songjongdevsecops/dsh-olhos
Eyes for an agent: captures a URL or local HTML file, measures how far the screen drifted, and tests written expectations against the image, with a score attached to every verdict.
The agent gets an olhos_verificar tool and the user gets a /olhos slash command over the same machinery. A capture of a URL or a local HTML file is compared against written expectations, and each expectation comes back with its cosine score, a confirmed or not verdict and the capture path, optionally with drift measured against the last saved reference. The chat command can also keep a good capture as the new reference or find an older capture by meaning, which makes the plugin useful for catching visual regressions rather than only taking screenshots.
Source repository unavailable. The repository songjongdevsecops/dsh-olhos no longer exists on GitHub (HTTP 404 when checked on 2026-10-28), so the install command and any further commands on this page cannot currently be run. They are kept below as a record of what the plugin documented while it was available.
Install
npx @deepseek-ai/dsh plugin --profile web add github:songjongdevsecops/dsh-olhosCopied verbatim from the README's install block, which precedes it with npm install -g pnpm. There is no registry package for this name, so the documented GitHub spec is the install path. Restart the harness afterwards, because the bundle layer is read at boot, then confirm the loaded line with the README's --dump-config and grep command.
Compatibility
Requires Node.js ^22.19.0 or >=24.0.0, the range DeepSeek Harness supports. Captures attach to a Chrome instance exposing the DevTools protocol on port 9222, so the browser must be started with remote debugging before captures work. Verdicts are cosine scores in a 768-dimensional embedding space, so the comparison is numeric rather than a model opinion.
Details
- Repo:
songjongdevsecops/dsh-olhos(source unavailable) - Category: Vision & Multimodal
- Stars: 0
- Version: 0 stars, JavaScript, MIT, created 2026-10-18; install source is GitHub; English README only.
Alternatives
wads-dev/dsh-live-voice · TaurusWood/dsh-plugin-appshot · gloryxpnv/dsh-tool-vision