knqiufan/powercontext-dsh
101Last commit Aug 16, 2026
powercontext-dsh DSH plugin
PowerContext-DSH is a DeepSeek Harness plugin that integrates with a running PowerContext Server over HTTP. It automatically recalls context and captures user input before each model step, and exposes named tools for memory, handoff, experience, skill, and review operations. The plugin does not embed storage, start the server, or import the Python package.
How to install the powercontext-dsh DSH plugin
dsh plugin --profile web add /path/to/powercontext/integrations/dsh/plugins/powercontextCopying does not run this command. Review the repository and version before installing the powercontext-dsh DSH plugin.
powercontext-dsh DSH plugin data source
powercontext-dsh DSH plugin snapshot date: Aug 16, 2026
discovered
What the powercontext-dsh DSH plugin can do
- Recall bounded context via POST /v1/context/prepare and inject as historical evidence
- Capture user input as a Content Source via POST /v1/sources/content
- Provide pc_* tools for memory (search, remember, list, get, revise, retire)
- Provide pc_* tools for handoff (activate, prepare, finalize, commit, continue)
- Provide pc_* tools for experience and skill generation and retrieval
- Provide pc_* tools for review listing and retrieval
- Support remote PowerContext Server with authentication via environment variables
- Check server reachability with /pc doctor chat command
Where the powercontext-dsh DSH plugin fits
- Persistent memory across DSH conversations using PowerContext Server
- Handoff complex tasks to other agents or workflows
- Generate and retrieve experience and skill entries to improve model performance
- Review and manage artifact candidates before committing
- Maintain context consistency across turns by automatically recalling relevant information
Who the powercontext-dsh DSH plugin is for
- Developers using DeepSeek Harness who need a persistent memory and context management backend
- Teams deploying PowerContext Server and wanting to integrate it with DSH
powercontext-dsh DSH plugin limitations
- Requires a running PowerContext Server (not included); plugin does not start or manage the server
- Server must be reachable over HTTP; if unreachable, recall is skipped and turn continues without context
- Default server binds to 127.0.0.1:8000 with no authentication; remote deployment requires TLS and auth setup
- Plugin does not support MCP; it uses OpenAPI /v1/... endpoints only
- On Windows, replacing a link-type install with a tarball fails due to pnpm symlink issues
powercontext-dsh DSH plugin: from the repository README
Quoted from the knqiufan/powercontext-dsh README, the upstream source of the powercontext-dsh DSH plugin. Copyright remains with the original authors.
**English** | [中文](README.zh.md) DeepSeek Harness plugin that connects to a running [PowerContext](https://github.com/oceanbase/powercontext) Server over HTTP for recall, memory, handoff, experience, and skills. It does not embed storage, start the Server, or import the Python package. The same plugin is also integrated into official PowerContext at [`integrations/dsh/plugins/powercontext`](https://github.com/oceanbase/powercontext/tree/master/integrations/dsh/plugins/powercontext). This standalone repository and the in-tree plugin stay in sync: fixes and improvements land in both places. ```bash dsh plugin --profile web add <path-or-tarball> ``` ## Features The plugin calls the Server’s `/v1/...` OpenAPI surface over HTTP. It does not use MCP. Before each model step it automatically: 1. **Recalls** bounded context with `POST /v1/context/prepare` and injects it as untrusted historical evidence. 2. **Captures** the current user input as a Content Source with `POST /v1/sources/content`. Named `pc_*` tools expose the agent-safe Memory, handoff, experience, skill, and read-only review operations. DSH requests one-time user approval before named mutations run. Review mutations r
Read the full READMERepository license: Apache-2.0
powercontext-dsh DSH plugin questions
How do I install the PowerContext-DSH plugin?
Make sure you have DeepSeek Harness installed and the web profile exists (run `dsh web` once). Then use `dsh plugin --profile web add <path-or-tarball>`. You can also use the PowerContext CLI: `powercontext setup dsh --source oceanbase/powercontext --ref master`. If you have a local checkout of PowerContext, point the source to the path. For a released tarball, use `dsh plugin --profile web add ./powercontext-dsh-X.Y.Z.tgz`.
Do I need to run a PowerContext server separately?
Yes, the plugin connects to a running PowerContext Server over HTTP. It does not embed storage or start the server. You must start the server with `powercontext server run` (defaults to http://127.0.0.1:8000). Both the server and the plugin must be using the same Git ref. The server is not included in the plugin package — install it separately with `uv tool install "powercontext[cli,server] @ git+https://github.com/oceanbase/powercontext.git@master"`.
What tools does the plugin provide?
The plugin exposes named tools prefixed with `pc_`: memory tools (pc_search, pc_remember, pc_memory_list, pc_memory_get, pc_memory_revise, pc_memory_retire), handoff tools (pc_handoff_activate, pc_handoff_prepare, pc_handoff_finalize, pc_handoff_commit, pc_handoff_continue), experience/skill tools (pc_experience_generate, pc_experience_get, pc_skill_generate, pc_skill_get), and review tools (pc_review_list, pc_review_get). It also automatically recalls context and captures user input before each model step.
Can I use the plugin with a remote PowerContext server?
Yes, you can configure the server URL via the `POWERCONTEXT_DSH_BASE_URL` environment variable (default http://127.0.0.1:8000). For remote servers, enable authentication by setting `POWERCONTEXT_SERVER_AUTH_ENABLED=true` and `POWERCONTEXT_SERVER_AUTH_TOKEN=<long-random-secret>` on the server side, and pass `POWERCONTEXT_DSH_AUTHORIZATION="Bearer <token>"` to the plugin. Ensure TLS is in front of the server before exposing it on a network.
What happens if the PowerContext server is unreachable?
If the server is unreachable during a recall step, the plugin skips the recall and the turn continues without injecting context. The user input capture may also be skipped depending on the configuration. The model will not have access to memory or other tools until the server is reachable again. You can check connectivity with the `/pc doctor` chat command.