| name | argus |
| description | Use this skill to process one to 99 local exact 2:1 equirectangular panorama images with Realsee Argus, producing depth maps, a merged GLB point cloud, camera poses, optional intrinsics, and a validated local output index. Trigger for Argus panorama reconstruction, Argus ZIP input, or explicit Argus start/status/collect lifecycle requests. Do not trigger for panorama editing or stitching, arbitrary-photo 3D generation, existing GLB inspection, or research-only questions. |
| compatibility | Requires a POSIX shell, Node.js 22+, npm 10+, npm registry access, and network access to app-gateway.realsee.ai or app-gateway.realsee.cn |
| metadata | {"version":"2.0.0","documentation":"README.md"} |
argus
Use this Skill to submit 1–99 exact 2:1 equirectangular panoramas to Realsee Argus and collect a validated output.zip. The public entrypoint is scripts/run-argus.mjs; <skillDir> below means the directory containing this file.
Treat the official product, demo, research, and developer sites as background only. Do not infer arbitrary-photo input or other product workflows from them; follow the narrower Skill 2.0 contract in this file.
Argus is a remote upload. Do not upload until the user has selected the input and consented. Never print, log, or persist credentials, upload tokens, presigned URLs, or raw provider errors. Do not open output files unless the user asks.
1. Ensure runtime dependencies
Skill installers copy the canonical package but may not install its Node.js dependencies. Before the first Argus command in an installed Skill directory, check all four runtime packages without printing any sensitive data. If any are absent, install the exact lockfile once:
test -f "<skillDir>/node_modules/@realsee/universal-uploader/package.json" \
&& test -f "<skillDir>/node_modules/@aws-sdk/client-s3/package.json" \
&& test -f "<skillDir>/node_modules/ajv/package.json" \
&& test -f "<skillDir>/node_modules/yauzl/package.json" \
|| (cd "<skillDir>" && npm ci --omit=dev --ignore-scripts --no-audit --no-fund)
Do not replace npm ci with an unlocked install. If the package cannot be resolved, stop and report the install error; do not begin an upload.
2. Resolve credentials
The runtime requires REALSEE_APP_KEY, REALSEE_APP_SECRET, and REALSEE_REGION (global or cn). The Gateway base is unchanged: global uses app-gateway.realsee.ai; CN uses app-gateway.realsee.cn. In the CN-only Arkclaw distribution, use cn and do not offer the global region.
Resolve values in the existing order:
-
Probe the current shell environment without printing values:
printenv REALSEE_APP_KEY REALSEE_APP_SECRET REALSEE_REGION >/dev/null \
&& echo present || echo missing
-
If ~/.realsee/credentials already exists, load it into the shell and probe presence. Never display the file:
[ -f ~/.realsee/credentials ] && set -a && . ~/.realsee/credentials && set +a; \
[ -n "$REALSEE_APP_KEY" ] && [ -n "$REALSEE_APP_SECRET" ] && [ -n "$REALSEE_REGION" ] \
&& echo present || echo missing
-
Otherwise ask for region, APP_KEY, and APP_SECRET one field per turn. Never repeat a supplied value. If the user explicitly chooses to persist them, retain the existing mode-0600 ~/.realsee/credentials flow. Never place credential values in a CLI argument or environment prefix recorded by the host.
3. Select the input
Two mutually exclusive modes are supported:
- repeat
--image <absolute-path> for 1–99 local images; or
- pass one
--zip <absolute-path> containing root-level images.
The CLI performs authoritative validation and deterministic packaging. Inputs must be JPEG, PNG, or WebP, RGB, 8-bit, and exactly width == 2 * height. A resolution below 2048×1024 emits a warning. Square 1:1 images are rejected; users who require the old square/single-GLB workflow must pin v1.0.2.
ZIP mode is not a validation bypass. The CLI safely extracts, validates, Unicode-normalizes, sorts, and repacks it before upload. Do not manually rename output IDs: consumers trust the algorithm's name_mapping.
The Skill ships only examples/manifest.json, which lists two first-party example sets and their CDN URLs, byte lengths, and SHA-256 digests. Panorama JPEGs are absent from the current release tree and every generated distribution. If the user wants official examples, ask them to choose a region and an absolute output directory outside <skillDir>, then run:
node <skillDir>/scripts/download-examples.mjs \
--region cn \
--output "/absolute/example-output"
The downloader publishes the directory only after every file passes its manifest byte-length and SHA-256 checks. Do not create, rename, or replace the requested output path or its parent while the command is running. Use global instead of cn when appropriate; the CN-only Arkclaw distribution only allows cn. Downloading does not consent to a later Argus upload. Before start, the user must still select the downloaded files and consent to sending them to Realsee. See the example panorama guide and its Chinese version.
4. Start once
The user selecting files for an Argus request is upload consent. If the user has not selected files, ask one question that also states the files will leave the machine for remote processing. Do not ask a redundant second confirmation.
For repeated images:
set -a; . ~/.realsee/credentials; set +a; \
node <skillDir>/scripts/run-argus.mjs start \
--image "/absolute/path/a.jpg" \
--image "/absolute/path/b.webp" \
--workspace "/absolute/workspace-root" \
--yes --json
For an existing ZIP:
set -a; . ~/.realsee/credentials; set +a; \
node <skillDir>/scripts/run-argus.mjs start \
--zip "/absolute/path/input.zip" \
--workspace "/absolute/workspace-root" \
--yes --json
If credentials already exist in the inherited shell, omit the source prefix. Capture workspace_dir from the JSON response; it is the durable run handle for later commands.
start validates and packages locally, uploads one ZIP, submits once, persists task_code, and returns. It does not poll in the background. Never automatically rerun start after submission_unknown: the submit operation is not idempotent and a blind retry may create a duplicate task.
5. Query status explicitly
Run one status query:
set -a; . ~/.realsee/credentials; set +a; \
node <skillDir>/scripts/run-argus.mjs status \
--workspace "<workspace_dir>" --json
Interpret task_status as queued, processing, succeeded, or failed. When queued or processing, report the current state and query again later only when appropriate. There is no detached poller and no --resume mode.
6. Collect a terminal result
When the task succeeds, run:
set -a; . ~/.realsee/credentials; set +a; \
node <skillDir>/scripts/run-argus.mjs collect \
--workspace "<workspace_dir>" --json
collect retains the original output.zip, safely extracts it, validates output.json and all referenced artifacts, and writes a local result.json index. Repeating collect is safe: a completed run is not resubmitted or downloaded twice.
Report these fields separately:
task_status: remote lifecycle state;result_status: algorithm result (success, partial, or error);- local
output_zip_path, output directory, manifest, merged GLB, depth-map, pose, and optional intrinsics paths;
missing_ids and warnings.
For partial, the CLI exits 0. Still show a prominent warning and the complete non-empty missing_ids list. For error, surface the sanitized error and treat the command as failed. Do not present temporary result URLs as durable output.
References
