사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/islo-labs/skills --skill factory-lines시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Islo Factories
A Factory contains Machines, reusable Agents, and Lines. Agents are the execution units for new Lines. A Line connects Agent interactions and routes their results. Managers route a Line between stages.
Log in before Line creation
Before starting a Line creation workflow, verify that the Islo CLI is logged in with this read-only check:
islo factory list --output jsonOn success, including an empty array, continue without asking the user to log in. Only for missing or invalid authentication, ask the user to run islo login in their local terminal outside a sandbox. Wait for confirmation and recheck before inventory, design, or resource creation. Diagnose other failures without requesting login. Provider connections still use the integration cards below.
Setup contract
For a request to build or substantially change a Factory, follow factory-design.md before create-a-line.md. Inventory with read-only commands first. Distinguish observed resources, inferred defaults, and decisions still needed from the user. Ask one batch containing only unresolved decisions, then present the blueprint as a Line draft card. Wait for the user to write "approve" for that card before any create, update, deploy, provider event, or external write.
Keep the discussion in product terms: Factory, Machines, Agents, their interactions, shared workspaces, semantic inputs and outputs, integrations, side effects, and verification. Choose a Script Agent for deterministic commands, a Decision Agent for typed choice or score questions, and a Prompt Agent for open-ended judgment or tool use. Build new Lines from Agent stages only. Leave existing Job stages in place when changing a Line, and do not add new ones. Load Job references only when the user explicitly asks to maintain an existing Job or job.toml.
Never write a manifest from memory. Check the installed islo schema factory --short and islo schema agent; validate before deploy. If these schemas cannot express a required capability, or the installed CLI lacks a required command, report the missing capability and stop the dependent design and deployment. Do not author a Job, consult the Job schema for a fallback, or propose an implementation workaround. Do not infer support from a similarly named field or ask the user to solve the implementation gap.
Integration connections
Before creating or deploying resources, check providers with the integrations skill. Follow the chat connection gate for connect cards and waiting.
Chat links
Always use explicit /factories/FACTORY_KEY/... routes for Factory objects. Resolve the object's Factory key with islo factory current --factory KEY_OR_UUID --output json and read factory.key; use the same selection as the object operation. Never emit legacy singular-Factory run URLs or guess a Factory from the current conversation. If the key cannot be resolved, keep a known explicit link as a badge with view=chip, rather than requesting a card. If no explicit URL is known, describe the run in text until its Factory is identified. Follow chat-embeds.md for views and filters.
Waiting for line runs you start
Read and follow Waiting for line runs you start whenever you start a run for a chat request. Wait through Manager recovery, read the final outputs, and continue the original task in the same turn.
References
When authorized to ask Line-run questions, send choice question objects with 1 to 8 explicit options and prompts of 30 words or fewer. One option is enough because every choice includes Other for a custom answer. Follow questions.md for the request format.
| Need | Read |
|---|---|
| Discover resources, design a Factory, get approval | factory-design.md |
| Create resources and verify a Line | create-a-line.md |
| Machine lifecycle and readiness | machines.md |
| Agent, Manager, harness, model, Skills, or MCP | agents-and-managers.md |
| Line manifest, triggers, or run debugging | line-manifest.md, triggers.md, run-control-and-debugging.md |
| Show an outcome card, Line, run, question, Machine, environment, Job, or Knowledge item in Agent chat | chat-embeds.md |
Only when explicitly asked to maintain an existing Job or job.toml | harness-models-knowledge.md, job-manifest.md, and standalone-jobs.md |
Use the platform skill for sandbox and snapshot lifecycle, environments, gateway profiles, provider connections, incoming webhooks, and Knowledge CRUD. Treat nearby examples as shape references; never copy their repo, selector, destination, or gateway identity into a new Factory.
scripts/validate_commands.sh and scripts/validate_examples.sh validate documented commands and manifest examples against the installed CLI.