Coding & Development

Update CHANGELOG.md from commits since the last tag, following Keep a Changelog. Use when the user asks to update the changelog or prepare a release.

Branch naming convention and the rule to branch before changing code on the default branch (main/master). Use when starting new work, creating a branch, or when about to edit code while on the default branch.

Skill router. Use ONLY when the user's message contains "/route" (at the start, at the end, or sent alone after a request); never on your own initiative. Picks the best 1 to 5 skills from every installed skill for the user's request, shows which ones it loads and why, loads them, does the work, then suggests which skills to run next and when. "/route build" creates the personal skill catalog (run once), "/route refresh" re-syncs it after skills are installed or removed.

Systematic root-cause debugging. Use whenever something is broken, failing, flaky, slow, or "weird" — error messages, stack traces, failing CI, 5xx in production, a Lambda timing out, a Terraform apply erroring, a pod in CrashLoopBackOff. Use it before proposing any fix so the fix targets the cause, not the symptom.

Onboard an autonomous agent onto Proof of Agent (minepoa.com) and Proof Floor (poa.grok.me) on Linux, macOS, or Windows. Registers or imports an identity into ~/.poa/key with 0600 permissions, configures header-authenticated MCP, runs the first pulse check, and sets up twice-hourly scheduling (:21 and :51).

Interact with Proof Floor (https://poa.grok.me) using poa-agent-kit (`poa.py floor`) over JSON-RPC MCP (`https://poa.grok.me/api/mcp`). Read live network pulse and history, check weekly standing, inspect peer addresses, post to the Trollbox, and coordinate via verified agent-to-agent DMs while maintaining strict OPSEC.

Use SEMPRE ao terminar qualquer solicitação que tenha alterado arquivos do projeto (implementação, correção, ajuste, refatoração, configuração, documentação), mesmo que o trabalho ainda vá ter alterações depois. Use também quando o usuário disser "finaliza", "salva", "salvar memória", "commita", "sobe", "encerrar sessão", "deixa recado", "lembra a próxima sessão", "o que as outras sessões fizeram", e sempre que houver referência a trabalho anterior. Executa automaticamente o fluxo completo de documentação, git pull --rebase, commit convencional, push para dev, consulta de produção e salvamento compactado no Obsidian.

Runs Proof of Agent peer grading and governance on Linux, macOS, or Windows using poa-agent-kit. Audits open grading assignments for prompt injection, missing or inflected seed words, and author template reuse; submits calibrated 0-3 grades sequentially; and checks/casts governance votes.

Principal Software Architect skill that audits any workspace, resolves skill collisions, synthesizes an adaptive Mermaid execution pipeline, and generates a high-density, production-grade GEMINI.md flight manual (<75 lines, zero redundancy) with archetype-adaptive directory isolation and zero root pollution.

Overnight autonomy protocol — the user is going to sleep and the agent must keep working unsupervised: keep the GPUs busy with the most valuable experiment it can justify, chain the next job when one finishes, and have a report ready when they wake. Use when the user says "bedtime", "/bedtime", "going to bed", "good night", "gn", "sleeping now", "keep going overnight", or "don't let the GPUs idle". Not for short absences ("back in an hour") — that is normal away mode.

国内（中国大陆网络）GitHub 访问慢、clone/fetch 掉线时的加速方案。自动轮换 GitHub 加速前缀并带重试，配好 git 全局静态加速 + ghclone 别名。

Quick update — the ultra-terse status format for when the user comes back after the agent has been working alone for a while. Three fixed headers only: RUNNING / RESULTS / NEED YOU. Use when the user says "qu", "/qu", "quick update", "quick status", "what's up", "back, anything?", or returns from an absence and asks how it went. Also the default register for any unprompted update after a long autonomous stretch.

비개발자·초보자용 하네스 생성 스킬 (코딩 도메인 — 웹앱·API·모바일·데이터/AI). harness-100의 검증된 exemplar를 복제·치환하여 빠르고 안전하게 하네스를 만든다. 3개의 객관식 질문만 받으며, 자유 설계 대신 template 복제 전략을 사용한다. "쉽게 코딩 하네스 만들어줘", "처음 앱 만들어봐", "간단하게 개발 팀 만들어줘", "템플릿으로 웹앱 시작", "풀스택 팀 빠르게", "API 팀 템플릿", "ML 실험 팀", "비개발자 앱", "starter coding harness", "easy coding harness", 표현 시 사용. 단, 문서·분석·기획 등 워크스페이스 도메인은 harness-starter-workspace 를 사용. 호출 전에 사용자 의도를 3가지 관점(①도메인: 코딩·문서·혼합·검증 / ②숙련도: 처음·숙련·전문가 / ③목적: 시작·개선·검증)으로 간단히 확인한 뒤 진행한다. 사용자가 이미 명시한 경우 확인을 생략할 수 있다.

AI 가 한 일과 낸 결과, 그 의도를 글만으로는 이해하기 어려울 때, 읽는 사람이 어디서 막히는지 찾아 그 막힘을 푸는 HTML 한 장을 만든다. 형식은 정해 두지 않는다 - 직접 그린 구조도와 흐름도, 단계 재생, 값을 바꿔 보는 그림, 막대와 와플, 실제로 띄워 찍은 화면과 저장된 값, 같은 범위로 뽑은 도면, 바뀐 코드의 diff 까지 병목에 맞는 것을 고르고, 없으면 그 페이지 안에 만든다. 받는 사람은 문장을 골라 의견을 달고 결정을 골라 답을 묶어 회신한다. 남에게 건네는 보고(개선 전후, 기획서 대조, 원칙 점검, 상태 대조, 의사결정)면 판정을 세고 휴대폰과 A4 인쇄까지 맞춘다. 답이 길고 구조가 얽혀 글로는 안 읽히겠거나, 터미널 표가 커지거나, 화면에 드러나는 큰 작업을 마쳤으면 요청이 없어도 먼저 떠올린다. "보여줘", "html 로 보여줘", "시각화해줘", "그림으로 설명해줘", "다이어그램 그려줘", "구조 그려줘", "한눈에 보이게", "글로는 이해가 안 돼", "뭐가 바뀌었는지 보여줘", "diff 리뷰", "계획 리뷰", "비교해줘", "프로젝트 정리", "전후 비교", "캡처해서 설명", "구현한 거 보여줘", "기획서대로 됐는지 확인", "결과보고서", "고객에게 보낼 자료", "의사결정문"류 요청에 발동합니다. (pptx, docx 처럼 정해진 제출 파일 형식이 필요하면 그 문서 도구의 몫이다.)

对建筑/街道/街区做空间历史考古，挖出层层堆积的演变史，交付可核查的空间考古档案。触发：daoyou、「导游一下 XX」、「把 XX 考古一遍」。

비개발자·초보자용 하네스 생성 스킬 (워크스페이스 도메인 — 분석·기획·문서·리서치). harness-100의 검증된 exemplar를 복제·치환하여 빠르고 안전하게 하네스를 만든다. 3개의 객관식 질문만 받으며, 자유 설계 대신 template 복제 전략을 사용한다. "쉽게 하네스 만들어줘", "처음 만들어봐", "간단하게 팀 만들어줘", "템플릿으로 시작", "유튜브 팀 만들어줘", "블로그 팀", "계약서 검토 팀", "강의 만들기 팀", "비개발자", "코딩 모름", "starter harness", "easy harness", "template harness", 표현 시 사용. 단, 코드·API·앱·데이터 파이프라인 등 코딩 도메인은 harness-starter-coding 을 사용. 호출 전에 사용자 의도를 3가지 관점(①도메인: 코딩·문서·혼합·검증 / ②숙련도: 처음·숙련·전문가 / ③목적: 시작·개선·검증)으로 간단히 확인한 뒤 진행한다. 사용자가 이미 명시한 경우 확인을 생략할 수 있다.

把計劃、提案、草稿或完稿變成「可框選評論、可勾選」的 HTML 預覽頁，在電腦上用預設瀏覽器打開。解說型內容先改寫成白話、不重複、圖像化（對照、架構、時間軸、步驟、數字卡、重點框，用幾行短語法由產生器畫圖）再產頁。支援框選任意文字留評論、勾選確認、選擇自己是誰、一鍵複製全部評論。當使用者說「做成預覽」「預覽」「做個可評論的版本」「框選評論」「讓我勾選」「做成勾選的」「給客戶評論」「preview」時使用；要使用者從一串項目裡逐項決定（要／不要、保留／刪除）時也主動用勾選。

LITE SUBSET of ProGoals Permission Gate. Classify every planned agent action as READ, REVERSIBLE or IRREVERSIBLE, and require the human's explicit approval (with an exact approval card) before sends, DMs, payments, deletes and shares. Use before any tool call that leaves the agent's own workspace.

Run tarot reading practice sessions with Jeff. Use this skill whenever Jeff wants to practice tarot card readings, says "let's do tarot practice", "I want to practice readings", "give me a tarot scenario", "deal me some cards", or anything suggesting he wants to drill his card interpretation skills. This skill sets up realistic seeker scenarios, draws random cards (with reversals), coaches Jeff through his interpretations, and gives targeted feedback — including honest pushback when the read goes off the rails. Always use this skill when Jeff wants tarot practice — even if he just says "let's do a round" in the context of a prior tarot session.

Test a ready-made list of deliverables, documents or artifacts before anyone commits to tracking it. Use when someone is handed a list from a template, vendor, consultant, PMO framework or AI (\"standard project documents\", \"required artifacts\", \"deliverables checklist\") and asks whether to adopt it, what is really required, what to drop, or what is missing. Tags each item by who requires it (funder, law or regulation with citation, a named person, or standard practice that nobody requires), reads the funder's or agency's own template section by section to find required items the list leaves out, and returns a table plus a drop / keep / add list. Not for building a list from scratch with no list in hand, and not for checking whether finished documents are good.

Verify factual claims before delivering them. Use before any answer or deliverable the user will act on, forward, or share with others — rules, policy, legal, financial, date, or product facts — and whenever they say 'is that right', 'double-check that', 'verify this', 'fact-check', 'are you sure', or 'source?'. Exception: vetting a claim someone else handed the user is critical-thinking-walkthrough; fetching page text is reliable-fetch (this skill calls it).

Read-only preflight check of a repo's git/GitHub setup before pushing — especially the first push. Reports which GitHub account the push will authenticate as, which identity the commits will be authored with, whether the remote URL is being rewritten by insteadOf, and whether the branch and remote state will actually accept the push. Use this whenever the user says "check my github setup", "am I ready to push", "before first push", "which account will this push as", "did I set the remote right", "verify my remote", "wrong git account", "pushing as the wrong user", or hits "src refspec main does not match any", "Repository not found", "403", "GH007", "updates were rejected", or "Permission denied (publickey)". Also use it proactively before helping someone set up or push to a brand-new repository, and whenever a machine has more than one GitHub account or SSH host alias, since the failure mode there is silent — the push succeeds under the wrong identity.

Load general development standards. Applies to all repos. Invoke at the start of a session or when standards need reinforcing.

Commit staged and unstaged changes with an auto-generated Conventional Commit message, then sync the branch with the remote. Use this skill whenever the user says "commit", "save my changes", "push this", "commit and push", "sync", or any variation of wanting to commit their work and get it to the remote. Also trigger when the user says "ship it", "land this", "save and sync", or asks to wrap up a coding session. Do NOT use for complex git operations like rebasing, cherry-picking, or merge conflict resolution.