dsh-session-allow
Manifest valid★ 1DSH Web GUI plugin: session-scoped "Allow for this session" option in the approval dialog (per-mode standing grants, localStorage)
dsh-session-allow
简体中文: README.zh.md
A DSH (DeepSeek Harness) Web GUI plugin that adds an "Allow for this session" option to the approval dialog.
When you pick it, a session-scoped, per-mode standing grant is recorded (persisted in localStorage, isolated by session id). Later escalation requests of the same mode in this session are auto-approved without a dialog; other modes keep prompting as usual.
What it does
| Scenario | Behavior |
|---|---|
A tool asks to escalate to danger-full-access | Dialog shows: Reject / Allow once / Allow danger-full-access for this session |
| You click "Allow for this session" | The current request is allowed and the mode is granted for this session |
| Same mode is requested again in this session | Silently auto-approved, no dialog |
| A different mode is requested (not yet granted) | Still prompts normally |
| Bottom of the dialog | Shows this session's granted modes as revocable chips (click × to revoke) |
| Page refresh / new session | Grants persist per session id / stay isolated per session |
How it works (no core package changes)
The core approval panel is registered on the conversation.composer chain slot at priority 1. This plugin registers the same slot at priority 0 with an identical selector — the chain slot elects the lowest-priority entry whose select returns non-null, so this panel fully replaces the core one. Grants live in localStorage["dsh.sessionAllow.v1"]; a matching request is auto-answered allowed-once on the wire (the host audit still records approval/asked + approval/decided as usual).
Grant keys: mode:<mode> (parsed from the reason escalate sandbox to <mode>: ...); non-escalation reasons fall back to tool:<tool>:<reason>.
Install
Method 1 — recommended, one command (restart dsh web afterwards):
dsh plugin --profile web add https://github.com/AnakinCao/dsh-session-allow.git
Method 2 — manual: append to ~/.dsh/profiles/web/cordis.patch.yml:
- insert:
- id: dsh-session-allow
name: 'dsh-session-allow'
and place the repository content under ~/.dsh/profiles/web/node_modules/dsh-session-allow/ (or pnpm add), then restart dsh.
Pick ONE method — never mount twice.
Verify
- The dialog shows three buttons: Reject / Allow once / Allow … for this session.
- Click "Allow for this session" → the current request is allowed.
- Trigger the same mode again → no dialog, executes directly.
- The chips at the bottom show the granted mode; click × to revoke and the mode prompts again.
- Open another session → that mode still prompts (grants are per-session).
Uninstall
dsh plugin --profile web remove dsh-session-allow
or delete the - insert entry from the profile's cordis.patch.yml and remove node_modules/dsh-session-allow. Grant data lives in the browser's localStorage["dsh.sessionAllow.v1"] and can be cleared manually.
File structure
dsh-session-allow/
├── package.json # dsh.client / dsh.bundle declaration
├── cordis.patch.yml # loader entry registration
├── lib/
│ ├── index.js # host-side placeholder plugin
│ └── client.js # client bundle (panel takeover + session grants)
├── test/
│ └── behavior.test.mjs # grant / auto-allow / revoke regression tests
└── README.md
License
MIT
Comments
Loading…
From the same category
by toby-bridges
Local security audit for AI API relays and LLM proxies: detects prompt injection, model substitution, tool-call rewriting, SSE anomalies, error leakage, and Web3 wallet risks.
★ 857
AGPL-3.0
Python
Sep 16, 2026
dsh plugin --profile web add dsh-api-relay-auditby sandbaseai
Local-first, self-hosted AI agent runtime and MCP bridge with sandboxed sessions, memory, credentials, audit/replay, and a local Console.
★ 650
↓ 5/wk
Apache-2.0
TypeScript
Sep 22, 2026
dsh plugin --profile web add managed-agentsby hashgraph-online
Open-source antivirus for AI agents: block risky tools, secret access, prompt injection, malicious packages, MCP servers, plugins, and skills at runtime.
★ 650
Apache-2.0
Python
Sep 22, 2026
by SeaOf0
基于dsh web实现的多种模式,目的是服务于redteam进行授权的安全研究,覆盖渗透测试、红队评估、代码审计等范围领域,请勿用于非法行为。(允许二开,赋予模块各位自己的业务逻辑,方法论只有自己熟练的才好用,好的方法论=好的生态)
★ 576
MIT
Python
Sep 22, 2026
dsh plugin --profile web add @dsh-external/dsh-redteam-modelby howmp
面向 DeepSeek Harness(dsh)的渗透测试模式 @CloverSecLabs
★ 530
JavaScript
Sep 22, 2026
dsh plugin --profile web add @howmp/dsh-pentestby xiaods
k8e.sh - OpenSource Agentic AI Sandbox Matrix
★ 497
↓ 17/wk
Apache-2.0
Go
Sep 22, 2026
dsh plugin --profile web add @k8e-sandbox/dsh-k8e-sandbox-bundle