Skip to main content
Run any Skill in Manus
with one click
GitHub repository

ai-roots

ai-roots contains 22 collected skills from siner308, with repository-level occupation coverage and site-owned skill detail pages.

skills collected
22
Stars
2
updated
2026-07-07
Forks
1
Occupation coverage
4 occupation categories ยท 100% classified
repository explorer

Skills in this repository

model-effort-delegation
computer-occupations-all-other

executor(๋ฉ”์ธ ์„ธ์…˜ vs subagent vs team), ๋ชจ๋ธ(Opus/Sonnet/Haiku), effort ๋ ˆ๋ฒจ ์ค‘ ๋ฌด์—‡์ด ์ž‘์—…์— ๋งž๋Š”์ง€ ์ •ํ•  ๋•Œ ์ ์šฉ โ€” ์ฆ‰ ์‚ฌ์†Œํ•˜์ง€ ์•Š์€ ์ž‘์—…์„ ์œ„์ž„ํ•˜๊ธฐ ์ „์—. ๋‹ค์šด๊ทธ๋ ˆ์ด๋“œ ์—„๊ฒฉ ์กฐ๊ฑด(plan ์ •๋ฐ€๋„ + ๊ฒ€์ฆ ๋ฃจํ”„), ์—์Šค์ปฌ๋ ˆ์ด์…˜ ํŠธ๋ฆฌ๊ฑฐ, blast-radius ์˜ค๋ฒ„๋ผ์ด๋“œ, subagent ๋ธŒ๋ฆฌํ•‘ ๊ธฐ์ค€์„ ๋‹ค๋ฃฌ๋‹ค.

2026-07-07
model-effort-delegation
computer-occupations-all-other

Apply when deciding which executor (main session vs subagent vs team), which model (Opus/Sonnet/Haiku), and what effort level fits a task โ€” i.e. before delegating non-trivial work. Covers strict downgrade conditions (plan precision + verification loop), escalation triggers, blast-radius override, and the subagent briefing standard.

2026-07-07
background-task-monitoring
software-developers

์žฅ์‹œ๊ฐ„ ์ž‘์—…์ด ๋ฐฑ๊ทธ๋ผ์šด๋“œ์—์„œ ๋Œ๊ณ  ์‚ฌ์šฉ์ž๊ฐ€ ์™„๋ฃŒ๋‚˜ ์ง„ํ–‰ ์ƒํ™ฉ์„ ๋ด์•ผ ํ•  ๋•Œ ์ ์šฉ โ€” ์™„๋ฃŒ ์•Œ๋ฆผ๋งŒ(๊ธฐ๋ณธ), Monitor๋ฅผ ํ†ตํ•œ ์ด๋ฒคํŠธ ์ŠคํŠธ๋ฆผ, ScheduleWakeup์„ ํ†ตํ•œ ์ธํ„ฐ๋ฒŒ ํด๋ง ์ค‘ ์„ ํƒ. ์„œ๋ธŒํ”„๋กœ์„ธ์Šค๋ฅผ tmux split pane, sentinel ๋ฌธ์ž์—ด, foreground tail/grep ๋ฃจํ”„๋กœ ๊ฐ์‹œํ•˜๊ณ  ์‹ถ์–ด์งˆ ๋•Œ๋„ ์ ์šฉ โ€” ๊ทธ ํŒจํ„ด์€ ์‹คํŒจํ–ˆ๊ณ  post-mortem์ด ์—ฌ๊ธฐ ์žˆ๋‹ค. ๊ฒฐ์ • ์‚ฌ๋‹ค๋ฆฌ, ์•ˆํ‹ฐํŒจํ„ด, tee ๊ธฐ๋ฐ˜ ์‚ฌ์šฉ์ž ๊ฐ€์‹œ ์ŠคํŠธ๋ฆฌ๋ฐ์„ ๋‹ค๋ฃฌ๋‹ค.

2026-07-07
web-research
market-research-analysts-and-marketing-specialists-131161

์›น์„ ๋‘˜๋Ÿฌ๋ณด๊ฑฐ๋‚˜, ํŽ˜์ด์ง€ ๋‚ด์šฉ์„ ์ถ”์ถœํ•˜๊ฑฐ๋‚˜, ๋ฐ์ดํ„ฐ๋ฅผ ๊ธ๊ฑฐ๋‚˜, ์‚ฌ์ดํŠธ์—์„œ ์ˆ˜์น˜๋ฅผ ๊ฐ€์ ธ์˜ฌ ๋•Œ ์ ์šฉํ•˜์„ธ์š” โ€” agent-browser(๊ธฐ๋ณธ lightpanda, ๋ฌด๊ฒ๊ฑฐ๋‚˜ ์ƒํ˜ธ์ž‘์šฉ์ด ํ•„์š”ํ•œ ํŽ˜์ด์ง€๋Š” chrome), ๋ฐœ๊ฒฌ์šฉ WebSearch, ์ฐจ๋‹จ ๋ณธ๋ฌธ ํšŒ์ˆ˜์šฉ insane-search ์Šคํ‚ฌ ์ค‘์—์„œ ๊ณ ๋ฆ…๋‹ˆ๋‹ค. agent-browser๊ฐ€ ์ฐจ๋‹จ/๋นˆ/๋™์  ์ฝ˜ํ…์ธ ๋ฅผ ๋Œ๋ ค์ค˜์„œ ๋‹ค๋ฅธ ์—”์ง„์œผ๋กœ ์žฌ์‹œ๋„ํ•˜๊ฑฐ๋‚˜ ํ˜•์ œ URL์„ ์ถ”์ธกํ•˜๊ณ  ์‹ถ์–ด์ง€๋Š” ์ˆœ๊ฐ„์—๋„ ์ ์šฉ โ€” ๋ธŒ๋ผ์šฐ์ € ์—์Šค์ปฌ๋ ˆ์ด์…˜์„ ๋ฉˆ์ถ”๊ณ  ๊ฒ€์ƒ‰ ๊ธฐ๋ฐ˜ ๋„๊ตฌ๋กœ ์ „ํ™˜ํ•˜์„ธ์š”. Codex --search๋Š” ๊ต์ฐจ๊ฒ€์ฆ ์ „์šฉ GPT ์˜์กด ์ตœํ›„ ์ˆ˜๋‹จ์ž…๋‹ˆ๋‹ค.

2026-07-07
background-task-monitoring
software-developers

Apply when a long-running task runs in the background and the user needs visibility into its completion or progress โ€” choosing between completion-notification-only (default), streamed events via Monitor, and interval polling via ScheduleWakeup. Also apply when tempted to monitor a subprocess via tmux split panes, sentinel strings, or a foreground tail/grep loop โ€” that pattern failed and its post-mortem lives here. Covers the decision ladder, anti-patterns, and tee-based user-visible streaming.

2026-07-07
web-research
market-research-analysts-and-marketing-specialists-131161

Apply when browsing the web, extracting page content, scraping data, or fetching figures from sites โ€” choosing between agent-browser (lightpanda default, chrome for heavy/interactive pages), WebSearch for discovery, and the insane-search skill for retrieving blocked content. Also apply the moment agent-browser returns blocked/empty/dynamic content and you are tempted to retry with another engine or guess sibling URLs โ€” stop escalating the browser and switch to a search-based tool. Codex --search is the GPT-dependent last resort for cross-provider verification only.

2026-07-07
codex-delegation
computer-occupations-all-other

OpenAI Codex CLI์— ์ž‘์—…์„ ์œ„์ž„ํ•  ๋•Œ ์ ์šฉ โ€” Claude๊ฐ€ ๋ง‰ํžŒ ๋’ค์˜ rescue ๋””๋ฒ„๊น…(three-turn ์ƒํ•œ), cross-provider/๋ณด์•ˆ ๋ฏผ๊ฐ ๋ฆฌ๋ทฐ, ์ตœ์‹  docs ์›น ๋ฆฌ์„œ์น˜, ๋˜๋Š” ๋ฒ”์œ„๊ฐ€ ์ •ํ•ด์ง„/๋ฌด์ธ ๊ตฌํ˜„. ๋ชจ๋“œ์™€ ํ”Œ๋ž˜๊ทธ ์„ ํƒ, reasoning effort, rescue protocol, plan-stage review, ์‹คํ–‰ ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ๋‹ค๋ฃฌ๋‹ค. Codex CLI๊ฐ€ PATH์— ์žˆ์„ ๋•Œ๋งŒ ํŠธ๋ฆฌ๊ฑฐํ•  ๊ฒƒ.

2026-07-07
incremental-verification
software-developers

์ž‘์—… ๊ฒฐ๊ณผ๊ฐ€ ๋ถˆํ™•์‹คํ•  ๋•Œ ์ ์šฉํ•˜์„ธ์š” โ€” ์™ธ๋ถ€ API, ๋ธŒ๋ผ์šฐ์ € ์ž๋™ํ™”, ์…ธ ์ด์Šค์ผ€์ดํ”„, ์ต์ˆ™์ง€ ์•Š์€ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ, ๋ฐ์ดํ„ฐ ํŒŒ์ดํ”„๋ผ์ธ. ์ „์ฒด ์Šคํฌ๋ฆฝํŠธ๋ฅผ ์งœ์„œ ๋Œ€๊ทœ๋ชจ๋กœ ํ•œ ๋ฒˆ์— ๋Œ๋ฆฌ์ง€ ๋ง๊ณ , ๊ฐ€์žฅ ์ž‘์€ ๊ฒ€์ฆ ๋‹จ์œ„๋กœ ์ชผ๊ฐœ ๊ฐ ๋‹จ๊ณ„์˜ ์‹ค์ œ ์ถœ๋ ฅ์„ ํ™•์ธํ•œ ๋’ค ๋‹ค์Œ ๋‹จ๊ณ„๋ฅผ ์Œ“์œผ์„ธ์š”.

2026-07-07
parallel-execution-modes
computer-occupations-all-other

์ž‘์—…์„ ๋ณ‘๋ ฌํ™”ํ•  ์ˆ˜ ์žˆ์„ ๋•Œ ์ ์šฉ โ€” sequential, subagent, team ์ค‘ ์„ ํƒ; inline vs subagent ๊ฒฐ์ •; foreground vs background ์‹คํ–‰. ์—์ด์ „ํŠธ๋ฅผ ๋„์šฐ๊ธฐ ์ „์—, ๋˜๋Š” ์ž‘์—…์ด ํŒŒ์ผ 3๊ฐœ ์ด์ƒ ๋˜๋Š” ๋…๋ฆฝ ์„œ๋ธŒํƒœ์Šคํฌ 2๊ฐœ ์ด์ƒ์œผ๋กœ ํผ์งˆ ๋•Œ ์‚ฌ์šฉํ•œ๋‹ค.

2026-07-07
parallel-hypothesis-investigation
software-developers

๋ฌธ์ œ์˜ ์›์ธ์ด ์—ฌ๋Ÿฌ ๊ณ„์ธต์— ๊ฑธ์ณ ๊ทธ๋Ÿด๋“ฏํ•˜๊ฒŒ ๋‚˜๋‰˜์–ด ์žˆ๊ฑฐ๋‚˜, ์ถœ๋ ฅ์ด ์„œ๋กœ ๋…๋ฆฝ์ ์ธ ์—ฌ๋Ÿฌ ํŒ๋‹จ ๊ธฐ์ค€์„ ๋ชจ๋‘ ํ†ต๊ณผํ•ด์•ผ ํ•  ๋•Œ ์ ์šฉํ•˜์„ธ์š” โ€” ๊ฐ€์„ค์„ ์กฐ์‚ฌํ•˜๊ฑฐ๋‚˜ ๊ธฐ์ค€์„ ํ‰๊ฐ€ํ•  ๋•Œ ์ˆœ์ฐจ๋กœ ํ•˜์ง€ ๋ง๊ณ  ๋™์‹œ์— ํ•˜๋ผ๋Š” ๊ฒ๋‹ˆ๋‹ค. ํŠนํžˆ ๋ป”ํ•œ ์ˆ˜์ •์ด ์ด๋ฏธ ์‹คํŒจํ–ˆ๊ฑฐ๋‚˜, ๋ฌธ์ œ๊ฐ€ ๊ฒฝ๊ณ„๋ฅผ ๋„˜๋‚˜๋“ค ๋•Œ(config๋Š” ์„ค์ •๋ผ ์žˆ๋Š”๋ฐ ๋™์ž‘์€ ํ‹€๋ฆฐ ๊ฒฝ์šฐ) ์“ฐ์„ธ์š”.

2026-07-07
simulate-dont-just-scan
software-developers

์ฝ๊ธฐ๋Š” ํ–ˆ์ง€๋งŒ ์‹คํ–‰ํ•ด ๋ณด์ง€ ์•Š์€ ์ฝ”๋“œ๋ฅผ ํฌํŒ…ยท๋””๋ฒ„๊น…ยท๊ตฌํ˜„ํ•  ๋•Œ ์ ์šฉํ•œ๋‹ค โ€” ์†Œ์Šค๋ฅผ ์ฝ์€ ๊ฒƒ์ด ๋™์ž‘์„ ์ดํ•ดํ•œ ๊ฒƒ์ด๋ผ๊ณ  ๊ฐ€์ •ํ•˜์ง€ ๋ง๊ณ , ๊ตฌ์ฒด์ ์ธ ์ž…๋ ฅ์œผ๋กœ ์ฝ”๋“œ๋ฅผ ๋จธ๋ฆฟ์†์—์„œ ๋Œ๋ ค ์‹ค์ œ ๋Ÿฐํƒ€์ž„ ์ถœ๋ ฅยทDOMยท์‘๋‹ตยท๋ฐ์ดํ„ฐ ๋ชจ์–‘ยท๋ถ€์ˆ˜ํšจ๊ณผ๋ฅผ ์˜ˆ์ธกํ•œ๋‹ค.

2026-07-07
codex-delegation
computer-occupations-all-other

Apply when delegating work to the OpenAI Codex CLI โ€” rescue debugging after Claude is stuck (three-turn cap), cross-provider/security-sensitive review, current-docs web research, or bounded/unattended implementation. Covers mode and flag selection, reasoning effort, the rescue protocol, plan-stage review, and execution mechanics. Trigger only when Codex CLI is on PATH.

2026-07-07
css-discipline
software-developers

Apply when editing, writing, or reviewing CSS โ€” stylesheets, scoped styles, inline style attributes, CSS-in-JS, or any framework styling (Tailwind, CSS Modules, Svelte/Vue scoped styles). Covers the closed spacing scale, box-model, cascade/specificity, overflow, flex/grid shrink, color/z-index tokens, and unit discipline. Do not trigger for tasks that do not touch styles.

2026-07-07
incremental-verification
software-developers

Apply when a task outcome is uncertain โ€” external APIs, browser automation, shell escaping, unfamiliar libraries, or data pipelines. Break the work into the smallest verifiable steps and confirm each step's actual output before building the next, instead of writing a full script and running it at scale.

2026-07-07
parallel-execution-modes
software-developers

Apply when work can be parallelized โ€” choosing between sequential, subagents, and teams; deciding inline vs subagent; and foreground vs background execution. Use before spawning agents, or when a task fans out across three or more files or two or more independent subtasks.

2026-07-07
parallel-hypothesis-investigation
software-developers

Apply when a problem has multiple plausible causes across system layers, or output must pass multiple independent judgment criteria โ€” to investigate hypotheses or evaluate criteria concurrently rather than sequentially. Use especially when the obvious fix already failed or the problem spans boundaries (config set but behavior wrong).

2026-07-07
simulate-dont-just-scan
software-developers

Apply when porting, debugging, or implementing against code you have read but not executed โ€” to mentally run the code with concrete inputs and predict the actual runtime output, DOM, response, data shape, or side effect, rather than assuming that reading the source equals understanding its behavior.

2026-07-07
github-pr-markdown
software-developers

pull request ๋ณธ๋ฌธ์ด๋‚˜ ์ œ๋ชฉ์„ ๋งŒ๋“ค๊ฑฐ๋‚˜ ์ˆ˜์ •ํ•  ๋•Œ(gh pr create, gh pr edit, gh api PR ์—…๋ฐ์ดํŠธ), ๋˜๋Š” PR ์„ค๋ช…์„ ์ž‘์„ฑํ•  ๋•Œ โ€” ๋‹ค๋ฅธ ์Šคํ‚ฌ์ด๋‚˜ ์›Œํฌํ”Œ๋กœ๊ฐ€ PR ์„ค๋ช…์„ ๋งŒ๋“ค์–ด๋‚ด๋Š” ๊ฒฝ์šฐ๋„ ํฌํ•จ โ€” ์ ์šฉํ•œ๋‹ค. ๋ณธ๋ฌธ์€ ์ตœ๋Œ€ 3์ค„ โ€” ๊ธฐ๋ณธ๊ฐ’์ด ์•„๋‹ˆ๋ผ ํ•˜๋“œ ์บก์ด๋ฉฐ, ์œ ์ผํ•œ ์˜ˆ์™ธ๋Š” ๋ ˆํฌ์˜ PR ํ…œํ”Œ๋ฆฟ์ด๋‹ค. GitHub-flavored Markdown์„ ๊ฐ•์ œํ•œ๋‹ค: ASCII ๋Œ€์‹œ ๋ถˆ๋ฆฟ, task-list ์ฒดํฌ๋ฐ•์Šค, ๋ฐฑํ‹ฑ ์ฝ”๋“œ ์ฐธ์กฐ, ๊ทธ๋ฆฌ๊ณ  gh CLI ๋งˆํฌ๋‹ค์šด ์†์ƒ์„ ํ”ผํ•˜๋Š” ์•ˆ์ „ํ•œ API-PATCH ๋ณธ๋ฌธ ์ „๋‹ฌ.

2026-07-06
github-pr-markdown
software-developers

Apply when creating or editing a pull request body or title (gh pr create, gh pr edit, gh api PR updates) or composing any PR description, including when another skill or workflow produces one. The body is capped at three lines โ€” a hard ceiling, not a default โ€” with the repo's PR template as the only exception. Enforces GitHub-flavored Markdown: ASCII dash bullets, task-list checkboxes, backtick code refs, and the safe API-PATCH body delivery that avoids gh CLI markdown corruption.

2026-07-06
review
software-quality-assurance-analysts-and-testers

[ai-roots] ๋ฌด์—‡์ด๋“  ๋ฆฌ๋ทฐํ•˜๋Š” ๋‘ ํ‰๊ฐ€์ž ๋ฆฌ๋ทฐ โ€” ์ฝ”๋“œ ๋ณ€๊ฒฝ, Claude๊ฐ€ ๋ฐฉ๊ธˆ ๋งŒ๋“  planยท์„ค๊ณ„, ๋ฌธ์„œ, config, ๊ทธ ์™ธ ๋ฆฌ๋ทฐ ๊ฐ€๋Šฅํ•œ ๋ฌด์—‡์ด๋“ . ์ž์—ฐ์–ด ์š”์ฒญ์—์„œ ์‚ฐ์ถœ๋ฌผ ์ข…๋ฅ˜๋ฅผ ํŒ๋ณ„ํ•˜๊ณ , ๋‘ ํ‰๊ฐ€์ž๊ฐ€ ๋™์ผํ•˜๊ฒŒ ์ ‘๊ทผํ•  ํ•˜๋‚˜์˜ ๊ตฌ์ฒด์  ์‚ฐ์ถœ๋ฌผ(git diff, ํŒŒ์ผ, ๋˜๋Š” temp ํŒŒ์ผ๋กœ ๊ณ ์ •ํ•œ ์ธ๋ผ์ธ ํ…์ŠคํŠธ)๋กœ ๊ฒฐ์ •ํ•œ ๋’ค, ๊ทธ ๋™์ผํ•œ ์‚ฐ์ถœ๋ฌผ์— ๋Œ€ํ•ด Claude Code subagent์™€ Codex๋ฅผ ์ข…๋ฅ˜๋ณ„ ๊ธฐ์ค€์œผ๋กœ ๋ณ‘๋ ฌ ์‹คํ–‰ํ•˜๊ณ  rules/evaluation-integrity.md ยงMulti-advisor synthesis์˜ Agreed / Conflicting / Chosen-direction ํฌ๋งท์œผ๋กœ ์ข…ํ•ฉํ•œ๋‹ค. ๋Œ€๊ธฐ ์ค‘์ธ ์ฝ”๋“œ, plan, ๋ฌธ์„œ, ๊ทธ ์™ธ ์‚ฐ์ถœ๋ฌผ์˜ ๋ฆฌ๋ทฐ๋ฅผ ์š”์ฒญํ•˜๊ณ  Claude์™€ Codex๊ฐ€ ๋‘˜ ๋‹ค ์žˆ์„ ๋•Œ ์‚ฌ์šฉํ•œ๋‹ค.

2026-06-17
review
software-quality-assurance-analysts-and-testers

[ai-roots] Two-evaluator review of any artifact โ€” code changes, a plan or design Claude just produced, a document, a config, or anything reviewable. From a natural-language request it determines the artifact kind, resolves ONE concrete shared artifact (a git diff, a file, or inline text captured to a temp file), then spawns a Claude Code subagent and a Codex run in parallel on that same artifact with kind-appropriate criteria, and synthesizes findings using the Agreed / Conflicting / Chosen-direction format from rules/evaluation-integrity.md ยงMulti-advisor synthesis. Use whenever the user asks to review pending code, a plan, a document, or any other artifact and both Claude and Codex are available.

2026-06-17
css-discipline
software-developers

CSS๋ฅผ ํŽธ์ง‘ยท์ž‘์„ฑยท๋ฆฌ๋ทฐํ•  ๋•Œ ์ ์šฉ โ€” ์Šคํƒ€์ผ์‹œํŠธ, scoped ์Šคํƒ€์ผ, ์ธ๋ผ์ธ style ์†์„ฑ, CSS-in-JS, ๋˜๋Š” ํ”„๋ ˆ์ž„์›Œํฌ ์Šคํƒ€์ผ๋ง(Tailwind, CSS Modules, Svelte/Vue scoped ์Šคํƒ€์ผ) ์ „๋ฐ˜. ๋‹ซํžŒ spacing scale, ๋ฐ•์Šค ๋ชจ๋ธ, ์บ์Šค์ผ€์ด๋“œ/specificity, overflow, flex/grid shrink, color/z-index ํ† ํฐ, ๋‹จ์œ„ ๊ทœ์œจ์„ ๋‹ค๋ฃฌ๋‹ค. ์Šคํƒ€์ผ์„ ๊ฑด๋“œ๋ฆฌ์ง€ ์•Š๋Š” ์ž‘์—…์—์„œ๋Š” ํŠธ๋ฆฌ๊ฑฐํ•˜์ง€ ๋ง ๊ฒƒ.

2026-06-02