Semantic search over video files using Gemini embeddings. Index dashcam, security camera, or any mp4 footage, then search with natural language queries to fi...
Security Analysis
high confidenceThe skill's requirements and instructions line up with its stated purpose (video semantic search using Gemini embeddings); the only noteworthy risks are that it clones/installs code from the linked GitHub repo and uploads video chunks to the Gemini API (so sensitive footage and billing exposure are the main concerns).
Name/description (semantic search over video using Gemini embeddings) match what the skill asks for: GEMINI_API_KEY is the primary credential, ffmpeg is needed for trimming/preprocessing, python3 and the 'uv' tool are required to install/run the repo. Required binaries and env var are expected for the stated functionality.
SKILL.md instructs the agent/user to git clone the GitHub repo and run 'uv sync', then run the sentrysearch CLI which splits video into chunks, pre-processes them, and sends raw video chunks to Gemini Embedding 2. The instructions are scoped to the job (index/search/trim/overlay). Important behavioral detail: chunks of video (potentially sensitive footage) are sent to an external API (Gemini), which is necessary for embeddings but is a privacy/billing consideration.
There is no registry install spec; the SKILL.md instructs cloning https://github.com/ssrajadh/sentrysearch.git and running 'uv sync' which will fetch and install code/dependencies from that repository. This is a standard but higher-risk install pattern than instruction-only: the user/agent will pull and execute code from the remote repository, so the repo contents should be reviewed before running.
Only GEMINI_API_KEY is required and is appropriate for a skill that calls Gemini embeddings. Note: the API key enables outbound uploads of video to an external service and could incur cost and privacy exposure; the SKILL.md warns about cost and validates the key via a test embedding.
Skill is not always-on and does not request system-wide config paths. The CLI writes a local .env when run (expected behavior). Autonomous invocation is allowed by default (platform normal) but not combined with other high-risk flags here.
Guidance
This skill appears coherent for its stated purpose, but take these precautions before installing: 1) Review the GitHub repository (https://github.com/ssrajadh/sentrysearch) before running 'git clone'/'uv sync' so you know what code and dependencies will be installed. 2) Be aware that indexing sends video chunks to Google Gemini — do not use with sensitive private footage unless you accept that it will be transmitted to an external service and may be stored/processed by that provider. 3) The GEMINI_API_KEY grants the skill the ability to call the external API and will incur costs; consider using a scoped or billing-limited key and monitor usage. 4) Run installation and indexing in a controlled environment (isolated VM or container) if you have concerns about executing unreviewed code. 5) If you only want local-only processing, confirm the repo does not send data to any endpoints other than Gemini and that preprocessing is acceptable for your privacy needs.
Latest Release
v0.2.3
Added local model support, Tesla metadata overlay, similarity threshold, switched to uv
Popular Skills
Published by @ssrajadh on ClawHub
