ae-system

ae-system

Use the system domain for Agent system administration:

ae-cli system +<command> [options]
ae-cli system <resource> <action

Mandatory Rules

• These commands are only for users whose Agent role is root or agent_admin.
• The te-agent /api/admin/** and /api/cli/channel/v1/** endpoints are the final authorization boundaries. A member receives a permission error (HTTP 403). Do not retry login or attempt a different endpoint after a 403.
• Run ae-cli auth login --host <host> before using this domain. System administration requires a valid user CLI-token session; sandbox identity headers are not an authorization substitute.
• +npm-install is the exception that must also run inside a Linux te-agent sandbox because it packages the installed Linux files. It still requires the logged-in user to be root or agent_admin.
• Discover real IDs with a list command before any update or delete. Never guess a user, sandbox, model, quota rule, or channel ID.
• Before every write, run --dry-run, show the target and effect, and obtain explicit user confirmation. The CLI itself prompts only for high-risk-write; --yes can bypass that prompt and is not a security boundary.
• Use --dry-run to inspect method, path, query, and redacted body without executing.
• JSON inputs accept inline JSON, @file, or - for stdin. Prefer @file for channel credentials and other sensitive values.
• Successful output is JSON by default. Use --format table only when a human-readable table is more useful.
• After each command, check stderr and _notice.host_compat. If present, show the version warning and its update commands before the business result.
• Do not treat an absent ae-cli command as proof that an HTTP endpoint is unreachable. An Agent with Bash/network access can construct requests directly; server authentication, role checks, company isolation, and resource ownership are the actual controls.
• Do not call DELETE /api/admin/members?openId=... or /api/internal/sandboxes/** through ad-hoc HTTP. They are intentionally excluded from this Skill because other systems own those integration contracts.

Command Groups

Members

Examples:

+list-members filters:

• --q: login/display name search.
• --status: all | enabled | disabled.
• --page, --page-size: page size is 1-100.
• --all: return all matches.
• --sort-field periodUsedAmount, --sort-dir asc|desc: central usage sort.

+add-members --members schema:

Optional flags are --rule-id and --create-sandbox true|false.

Sandboxes

Examples:

Use Agent database user IDs from +list-members, not AE openIds, for sandbox commands.

Shared Sandbox Tools

Uploaded tools are registered for the current company with enabled=false. Upload does not activate the tool in any running sandbox. Review and enable/activate it through sandbox tool management after upload.

For activate/deactivate/status operations:

• --target-mode selected requires 1-50 --sandbox-ids; all-running forbids them.
• --tool-ids contains 1-20 real IDs from +list-sandbox-tools.
• --command-names-by-tool-id optionally limits an operation to named commands.
• --expected-tool-snapshots-by-id carries the version/package/command snapshot returned by the server for optimistic concurrency checks.
• JSON maps accept inline JSON, @file, or stdin. Use dry-run and user confirmation before distribution changes.

Preferred npm Flow

Run this inside the target Linux te-agent sandbox:

For a scoped package or a custom shared-tool identifier:

Requirements and behavior:

• --package must be an exact registry package version. Tags, ranges, URLs, Git sources, npm aliases, and local paths are rejected.
• The installed package must expose at least one package.json bin entry. Each bin becomes one tool command.
• The default tool name is the unscoped package name. Use --name only when a different valid lowercase tool identifier is required.
• npm lifecycle scripts are disabled with --ignore-scripts by default. Use --allow-scripts true only after reviewing and trusting the package and all transitive dependencies.
• The command calls the admin upload-policy endpoint before starting npm. A disabled feature, expired session, or non-admin role fails before installation.
• Installation uses a temporary prefix with development dependencies omitted. Temporary installation and ZIP files are removed whether upload succeeds or fails.
• npm-created node_modules/.bin symlinks are converted to regular executable wrappers in the ZIP. All other symlinks, special files, and links resolving outside the package root are rejected.
• Pure JavaScript Node.js CLIs are the supported baseline. Packages that require native addons, downloaded platform binaries, build tools, system libraries, or lifecycle setup may fail when scripts are disabled or when activated in a different runtime image.
• If lifecycle scripts are necessary, install and upload from the same Linux sandbox image family that will execute the tool. Upload never makes an incompatible native artifact portable.

Existing Directory Flow

Use the low-level command when the tool is already installed or assembled:

The directory root must contain exactly one tool.json. An external manifest is allowed only when the root has no tool.json:

Minimal manifest:

Upload contract:

• name and command names must start with a lowercase letter and contain only lowercase letters, numbers, ., _, or -, with a maximum length of 64.
• Command names must not replace reserved runtime commands such as node, npm, npx, bash, python, git, curl, or sudo.
• Every command entry must be a regular file under the upload root. Use runtime: "node" for JavaScript entry files and runtime: "native" only for an executable compatible with the sandbox Linux image.
• Paths must be relative and normalized. Absolute paths, .., backslashes, control characters, empty segments, and a current path segment are rejected.
• ZIP limits are 50 MB compressed, 500 MB unpacked, 50 MB per file, and 10,000 files. The server publishes only after independently validating the same boundaries.
• Do not pre-create or write /data/app/te_agent_ta/share/tools from a sandbox. Sandboxes are read-only for that directory; the authenticated te-agent upload endpoint owns the final write and registration.
• A tool name can be registered only once per company in this first static-version flow. Choose the final name and version before upload.

Models

Examples:

--biz-type is AE_AGENT | AI_QA and defaults to AE_AGENT. Use the database id returned by a model list, not the provider model name.

Usage

Examples:

Summary range:

• Use --days 1..365, or provide both --start-date and --end-date.
• Dates use YYYY-MM-DD.
• Do not combine --days with an absolute date pair.
• --refresh true is available on +get-usage-summary and +get-agent-tool-calls and bypasses the overview cache.

Details flags:

• --start-date and --end-date are required.
• --group-by: user | model | date | app_type.
• Optional filters: --search, --open-id, --model-id, --model-scope, --app-type.
• --model-scope requires --model-id.
• --sort-by: totalTokens | cost | share | requestCount.
• --sort-dir: asc | desc.

Combination drill-down requires exactly one parent selector:

• user → --open-id only.
• model → --model-id and --model-scope only.
• app_type → --app-type only.
• date → --date only, inside the selected range.

CSV exports require an explicit --output. The target is created exclusively: an existing file is never overwritten, and an HTTP or stream failure removes the incomplete file. The JSON result reports the absolute local path, bytes written, server filename, and content type.

Cost Control

Examples:

Quota rule JSON:

Rules:

• subjectType: USER | COMPANY.
• periodType: DAY | WEEK | MONTH.
• quotaType: COST | TOKEN.
• COST uses budgetAmount; TOKEN uses totalTokens. Token values are expressed in millions.
• allowedModels and modelLimits are optional according to the server rule type.
• Update accepts a partial rule object.

Channels

For channel setup, routing, WhatsApp Web linking, or Feishu user binding, read references/channel-management.md before taking action. It defines the two confirmation phases and the ae-cli plus Feishu OpenAPI MCP workflow.

Use canonical snake_case request fields. The four original commands also accept their legacy camelCase JSON fields for compatibility. Always use @file for credentials and batch rosters:

Permission Errors

A permission response looks like:

On this response:

1. Do not retry with another admin path.
2. Do not recommend re-login unless the server returned 401 instead.
3. Tell the user that root or agent_admin is required.

An authenticated root or agent_admin is still scoped to their own company. The current service checks the database role and company against the session, and audited member, channel, sandbox-tool, and sandbox-management routes apply company/resource ownership checks. Never use that statement as a claim that every unreviewed admin route is safe.

Transport Status

This is a Transitional domain backed by te-agent /api/admin/** and /api/cli/channel/v1/**.

• Maintainer: te-agent admin/channel-management routes and src/commands/te-system/**.
• Migration target: system and channel Capability Gateways.
• Review date: 2026-10-24.
• Exit condition: migrate after equivalent gateway schema, auth, risk, dry-run, and output contracts are stable.