Realsee Skills

English | 简体中文

Realsee Skills provides installable agent skills for Realsee workflows. Users and agent runtimes can install these skills to generate Realsee outputs from local inputs.

The current skill is argus. It generates Realsee Argus GLB output from a local JPEG image or panorama.

Credentials

Every install path needs three values:

Key	Purpose	Sensitive
REALSEE_APP_KEY	Realsee Open Platform APP_KEY	✅
REALSEE_APP_SECRET	Realsee Open Platform APP_SECRET	✅
REALSEE_REGION	global (app-gateway.realsee.ai) or cn (app-gateway.realsee.cn)	—

Register at my.realsee.ai (global) or my.realsee.cn (cn), then email developer@realsee.com with your account region, UserID, and IdentityID to request Argus VGGT API capability.

Install & Use — Claude Code

One-line install (inside a Claude Code session):

/plugin marketplace add realsee-developer/skills
/plugin install realsee-skills@realsee-developer-skills

Claude Code will prompt you for REALSEE_APP_KEY / REALSEE_APP_SECRET / REALSEE_REGION via the plugin config dialog. The two secrets are stored in the system keychain (not in settings.json).

Use — just describe the task in chat; Claude picks the skill based on its SKILL.md description. Examples:

Turn /path/to/photo.jpg into a Realsee Argus GLB (image mode).
Generate an Argus GLB from /path/to/pano.jpg (panorama, 2:1 aspect).

Need an explicit handle? The plugin-namespaced id is realsee-skills:argus.

Development install from a clone:

git clone https://github.com/realsee-developer/skills.git
cd skills
npm install && npm run rebuild
claude --plugin-dir ./plugins/realsee-skills

Install & Use — Codex

One-line install (host machine):

npx skills add realsee-developer/skills --skill argus --agent codex

This copies .agents/skills/argus/ into $CODEX_HOME/skills/argus (default $HOME/.codex/skills/argus).

Use — export credentials, then reference the skill in your Codex prompt:

export REALSEE_APP_KEY=...
export REALSEE_APP_SECRET=...
export REALSEE_REGION=global   # or cn

Use $argus on /path/to/photo.jpg (image mode) and report the GLB path.

Install from a clone instead (sets CODEX_HOME if you want a custom location):

CODEX_HOME=$HOME/.codex npm run install:codex-skills

Install & Use — npx skills (any detected host)

One-line install for the currently-active agent:

npx skills add realsee-developer/skills --skill argus

Install for all detected hosts in one call:

npx skills add realsee-developer/skills --skill argus --agent '*'

Or target a specific host:

npx skills add realsee-developer/skills --skill argus --agent claude-code
npx skills add realsee-developer/skills --skill argus --agent codex

List skills without installing:

npx skills add realsee-developer/skills --list

Install from a local checkout:

npx skills add . --skill argus

Use — once installed, invoke the skill exactly as you would on each host (see the Claude Code / Codex sections above).

Direct CLI Use (no host required)

Synchronous run (blocks until GLB download; Argus inference can take several minutes):

node .agents/skills/argus/scripts/run-argus.mjs --image /absolute/path/input.jpg --workspace ./workspace --yes --json

Asynchronous (returns immediately with status: in_progress; a detached process polls + downloads):

node .agents/skills/argus/scripts/run-argus.mjs --image /absolute/path/input.jpg --workspace ./workspace --yes --json --async

Resume or recover an async run from its workspace directory:

node .agents/skills/argus/scripts/run-argus.mjs --resume --workspace ./workspace/<run-dir> --json

Input type is auto-detected from the JPEG dimensions and strictly enforced: 2:1 (±0.05) → panorama, 1:1 (±0.05) → pinhole image, anything else is rejected before upload. --type panorama / --type image may be passed to override auto-detection, but the override is still validated against the file's dimensions.

Argus generation uploads the selected local image to Realsee remote services. Confirm user consent before any upload.

Open The Result

The skill does not ship an opener script. Read the workspace's result.json directly:

cat ./workspace/<run-dir>/result.json

When result.json#status === "success", ask the user whether to open the local GLB, the H5 preview, both, or neither. Then invoke the OS-native opener (see SKILL.md "Step 5"):

case "$(uname -s)" in
  Darwin)               open "<path-or-url>" ;;
  Linux)                xdg-open "<path-or-url>" ;;
  CYGWIN*|MINGW*|MSYS*) start "" "<path-or-url>" ;;
esac

Do not open anything until result.json#status is success.

Agent Runtime Files

File	Purpose
llms.txt	Machine-readable repository map.
AGENTS.md	Safe operating rules for automation.
SKILL.md	Runtime-facing skill definition.
README.md	Skill-specific user documentation.
argus-gateway-openapi.json	Public Gateway contract used by the skill.

Skill

Skill	State	Description
argus	Stable	Generate Realsee Argus GLB output from a local JPEG image or panorama.

Release metadata lives in release-channel.json.

Local Checks

Check local prerequisites:

npm run doctor

Run skill tests:

npm run test:skill

Contribute

Source skill files live under .agents/skills/. The Claude plugin package under plugins/realsee-skills/ is generated from the source skill files.

After changing source skill files, run:

npm run rebuild
npm run ci

Documentation:

• Architecture / 架构
• Install guide overview / 安装指南总览
• Claude Code install / Claude Code 安装
• Codex install / Codex 安装
• Usage guide / 使用指南
• Maintainer guide / 维护者指南
• Release guide / 发布指南
• Public distribution checklist / 公开分发检查清单
• Support / 支持
• Community guide / 社区指南
• Contribution guide / 贡献指南
• Security policy / 安全政策
• License

License

This repository is source-available under the Realsee SDK License Agreement. It is not published under an OSI open source license.