Daily security scan
Date · 2026-08-22 (UTC — plugins first indexed that day)
Effective tier · new today
dynamic first · this report2 of 12 new entries take their tier from a dynamic audit conclusion; the rest use the static scan tier. Where a conclusion exists it is authoritative (it may raise or lower the tier).
Static scan · new today
rule heuristics · fallbackTier = scan verdict: critical hit → critical; warning / non-blocking critical → high / medium; none → low. Used only when no dynamic conclusion exists.
Dynamic audit · subset of new today
model reads source · no executionTier = highest finding. Of 12 new entries: 2 with a conclusion, 4 queued, 0 failed (no conclusion), 6 low-risk not queued.
Data file updated 2026-11-20 07:30 (UTC+8); latest conclusion audited 2026-11-11 07:33 (UTC+8)
| Plugin | Effective | Static | Dynamic | Findings C/W/I | Manifest | Indexed |
|---|---|---|---|---|---|---|
| agentscars A public commons of real AI-agent failure patterns ("scars") — searchable via API and MCP. Live at agentscars.com. Private for now, open-sourcing once ready. | low | low | not queued (low risk) | 0/0/4 | 0/0 | 2026-08-22 00:40 |
| dsh-plugin-session-manager-custom DSH Web plugin for local session data management | medium | medium | queued | 0/1/2 | 0/0 | 2026-08-22 00:40 |
| dsh-image-tiler DSH插件:将大图像分割成带标签的800像素图块,并保留概览图,同时保留视觉模型所需的细节。包含设置卡。DSH plugin: slice large images into labeled 800px tiles + overview, preserving detail for vision models. Settings card included. | low | low | not queued (low risk) | 0/0/0 | 0/0 | 2026-08-22 00:40 |
| dsh-hooks-git-ai A DSH plugin that records which files the agent edited, with which model, and in which session into git-ai | low | low | not queued (low risk) | 0/0/0 | 0/0 | 2026-08-22 00:40 |
| dsh-font-size DSH Web GUI plugin: 'Conversation font size' slider (12-22 px) in Settings → General. | low | low | not queued (low risk) | 0/0/4 | 0/0 | 2026-08-22 00:40 |
| dsh-jj-vcs Jujutsu version-control plugin and skill for DSH multi-agent teams | medium | medium | queued | 0/3/3 | 0/0 | 2026-08-22 00:40 |
| dsh-pm dsh-pm is the ChunSun × DSH reference plugin: an AI-native project-delivery loop driven by ChunSun. Requirements / Runs / Steps / acceptance scenarios & cases / work-memory, a session delivery panel, and 28 chunsun_* model tools — with the platform as the single source of truth. MIT. | low | low | not queued (low risk) | 0/0/0 | 0/0 | 2026-08-22 00:40 |
| dsh-novel-writing DSH 自动化小说写作发布流水线插件:claude-writing-workflow 迁移版 agent 预设 + 小说工作台(可视化/实时渲染/章节编辑)+ 多平台发布配置与数据驱动优化闭环 | lowD↓ | high | low | 1/0/4 | 0/0 | 2026-08-22 00:40 |
| dsh-one-gateway Private DSH One Gateway — loopback, identity-first ingress for DSH | low | low | not queued (low risk) | 0/0/2 | 0/0 | 2026-08-22 00:40 |
| dsh-lanchat | highD | high | high | 2/1/10 | 0/0 | 2026-08-22 00:40 |
| dsh-android DSH plugin for Android — build, run, and interact with a live emulator or USB device stream inside a conversation, driven entirely through adb. | medium | medium | queued | 0/2/7 | 0/0 | 2026-08-22 00:40 |
| Dsh-dev Version-aware workspace core for building DSH plugins | medium | medium | queued | 0/5/8 | 0/0 | 2026-08-22 00:40 |
Dynamic audit results · new today (2 with a conclusion)
The model reads redacted source context and never executes plugin code. When a dynamic conclusion exists it is authoritative (it may raise or lower the level); otherwise the static tier applies. The audit never writes the exemption list.
The runtime-loaded lib/index.js exposes an unauthenticated /lanchat hub that lets any LAN host or web page inject prompts into DSH sessions and exfiltrate chat via webhooks/peers.
- highUnauthenticated /lanchat/relay & /push inject arbitrary user messages into any DSH session and wake it · lib/index.js
- mediumUnauthenticated webhook registration turns every chat message into an outbound POST to an attacker-controlled URL · lib/index.js
- medium/fed/add allows unauthenticated SSRF and a registered peer receives room message text · lib/index.js
Novel-writing bundle: reviewed files are preset config, tool wrappers, installer and dev-only evidence scripts; no critical/high issue, only low supply-chain and configured-command exposure.
- lowAgent-facing tools spawn user-configured platform shell commands · lib/tools.js
- lowInstaller tracks mutable main branch and documents curl | sh execution · install.sh:35
Site-wide effective tiers (dynamic first)
Dynamic conclusions set the tier where they exist; the static scan applies only where there is none. 32,047 in-scope entries (31,684 scanned + 363 pending).
Static scan for reference: low 31,461 · medium 0 · high 223 · critical 0 (dynamic conclusions moved these into the effective tiers above).
Static heuristics + semantic manifest audit + a model-driven dynamic audit — not a full security audit; no code execution, no dependency resolution. Dynamic tiers come from the model reading source context (highest finding) and never exempt a static finding.
Daily report archive
95 reportsWhat does this daily report cover?
This report covers plugins first indexed by dsh.so on 2026-08-22. For each new plugin it records the name, risk tier, scan verdict, critical/warning/info findings, manifest audit signals, and index time. Plugins that were already in the registry before that date are not included.
How are the risk tiers assigned?
Risk tiers follow the same public model used across the site: four levels — critical / high / medium / low — read from finding counts, most severe first. Critical is reserved for secrets, exfiltration, destructive commands, or mining; a high-severity finding without a critical one reads high; medium (warning-level) findings read medium (needs review); no meaningful findings reads low. Where the model-driven dynamic security audit has a conclusion, that conclusion sets the tier (it can raise or lower it); the static scan tier (vet primary, dsh backstop) applies only when there is no conclusion.
Why is this not a full security audit?
The daily scan is a static heuristic scan plus manifest audit. It does not execute code or resolve dependencies, so it cannot prove that a plugin is safe. Use the report to prioritize inspection, not to replace it.
What does the "Manifest" column mean?
Manifest here is the vet engine's semantic analysis of the plugin's JSON manifest — not the dynamic security audit. It flags declaration-level signals such as install scripts, lifecycle hooks and broad file or network permissions.
Is the static scan the only layer?
No. A second, model-driven dynamic security audit runs on top of it for entries the static scan already flags as needing review: it reads source context without executing plugin code, covers that subset only, and its conclusion is authoritative when there is one — it can raise a verdict or lower it; the static scan tier (vet primary, dsh backstop) applies only when there is no dynamic conclusion. Exemptions stay manual and the script never writes the suppression list. This daily report shows both layers: every new entry is marked with its dynamic state (conclusion / queued / failed / out of the queue because it is low risk), and all conclusions of the day are listed below with their tier, model summary and findings. As of the latest run: 2941 dynamic audit records, 2341 with a model conclusion, 321 stricter than the static verdict, 794 softer.
Source: dsh.so security standards · Maintained by dsh.so · Published