reCamera Intellisense

Requirements

• - python3 (no external packages)
• Reachable reCamera HTTP API (default port 80)
• Credentials stored in ~/.recamera/devices.json (created automatically; declared in skill metadata)

Security considerations

• - Credential storage: Device tokens are stored in ~/.recamera/devices.json. Protect this file with appropriate permissions (chmod 600) and do not place unrelated secrets there.
• Plain HTTP transport: Communication to devices uses HTTP (port 80) by default — data including images and tokens travels unencrypted. Configure HTTPS on your devices if operating on untrusted networks.
• Trusted networks only: The skill polls devices and downloads snapshot/image files. Only use it with cameras on networks you trust.
• Camera-specific tokens: Use dedicated per-camera tokens (sk_xxx). Do not reuse tokens shared with cloud services.
• Source review: The bundle includes full Python sources under scripts/. Review them to verify behavior matches your expectations before granting autonomous execution.

Scripts

All scripts live under {baseDir}/scripts and accept one JSON object as CLI argument (optional for detect_local_device and list_devices).

• - device_manager.py: add/update/remove/list/get device credentials, file download
• detection_manager.py: models, schedule, rules, events, event-image fetch
• capture_manager.py: capture status/start/stop, one-shot image capture

Full API signatures and CLI schemas: See REFERENCE.md

Agent rules

1. 1. Always pass complete JSON; never use interactive prompts.
2. Use exactly one of device_name (preferred) or inline device.
3. Auth token format: sk_xxx (from Web Console → Device Info → Connection Settings → HTTP/HTTPS Settings).
4. To detect by label name: call get_detection_models_info, map name → label index, use index in label_filter.
5. Poll get_detection_events every 1–10s; pass start_unix_ms for incremental reads.
6. Prefer event metadata first; fetch images only when needed.
7. CLI output: success = JSON on stdout (mutating commands may produce no stdout, check exit code 0); failure = actionable stderr. On error, surface stderr and provide one concrete fix.

Execution checklist

Copy and track for multi-step tasks:

CODEBLOCK0

CLI quickstart

Run from {baseDir}:

CODEBLOCK1

Python pattern (long-running automation)

CODEBLOCK2

Use a loop with checkpointed start_unix_ms for incremental polling.

Workflows

Onboard a device

1. 1. add_device with host + token.
2. INLINECODE24 to verify.

Configure object detection by name

1. 1. get_detection_models_info → map object name to label index.
2. INLINECODE26.
3. INLINECODE27 with label_filter containing the index.
4. INLINECODE29 to start fresh.

Monitor events

1. 1. Poll get_detection_events with start_unix_ms every 1–10s.
2. Track last timestamp for next poll.
3. Fetch images only when needed via fetch_detection_event_image.

On-demand snapshot

• - CLI: capture_image with local_save_path → returns {capture, saved_path, bytes}.
• Python: capture_image → persist returned content bytes.
• Alternative: fetch_detection_event_image with local_save_path.

Troubleshooting

Symptom	Fix
401/403 auth error	Re-copy token from Web Console
Timeout / connection refused

Verify host, network path, device power | | Schedule rejected | Use Day HH:MM:SS format | | Empty rules or events | Enable rule/storage prerequisites; check region filter; poll more frequently | | Image fetch failed | Use fresh snapshot_path; data may rotate out | | Import errors in Python mode | Run from {baseDir}; append ./scripts to sys.path |