply` to adopt them and restore discovery symlinks. Daily upkeep uses this command, preserves conflicts without changes, and retains rollback snapshots for each changed run. Skills installed through an intact discovery link already live centrally; upkeep registers them in the index. Run `agentic doctor` to check links, indexed paths, and repository integrity; run `python3 scripts/audit-agentic-paths.py` to audit skill storage references. Use `agentic path blog_content` (or another index key) to resolve external data. Shell/Python/JavaScript helpers honor `AGENTIC_HOME` (default `$HOME/.agentic`). After skill source edits, run `./scripts/link-skills.sh`. Legacy synchronization entrypoints now call doctor without copying or deleting catalogs. For an existing checkout in the old Claude directory, first pause the obsolete catalog-sync automation, then run `python3 scripts/migrate-agentic-home.py --dry-run`. Resolve reported conflicts with an explicit `--prefer NAME=SOURCE`; run the same command with `--apply` when the inventory is correct. `--verify` checks the result. `--rollback MIGRATION_ID` restores original files and retains newer generated files under the transaction's `reverted-content/` directory. Never delete the rollback snapshot without the user's explicit request. App project registrations and automation settings must be restored through the app separately on rollback. State, memory, artifacts, and the private root `index.json` are outside these Git repositories. `config/agentic/index.schema.json` defines the index structure. Historical design documents retain historical paths. Managed `.system` installer and configuration integrations are exempt from the shared-state path audit. ## Skills  ### Learning - **[exam](./skills/learning/exam/SKILL.md)** — Generate a complete standalone exam and a separate visible answer key for one confirmed topic. Use when the user asks for an exam, knowledge test, multiple-choice or free-form questions, or selects the exam modality in /learn. It authors files but does not administer or grade them. - **[illustrate](./skills/learning/illustrate/SKILL.md)** — Create a reusable interactive diagram that explains one concept, its terminology, connections, flow, and mental model. Use when the user asks to illustrate, visualize, map, or show how a concept fits together. It explains visually but does not grade understanding. - **[learn](./skills/learning/learn/SKILL.md)** — Route a conversation, path, URL, file, or topic to one focused learning modality. Use when the user wants to learn but has not chosen between a Socratic lesson, interactive illustration, coding lab, or exam. It identifies and confirms the topic, recommends a modality, and hands off without teaching or grading. - **[socratic-teacher](./skills/learning/socratic-teacher/SKILL.md)** — Teach one confirmed topic interactively through brief explanations, explain-backs, and one guiding question at a time. Use when the user wants a lesson, Socratic teaching, guided understanding, or asks to be taught rather than shown a diagram, given a lab, or tested. ### Productivity - **[accountability-beeminder](./skills/productivity/accountability-beeminder/SKILL.md)** — Create and settle evidence-backed accountability commitments with user-chosen BeeMinder charges. Use when a user commits to a measurable or subjective outcome, asks to be held accountable, wants a due commitment checked, answers whether a proposed miss is fair, or when a scheduled task checks accountability commitments. - **[ai-detection](./skills/productivity/ai-detection/SKILL.md)** — Assess whether supplied text or source code was AI-generated or substantially AI-assisted by calling Jev and reporting its estimated probability. Use on /ai-detection or requests to check AI authorship. - **[clarify](./skills/productivity/clarify/SKILL.md)** — Clarifies purpose, constraints, and success criteria by proposing question groupings and asking one batch at a time (at most three questions), then submits a plan for review in Plannotator. Use when the user has a vague idea, says "clarify this", "help me figure out what I want", "what should I build", or wants discovery before implementation. - **[cognitive-fitness](./skills/productivity/cognitive-fitness/SKILL.md)** — Coach instead of replacing the user's thinking when a request delegates difficult reasoning, problem-solving, or creativity; asks for writing for human readers (not agent-only content); or uses broad summaries in place of reading substantive sources. Honor explicit finished-output overrides and never interrupt an ongoing run or walkthrough. Also use on /cognitive-fitness. - **[coherent](./skills/productivity/coherent/SKILL.md)** — Use when the user wants to practice explaining a clarified idea, especially when the explanation assumes hidden context, mixes concepts, lacks a concrete example, or needs a concise research pitch for a skeptical audience. - **[confounder](./skills/productivity/confounder/SKILL.md)** — Check an idea's atomic claims for conflated concepts, variables, scopes, or reasoning. Use when the user invokes /confounder or asks whether an idea mixes distinct things together. - **[dissenter](./skills/productivity/dissenter/SKILL.md)** — Stress-test a choice with an independent LLM view and Jev judgment (Choice, Score, or Noul). Use before giving user-facing advice, recommending, or evaluating a choice when two plausible courses exist, including low-stakes choices; skip routine execution, factual lookups, and status questions. Explicit /dissenter requests can also evaluate factual questions. - **[dunning-krueger](./skills/productivity/dunning-krueger/SKILL.md)** — Assess what knowledge the user demonstrated and which gaps the evidence confirms from the current conversation, supplied context, or completed questions and answers. Use for an evidence-based knowledge assessment or blind-spot check. Reports and persists both strengths and gaps without generating a grill or lesson. - **[experience](./skills/productivity/experience/SKILL.md)** — Update agent memory with diagnostic feedback from the current session. Use when the user invokes /experience. - **[field-snapshot](./skills/productivity/field-snapshot/SKILL.md)** — Produce a standalone, source-backed snapshot of a research area's recent venue programs, active researchers and groups, trends, and up to ten papers to read. Use for field updates, not idea-specific prior-work reviews. - **[focus](./skills/productivity/focus/SKILL.md)** — ADHD work assistant via Todoist — stuck sessions, Focusmate timeboxing (25/50/75 min), and overcommitment guard for any task you name. Use when user says stuck, avoiding a task, break down, Focusmates, timeboxing, overcommitted, or names a Todoist task/project to plan. - **[incentive-check](./skills/productivity/incentive-check/SKILL.md)** — Research a person's current financial and nonfinancial incentives related to a specific statement, then use Jev to estimate whether each incentive exists and whether accepting the statement would advance it. Use on /incentive-check or when asked who benefits if a person's statement is believed. - **[is-grass-greener](./skills/productivity/is-grass-greener/SKILL.md)** — Expose the pros and cons of named past, possible, or current life paths without recommending a choice. Use when the user invokes /is-grass-greener, asks whether the grass would have been greener, or wants an even tradeoff view of career, education, relationship, location, or another major life path. - **[jev-knowledge-check](./skills/productivity/jev-knowledge-check/SKILL.md)** — On /jev-knowledge-check, assess whether explored evidence suffices for a reliable answer. - **[linear](./skills/productivity/linear/SKILL.md)** — Convert conversations, current chat context, transcripts, meeting notes, or rough discussions into scoped Linear tickets. Use when the user invokes /linear or asks to turn a conversation into Linear issues, search for existing related tickets, merge context into existing issues, or create reviewed Linear tickets. - **[literature-review](./skills/productivity/literature-review/SKILL.md)** — Exhaustive prior-work discovery via background subagents — academic-heavy, all four lanes to saturation, deep paper treatment. Use when the user invokes /literature-review or wants prior work before a novelty judgment. Does not assess novelty (use research-gap). - **[lock](./skills/productivity/lock/SKILL.md)** — Keep one focused task in its owning chat using a persistent personal AgentMemory lock. Use on /lock, /lock unlock, requests to unlock, the shared chat-start focus check, and automatic entry from learn, mentor, socratic-teacher, coherent, or grilling. - **[meeting-feedback](./skills/productivity/meeting-feedback/SKILL.md)** — Use when the user wants quick, direct feedback on their communication performance in a meeting. Fetches Granola notes/transcripts and returns a short coaching note with what worked, what got in the way, and one behavior to try next time. Trigger on /meeting-feedback or when the user asks for meeting feedback, meeting review, or communication coaching. - **[mentor](./skills/productivity/mentor/SKILL.md)** — Guide an overwhelmed user through one task with one tiny, collaborative next action at a time, adapting feedback and pace from the work and similar-task history. Use when the user says they are overwhelmed, have too much on their plate, want to work one step at a time, need a guiding hand while writing or reviewing, or invokes /mentor. - **[momtest](./skills/productivity/momtest/SKILL.md)** — Audit a customer-interview transcript against The Mom Test. Use when the user invokes /momtest. - **[next-hour](./skills/productivity/next-hour/SKILL.md)** — Choose one concrete task for the upcoming hour from goals stored in AgentMemory. Use on /next-hour or when the user asks what to do for the next hour without choosing between options. - **[novelty-search](./skills/productivity/novelty-search/SKILL.md)** — Assess CS research novelty using cspapers.org and Consensus MCP, with contribution-level and overall Jev probabilities through dissenter. - **[obsidian](./skills/productivity/obsidian/SKILL.md)** — Navigate and write to the State and Church Obsidian vaults via CLI, including agent-managed notes under Agentic/. Use when the user wants to capture ideas, find notes, read vault context, or mentions Obsidian, either vault, zettels, or MOCs. Also covers Obsidian CLI syntax and plugin/theme development (reload, eval, screenshots, DOM). - **[obsidian-excalidraw-cli](./skills/productivity/obsidian-excalidraw-cli/SKILL.md)** — Create, inspect, edit, export, or test Obsidian Excalidraw drawings through the local Obsidian CLI, including timeout and BUSY recovery. - **[pragmatic](./skills/productivity/pragmatic/SKILL.md)** — Use when a non-trivial build needs a two-person reality check before implementation: limited bandwidth, rough deadline, scope check, technical risk, business value, MVP sizing, or pragmatic build brief. - **[ramble](./skills/productivity/ramble/SKILL.md)** — Extract actionable items from the conversation and create Todoist tasks. Use when the user invokes /ramble. - **[research-advisor](./skills/productivity/research-advisor/SKILL.md)** — PhD-advisor-style research brainstorming that bluntly stress-tests an idea's problem, research question, mechanism, contribution, evidence, prior work, and security claims. Use only when the user invokes /research-advisor or explicitly asks for direct, adversarial academic research feedback. - **[research-gap](./skills/productivity/research-gap/SKILL.md)** — Evidence-backed opposing novelty views from a literature-review report (runs literature-review first if none exists in chat). Use when the user invokes /research-gap or asks whether a research question or idea is novel. - **[research-zettels](./skills/productivity/research-zettels/SKILL.md)** — Split mixed research notes or a conversation into approved research questions, deduplicate their claims into atomic State zettels, and connect them with question outlines and an umbrella MOC. Use when the user wants reusable claim notes from one source or several overlapping research documents. - **[safeguard](./skills/productivity/safeguard/SKILL.md)** — REQUIRED before any non-trivial feature, build, or refactor: invoke this skill as your first action before code, scaffolding, plans, or exploratory edits. User and project gate rules mandate this. Interviews the user on company necessity, what/why, codebase impact, and scope by proposing question groupings and asking one batch at a time (at most three questions), then submits a plan for Plannotator approval. Also when user says "safeguard", "/safeguard", "understand before code", or "before building". Not for clear bug fixes (use clarify) or when a detailed plan/issue already exists. - **[scope-creep](./skills/productivity/scope-creep/SKILL.md)** — Detect and stop scope creep when a session contains two or more distinct ideas by selecting one highest-impact focus and sorting the rest. Use proactively when a second idea appears, or when the user invokes /scope-creep. - **[skill-radar](./skills/productivity/skill-radar/SKILL.md)** — Search relevant locally installed skills first, then available plugins and online catalogs; ask the user before invoking or installing any match. Use for the first substantive user request in every session, and when a user asks for skill or plugin recommendations. - **[stack-walkthrough](./skills/productivity/stack-walkthrough/SKILL.md)** — Explore a PR stack, connect each PR to available results, and ask one question at a time to decide what to keep, change, or remove. Use for an interactive stack walkthrough before implementation. - **[statement-walkthrough](./skills/productivity/statement-walkthrough/SKILL.md)** — Walk a chosen subject one statement at a time and ask one question per statement to decide what to keep, change, or remove. Use for an interactive decision walk before implementation, including when another skill supplies the statements. - **[todo-triage](./skills/productivity/todo-triage/SKILL.md)** — Review Todoist tasks due today or overdue plus the triage backlog, group them by topic, choose focus tasks, and triage the rest. Use for a morning Todoist review, daily focus selection, or when the user invokes /todo-triage. - **[unscramble](./skills/productivity/unscramble/SKILL.md)** — Extract and organize every distinct claim from a conversation, pasted text, readable file, or Granola meeting. Use when the user asks to unscramble, untangle, or separate claims without analysis. ### Writing - **[apset](./skills/writing/apset/SKILL.md)** — Organizes current work into APSET (Area, Problem, System, Evaluation, Takeaway) through a clarify-style interview. Use when the user explicitly asks for an APSET, says "/apset", or says "organize this with APSET". - **[blog](./skills/writing/blog/SKILL.md)** — Convert the current conversation into a focused blog post (clarify scope first, then distill). Use when the user invokes /blog. - **[paper-submission-audit](./skills/writing/paper-submission-audit/SKILL.md)** — Audit a research-paper draft and its target venue for writing, evidence, presentation, and submission-readiness problems. Use when checking a paper before submission; audit only, no rewriting or submitting. - **[post](./skills/writing/post/SKILL.md)** — Use when the user asks to draft, make, schedule, queue, or publish social posts from the current conversation, blog post, article, launch note, or existing post context, especially for X/Twitter and LinkedIn through Postiz. - **[research-idea-brief](./skills/writing/research-idea-brief/SKILL.md)** — Turn a research or product idea into a concise four-section brief: problem, prior work, remaining gap, and proposed approach. Use when the user asks to package, frame, structure, or write up an idea, research direction, benchmark proposal, or gap analysis in this format. - **[update-blog-refs](./skills/writing/update-blog-refs/SKILL.md)** — Scan blog posts and suggest related cross-links. Use when the user invokes /update-blog-refs. - **[video](./skills/writing/video/SKILL.md)** — Create, edit, analyze, script, animate, or repurpose video through one entry point. Use for footage cleanup, video questions, YouTube scripts and Shorts, launch videos, meme edits, visual intercuts, product demos, AI video generation, and Hyperframes composition or rendering. ### Setup - **[setup-devinat1-skills](./skills/setup/setup-devinat1-skills/SKILL.md)** — Sets up an `## Agent skills` block in AGENTS.md/CLAUDE.md and `docs/agents/` so devinat1's skills know the agent memory project ID, blog directory, and MCP integrations. Run before first use of the learning skills, `experience`, `blog`, or `update-blog-refs`.

リポジトリの README を読み込んでいます…

投稿

すべて見る →

Devin Ersoy

@devinersoy4962026年9月29日 21:12 UTC

Skill公開の投稿要件を明確にする

投稿の要約 · 英語

Devin introduces Scope Creep, which separates competing ideas, helps choose one current focus and sorts the rest into later, discard or done. Its source workflow also calls related Skills for idea extraction and a challenge to the focus choice.

元の投稿

Skill の詳細を見る · skills/skills-productivity-scope-creep

このリポジトリの skills

ソフトウェア開発者グラフィックデザイナーその他の高等教育教員プロジェクト管理専門家

4 件の職業カテゴリ · 10% 分類済み

未分類

Audit a customer-interview transcript against The Mom Test. Use when the user invokes /momtest.

原文の言語: 英語

2026-09-21

収集済み skill 48 件中 40 件を表示しています。