skills
來自 scenario-labs/skills 的 67 項代理技能 — 包含 scenario、scenario-game-assets、scenario-video。使用 npx skills add scenario-labs/skills 安裝。
AI 翻譯
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證A collection of reusable Agent Skills for software development workflows.
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
A collection of reusable Agent Skills for software development workflows. Built for Claude Code as a plugin; the individual SKILL.md files are also usable by any agent that reads the Agent Skills format.
• Spec authoring — turn requirements into a structured spec before coding.
• Dev flow — discuss → implement → wrap up a feature.
• Notion integration — bridge between local code/reports/meeting notes and Notion's Roadmap + Meetings databases.
• Utilities — standalone helpers.檢視完整 README ↗A collection of reusable Agent Skills for software development workflows. Built for Claude Code as a plugin; the individual SKILL.md files are also usable by any agent that reads the Agent Skills format.
Grouped by workflow:
| Skill | Purpose |
|---|---|
| spec | Generate a multi-file engineering specification (architecture, data models, APIs, features) from a system description. |
| spec-review | Evaluate a specification.md for clarity, completeness, and consistency — focused on data model correctness and Given-When-Then behavior coverage. |
Four phases — start-of-day sync, planning, wrap-up, and session housekeeping — plus one orchestrator that ties wrap-up together.
SYNC PLANNING IMPLEMENTATION WRAP-UP
┌────────┐ ┌────────────┐ ┌──────────────────┐
│ sync │──►│ discuss │ ───► (you write the code) ───► │ wrap-up │
└────────┘ └────────────┘ └────────┬─────────┘
│ orchestrates
┌────────────────────┼──────────────┐
▼ ▼ ▼
verify update-docs code-review
│ │ │
└─────────────► report ◄────────┘
│
▼
rename-session
│
▼
sync-report (if Notion-configured)
│
▼
git commit (optional)
│
▼
deploy (optional)
| Skill | Phase | Purpose |
|---|---|---|
| sync | start-of-day | Safety net for git-illiterate collaborators. Fetches, auto-merges upstream (current branch + origin/main + teammate branches updated in last 24h), pushes. Dirty tree → defers to /wrap-up Quick. Any merge conflict aborts cleanly. |
| discuss | planning | Senior design/architecture advisor that drives a collaborative back-and-forth before any plan is written. Auto-invoked during plan mode. |
| verify | wrap-up | Executes the implementation plan's Test plan / Verification section. Runner, not author — does not write new tests. |
| update-docs | wrap-up | Detection-driven docs updater. Scans the project's doc layout, classifies the diff, and proposes per-file edits. |
| code-review | wrap-up | Multi-agent review of the local git diff. Writes a single REVIEW.md with findings tiered Critical/Warning/Suggestion/Nit (drops only auto-zeroed false positives). |
| report | wrap-up | Distills the conversation, file changes, and decisions into docs/reports/YYYY-MM-DD-[title].md. /report transcript exports the raw conversation instead, listing the files read/modified. |
| rename-session | housekeeping | Renames the current Claude Code or Codex session with a short title. Auto-invoked after runs. |
How to use:
/sync first. It pulls the latest from origin/main + teammates' recent branches into your current branch so you start from the freshest state. If your tree is dirty, /sync will offer to run /wrap-up Quick to save first./discuss (or just enter plan mode — discuss auto-invokes) to surface design alternatives before writing code./wrap-up. Picks Full or Quick mode. Full chains the full wrap-up steps; Quick ships report + safe commit + push only (for collaborators with low git literacy or anyone short on time)./verify, /update-docs, /code-review, /report, /rename-session — and wrap-up will skip steps you've already run earlier in the session if nothing relevant changed.wrap-up: /verify → /update-docs → /code-review → fix Critical/Warning findings (Suggestion/Nit are advisory) → /report → /rename-session → /sync-report → commit → deploy.These skills bridge between local code/docs and a shared Notion workspace (Roadmap DB + Meetings DB). Notion is master for task and meeting state; local files are working copies.
┌────────────────┐
│ setup-notion │ (one-time, writes config to AGENTS.md)
└────────────────┘
MEETING FLOW TASK FLOW
──────────── ─────────
transcript / Plaud URL ┌──► Notion task
│ │ │
▼ /meeting-notes │ ▼ /fetch-task
docs/meetings/YYYY-MM-DD-*.md │ docs/tasks/<slug>.md
│ │ │
▼ /upload-meeting │ (you work)
Notion Meetings DB page │ │
│ │ ▼ /report
▼ /create-tasks │ docs/reports/YYYY-MM-DD-*.md
extracts commitments → Notion tasks ───────┘ │
(links back to meeting) ▼ /sync-report
appends to linked Notion task
| Skill | Direction | Purpose |
|---|---|---|
| setup-notion | config | One-time wizard. Records Roadmap DB URL, Meetings DB URL, and optional team roster in AGENTS.md; adds docs/tasks/ to .gitignore. Idempotent. |
| fetch-task | Notion → local | Pulls a Notion task into docs/tasks/<slug>.md. Read-only against Notion. Refreshes only the managed # Context section on re-run. With no args, picks from your active tasks. |
| sync-report | local → Notion | Pushes docs/reports/*.md to a Notion task — appends body, suggests Status, fills Github Link from HEAD. Linked reports fast-path; unlinked reports prompt a candidate picker. |
| meeting-notes | transcript → local | Turns a transcript (Plaud.ai share URL, local file, or pasted text) into a structured Traditional-Chinese meeting note at docs/meetings/YYYY-MM-DD-<title>.md. |
| upload-meeting | local → Notion | Uploads a meeting markdown to the configured Notion Meetings DB. Auto-chained after /meeting-notes. |
| create-tasks | local → Notion | Extracts commitments from a meeting note, classifies LINK / DRAFT / SKIP, pre-matches against the Roadmap's Current view, then creates new tasks + wires up the meeting's Tasks relation on approval. |
Authority model: Only /sync-report and /create-tasks write to Notion. /fetch-task, /setup-notion, /meeting-notes are read-only against Notion (setup writes to local config only). /upload-meeting writes one Meetings DB page.
How to use:
/setup-notion once. It writes Roadmap + Meetings DB URLs into AGENTS.md and gitignores docs/tasks/. Skipped silently on re-run./fetch-task <url> to pull it into docs/tasks/<slug>.md. With no args, you get a picker of your active tasks. Re-run any time to refresh the # Context section without touching your notes./wrap-up automatically chains /sync-report at step 6 — appends your report to the linked Notion task and suggests a Status. To run it standalone: /sync-report after /report./meeting-notes <plaud-url | file | paste> synthesizes a structured note. It then auto-chains /upload-meeting (push to Meetings DB) and offers /create-tasks (extract commitments → Notion tasks, linked back to the meeting).| Skill | Purpose |
|---|---|
| find-session | Search Claude Code or Codex sessions by topic or file-change evidence; --open refreshes/opens the log. Searches both by default; --host filters the source. |
| headless-shots | Screenshot or drive a locally running web app via headless Chrome + CDP with a minted login cookie — fallback when the Playwright MCP browser is locked; write-path checks verified against the DB. |
dev-skills/
├── .claude-plugin/
│ ├── plugin.json
│ └── marketplace.json
├── skills/
│ └── <skill-name>/
│ ├── SKILL.md
│ └── agents/ (optional)
│ └── <agent-name>.md
├── AGENTS.md
├── CLAUDE.md → AGENTS.md (symlink)
└── README.md
From GitHub:
/plugin marketplace add dropout-tech/dev-skills
/plugin install dev-skills
The first command registers this repo's marketplace.json as a plugin source; the second installs the dev-skills plugin from that marketplace. After install, restart Claude Code to load the skills.
For local development against a clone:
/plugin marketplace add ~/dev-skills
/plugin install dev-skills@dev-skills
Session operations use one cross-host entry point: bin/session-adapter.py
provides show, rename, export, and Codex log, then delegates to the Claude Code or
Codex backend. bin/codex-session.py remains as a compatibility shim.
bin/session-adapter.py log --open refreshes and opens a derived .log.md for
this Codex session. --watch refreshes while running; configured lifecycle
hooks refresh automatically after /hooks trust review. See
Codex session log hooks. skills/find-session/scripts/search.py --host codex searches
persisted App Server history and --open-id current opens the current log.
session-log.py also routes show/note/writes/export to Codex when its thread ID
is exposed. Write evidence covers completed fileChange events only, not shell
or MCP writes; the compatibility export applies credential-pattern redaction.
Claude history remains available through --host claude.
In shared skill instructions, Claude Code's AskUserQuestion maps to Codex's
request_user_input whenever that tool is exposed by the current host. Use the
host's approval mechanism for shell permissions; fall back to a concise chat
question only when the structured input tool is unavailable.
Link each skill from the clone (<clone>/skills/<name>) into ~/.codex/skills/;
edits are available through those links without copying or reinstalling. Invoke a skill
with $wrap-up, $verify, or its name in your request.
Keep the entire clone for skills that use bin/ helpers. Session operations
use bin/session-adapter.py to resolve the host, identity, and backend in code.
Skills describe their own task-specific commands; they do not load a mandatory
compatibility preamble. The runtime reference is
optional maintainer documentation for troubleshooting host differences.
The shared workflow supports both Claude Code and Codex. Codex uses its own
available tools, approval UI, inherited review-agent model, and session identity.
Claude-specific argument-hint frontmatter has been moved into the skill body
as Arguments so both hosts can parse the same metadata.
| Capability | Codex behavior |
|---|---|
| Planning, verification, reports, reviews | Shared workflow with host tool mappings |
| Notion operations | Require an available, authenticated connection; use its actual schemas |
| Session search | Searches Claude + Codex by default; --host codex or --host claude restricts the source |
| Session rename | bin/session-adapter.py rename selects the host and verifies the saved title |
| Raw transcript export | bin/session-adapter.py export selects the host exporter; Codex pages persisted turns |
| Working log | session-adapter.py log --open refreshes and opens a persistent .log.md; --watch polls while running; local notes survive refresh |
| Write attribution | Completed fileChange events only; shell writes and Claude hunk snapshots are not inferred |
| Session-log hooks | Four user hooks configured locally; /hooks trust review is required before they run. See hook setup |
Install all skills from this repo (user-level / global):
gemini skills install https://github.com/dropout-tech/dev-skills.git --path skills
Project-level install:
gemini skills install https://github.com/dropout-tech/dev-skills.git --path skills --scope workspace
Or, if you've cloned the repo locally, link it for live updates:
gemini skills link ~/dev-skills/skills
Verify with gemini skills list, or /skills reload inside an active session.
The gh skill command can install individual skills from this repo into any supported agent:
gh skill install dropout-tech/dev-skills <skill-name> --agent claude-code --scope user
gh skill install dropout-tech/dev-skills <skill-name> --agent gemini-cli --scope user
gh skill install dropout-tech/dev-skills <skill-name> # defaults to github-copilot, project scope
<skill-name> is one of: code-review, create-tasks, discuss, fetch-task, find-session, meeting-notes, rename-session, report, setup-notion, spec, spec-review, sync, sync-report, update-docs, upload-meeting, verify, wrap-up. See gh skill install for the full list of supported agents and flags.
Each skills/<name>/ folder follows the standard Agent Skills layout (a SKILL.md with YAML frontmatter). Drop a skill folder into whichever directory your agent reads from:
~/.claude/skills/ (user) or .claude/skills/ (project)~/.gemini/skills/ or ~/.agents/skills/.gemini/skills/ or .agents/skills/~/.copilot/skills/ or ~/.agents/skills/ or ~/.claude/skills/.github/skills/ or .agents/skills/ or .claude/skills/report| wrap-up | orchestrator | Two modes. Full runs verify → update-docs → code-review → report → rename-session → cleanup → commit → deploy → improve for the primary developer. Quick skips quality/deploy gates and ships report + safe commit + push only — for time-constrained or git-illiterate collaborators, or when auto-invoked from /sync. |