# Licell > Query and manage Alibaba Cloud through curated workflows and all protocol-covered Alibaba Cloud APIs. Routes natural-language intent through catalog/help first, then capability discovery and API invoke fallback. - Skill: `agents-infrastructure/licell` (Agent Skill) - Install (CLI): `npx skillmds@latest add agents-infrastructure/licell` - Raw SKILL.md: https://api.skillmd.com/api/skills/agents-infrastructure/licell/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: agents-infrastructure (https://skillmd.com/u/agents-infrastructure) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/agents-infrastructure/licell --- # licell CLI Skill Use the `licell` CLI as a structured deployment and operations tool for Alibaba Cloud. ## Scope - This file teaches an agent how to drive `licell` safely. - It is not the canonical command reference. - If a detail can be discovered from CLI help or catalog output, prefer the CLI over this file. ## Operating Contract - Do not guess command names, flags, argument order, or result fields. - Discover commands with `licell catalog --output json`. - Read per-command usage with `licell --help --output json`. - Execute real work with `licell --output json`. - For streamed output, parse only lines prefixed with `@@LICELL_JSON@@`. - Prefer structured fields like `nextActions[]`, `result`, `error`, and `details` over human-readable console text. - If the task is mutating or destructive, inspect help first and follow the command's structured guidance. - Verify `kind` before trusting a record shape, and then verify `schemaVersion`. - Do not scrape plain-text terminal output when `--output json` is available. - Do not assume one command's result shape applies to another; read the command's help contract first. - If the user requires Licell-only execution and both curated commands and protocol capabilities are missing, report the capability gap. Do not inspect Licell package internals or credential files, and do not write ad hoc SDK scripts to bypass Licell. - Never parse or pass `cs.DescribeClusterUserKubeconfig` output to an Agent or `kubectl`; `[REDACTED]` is an intentional security boundary. Use a curated `k8s` command, and report a capability gap when no safe command exists. ## Preconditions - `licell` is installed and on PATH. - Authentication is configured, usually via `licell login`. - If the command operates on a project, run it inside the target repo or initialized workspace. ## Canonical Invocation Sequence ```bash licell catalog --output json licell --help --output json licell --output json ``` Use the first command for discovery, the second for contract lookup, and the third for execution. ## Natural-Language Intent Routing - The Agent owns natural-language understanding. Extract the service/product, action, resource, region, identifiers, and constraints; do not treat the full user sentence as CLI syntax. - Always inspect `licell catalog --output json` first. When a curated command covers the outcome, inspect its help contract and execute it. - If no curated command covers the outcome, run `licell capability products --output json`, then `licell capability search --product --intent "" --action --output json`. - Keep capability search input concise. Use `inspect|create|update|delete|execute` for `--action` and use the resource name for `--intent`. - Describe the selected candidate with `licell capability describe --output json`. Follow `execution.preferred`; do not infer execution from prose or API names. - Do not conclude that Licell cannot handle an Alibaba Cloud request until both the curated catalog and the matching raw protocol capability space have been searched. - Raw reads may execute after required inputs are resolved; raw write operations require `--dry-run` review and explicit `--yes`. - After a mutation, follow `nextActions[]` and perform a read-back verification before reporting success. ## Schema Contracts - Raw CLI JSON output is emitted line-by-line with the `@@LICELL_JSON@@` prefix. Each record currently conforms to `licell-cli-record@1.0`, then branches by `type=event|result|error`. - For `licell --help --output json`, read `help.kind` and `help.schemaVersion`; the current contract is `licell-help@1.0`. - For `licell catalog --output json`, read `kind` and `schemaVersion`; the current catalog contract is `licell-agent-command-catalog@1.0`. - `licell catalog --output json` also declares the help schema and CLI record schema explicitly: `licell-help@1.0` / `licell-cli-record@1.0`. - Read `catalog.agentWorkflow` for the curated-first natural-language routing contract. Only conclude a request is unsupported after curated command discovery and raw capability search both fail. - Agents should prefer `nextActions[]` as the stable next-step surface; `recommendedFlow`, `decisionGuide`, and `remediation[]` are supporting guidance layers. - For command-specific business payloads, keep reading the command help/catalog `result` descriptor; the three sections below only describe the shared CLI record envelope. ### CLI Event Record ยท licell-cli-record@1.0 - Streaming CLI event record for progress tracking, log bridging, and stage-aware automation. - `kind`: Fixed to `licell-cli-record`. - `schemaVersion`: CLI record schema version; currently `1.0`. - `type`: Fixed to `event`. - `ts`: Event timestamp in ISO 8601 format. - `command`: Current command key, such as `deploy` or `oss upload`. - `stage`: Stable stage identifier, such as `deploy`, `deploy.api`, or `auth.restore`. - `action`: Stable action identifier, such as `run`, `execute`, or `stdout`. - `status`: `start` / `ok` / `failed` / `skipped` / `info`. - `source`: `command` / `console` / `stream`. - `terminal`: Whether this event marks the terminal state of the current action. - `ok` (optional): Present only on terminal success/failure events; `true` means success and `false` means failure. - `message` (optional): Human-readable supplemental message. - `data` (optional): Additional structured context object. - `stream` (optional): Stream type when `action=stdout|stderr`. ### CLI Result Record Envelope - Successful CLI result envelope; command-specific payload fields should still be read from the corresponding help/catalog `result` descriptor. - `kind`: Fixed to `licell-cli-record`. - `schemaVersion`: CLI record schema version; currently `1.0`. - `type`: Fixed to `result`. - `ts`: Result timestamp in ISO 8601 format. - `command`: Current command key. - `stage`: Command stage identifier; usually aligned with the command key or sub-stage. - `ok`: Fixed to `true`. ### CLI Error Record - CLI error envelope with compatibility remediation/nextCommands plus the preferred `nextActions` surface. - `kind`: Fixed to `licell-cli-record`. - `schemaVersion`: CLI record schema version; currently `1.0`. - `type`: Fixed to `error`. - `ts`: Error timestamp in ISO 8601 format. - `command`: Current command key. - `stage`: Error stage, such as `parse`, `runtime`, or `deploy`. - `ok`: Fixed to `false`. - `error`: Stable error object. - `code`: Stable error code, such as `CLI_INVALID_INPUT` or `AUTH_MISSING_CREDENTIAL`. - `category`: `auth` / `permission` / `input` / `network` / `quota` / `conflict` / `not_found` / `internal`. - `message`: Primary error message. - `retryable`: Whether the error is suitable for direct retry. - `provider` (optional): Alibaba Cloud provider-side context. - `service` (optional): Cloud product name, such as `fc`, `oss`, or `alidns`. - `action` (optional): Cloud API action name. - `code` (optional): Original cloud-side error code. - `requestId` (optional): Cloud-side requestId. - `httpStatus` (optional): Cloud-side HTTP status code. - `endpoint` (optional): Resolved cloud API endpoint. - `details` (optional): Additional structured error context. - `remediation[]`: Compatibility remediation suggestions. - `type`: Suggestion type, such as `note` or `command`. - `title`: Remediation title. - `reason`: Why this action is recommended. - `commandTemplate`: Suggested command template. - `commandKey` (optional): Stable command key when the command can be matched from the CLI registry. - `commandDescription` (optional): Matched command description. - `phase`: Remediation phase, such as `inspect`, `mutate`, or `verify`. - `priority`: `primary` / `secondary`. - `order`: Stable sort order. - `nextCommands[]`: Compatibility command suggestions. - `commandTemplate`: Suggested command template. - `commandKey` (optional): Stable command key when the command can be matched from the CLI registry. - `description` (optional): Command suggestion description. - `intent`: Command intent, such as `inspect`, `repair`, or `bind`. - `priority`: `primary` / `secondary`. - `nextActions[]`: Preferred unified next-step suggestions. - `title`: Next action title. - `description`: Why this action is recommended. - `commandTemplate`: Suggested command template. - `commandKey` (optional): Stable command key when the command can be matched from the CLI registry. - `phase`: Action phase, such as `inspect`, `verify`, or `mutate`. - `priority`: `primary` / `secondary`. - `source`: Action source, such as `error-remediation`. - When parsing strictly, match `kind` first and then verify `schemaVersion`; if a higher unknown version appears, fall back to a compatibility path instead of assuming the old shape.