Install verification: runtime verified (L5) · tested on dsh 0.2.1-rc.2 · last checked Install verification details →
What it doesAI
Analyze images from local files or URLs with Doubao vision models via Volcano Ark, including model fallback.
- Analyze local images or remote URLs with Doubao vision models
- Designed for DSH agents and MCP clients
- Includes model fallback for more reliable image analysis
Capability tags are AI-inferred or rule-extracted from the README (see badge) — vocabulary-limited, not install-verified. AI = inferred; README = literal keyword from the readme.
Installation
dsh plugin --profile web add github:hunyu-feng/dsh-visionInstall method: GitHub · runtime verified (L5)
Verification & Compatibility
Grouped by plugin version: each card lists the dsh versions actually tested against that artifact version and their install verdicts (L5 wins). Untested combinations are omitted — absence means untested, never assumed compatible. L1–L3 are a plugin-level static archive (independent of the dsh version — the same verdict on every dsh row): they sit in each row's level grid next to L4/L5, marked "static", while their evidence is stored once per plugin inside that row's "checks & evidence" instead of being duplicated per version.
External credentials (plugin-level): none detected
package manifest not collected — the declared range is unknown.
| plugin \ dsh | dsh 0.2.1-rc.2 current |
|---|---|
| 6cf030 | ✓ |
6cf030github✓ 1 runtime-verified2026-10-22
https://github.com/hunyu-feng/dsh-vision
✓dsh 0.2.1-rc.2current2026-10-22L5 · runtime verified
- • Node.js: v24.14.0
- • DSH: 0.2.1-rc.2
- • Plugin artifact: github · https://github.com/hunyu-feng/dsh-vision (version unstated)
ID: dsh-vision-21@6cf030@dsh0.2.1-rc.2 · profile: l4-sandbox
Issued: 2026-10-22 · expires: 2026-10-29
View checks & evidence
active in loader tree (150 entries)
L1–L3 are plugin-level static conclusions (baseline, identical across dsh versions): evidence is stored once per plugin.
Author badges
Embed the official badges in your README to showcase security & install-test status:
`[](https://www.dsh.so/artifact/dsh-vision-21/)``[](https://www.dsh.so/artifact/dsh-vision-21/)`Security Report
Two automated layers, neither a manual review: a static scan (per-plugin rule checks) and a model-driven dynamic security audit that covers only the subset static-flagged at warn or above and never executes plugin code.
Static medium, dynamic medium — stricter wins → overall medium.
v1.0.0dsh manifest1.0.0git2026-10-221.0.0same versionsuspicious · gitVet is an AST static scan (npm artifact or git source).
show static findings (5)
- criticalpipeShell · shell
install.sh:27warn ' curl -fsSL https://ollama.com/install.sh | sh # or https://ollama.com/download' - criticalpipeShell · shell
install.sh:48curl -fsSL https://ollama.com/install.sh | sh - warningnetIp · network
bin/mcp-server.js:38ollamaBaseUrl: process.env.OLLAMA_BASE_URL ?? 'http://127.0.0.1:11434/v1', - warningnetIp · network
install.sh:57"baseUrls": { "ollama": "http://127.0.0.1:11434/v1" }, - warningnetIp · network
lib/vision.js:22export const DEFAULT_OLLAMA_ENDPOINT = 'http://127.0.0.1:11434/v1';
show 3 audit findings
- mediumanalyze_image reads any local image path supplied by the agent and uploads it to the configured endpoint
lib/vision.js:128The tool argument is not confined to the project directory: absolute paths and ../ are resolved and read, and detectMime also falls back to the file extension when magic bytes do not match, so any file named *.png is base64-inlined. With the default local Ollama provider the image stays on the machine, but once VOLCANO_API_KEY/ARK_BASE_URL/VISION_BASE_URL are configured the same call uploads the file to a remote service. An agent steered by injected instructions (from an image, a web page or another tool) can therefore cause local screenshots, ID scans or UI captures to be disclosed off-machin
- lowInstaller advertises curl|bash from a moving branch and runs an unpinned npx package
install.sh:40The script header recommends `curl -fsSL https://raw.githubusercontent.com/hunyu-feng/dsh-vision/main/install.sh | bash` and then executes whatever the latest `dsh-vision` version on npm does. Neither the script nor the package version is pinned, so installation depends on mutable upstream content; the script itself performs no destructive action, so impact is limited to the supply-chain surface.
- lowVision model output (derived from untrusted image content) is returned to the agent verbatim without untrusted-content framing
index.js:47The text returned by the tool comes from a model that just read an attacker-supplied image or remote endpoint, and it is fed back into the agent context as plain tool output. Text rendered inside a screenshot (e.g. 'ignore previous instructions and ...') can therefore act as an indirect prompt-injection carrier; the plugin adds no delimiters or warning label.
Focused vision/MCP plugin with no code execution or hidden exfiltration path; real risk is agent-driven reading of any local image file and uploading it to the configured (possibly remote) endpoint.
Rule: the dynamic audit can only raise risk, never lower it (raise-only, never an exemption); a wrong raise is withdrawn by human review and the tier falls back to the static verdict. Overall verdict = the stricter of the two layers.
Disclaimer: the static scan and the dynamic security audit are both automated analysis, not a security guarantee. Always review what you install.
Activity
Last commit 2026-10-22 · activity: Active
• Repo created: 2026-10-22
• Stars: ★ 0 · Forks: 0
• Health: Active — committed within last 30 days
Source
GitHub: github.com/hunyu-feng/dsh-vision
Author README
View on GitHub ↗The author's original README from the repository — fetched live, unedited (the English page loads the default README).
