reshuibuduo/tmcra-plugin-deepseek-harness
TMCRA × DeepSeek Harness: Recall and update project memory across software and sessions, switching tools without reintroducing the project
Moves a project's memory across apps and conversations. The plugin connects DeepSeek Harness lifecycle events to the TMCRA Memory API: a new human prompt recalls account-global and current-project evidence in parallel, injects the resulting evidence into the model-visible session log, and writes the completed user/assistant turn back as two role-separated records. Because TMCRA gives every account a global scope and every project an isolated project scope, work done in Harness can be recalled in a later Codex session and vice versa, as long as both tools use the same TMCRA account and resolve the same project identity.
Install
dsh plugin --profile web add https://github.com/reshuibuduo/tmcra-plugin-deepseek-harness/releases/download/v1.0.0-rc.1/dsh-tmcra-memory-1.0.0-rc.1.tgz
dsh plugin --profile web exec dsh-tmcra-memory loginREADME's 'Install the preview tarball' block quoted verbatim: the prebuilt release tarball is the preferred route and the second line authenticates the plugin against your TMCRA account. Release asset verified live 2026-09-17 (v1.0.0-rc.1 ships dsh-tmcra-memory-1.0.0-rc.1.tgz plus SHA256SUMS). The README also documents dsh --profile web --dump-config to confirm the package contributed its cordis.patch.yml, plus dsh plugin --profile web exec dsh-tmcra-memory status|logout and a setup command for the local writer/organizer providers. Registry check 2026-09-17: no dsh-tmcra-memory npm package resolved, consistent with the tarball route.
Compatibility
README: technical preview, verified with @deepseek-ai/dsh 0.1.1-rc.2 and TMCRA API 0.2.2, and it repeats that DeepSeek Harness itself is a developer preview that may introduce breaking changes. It needs a TMCRA account and, for the fully local path, the bundled E5/MiniLM, BGE-M3/BGE reranker and Qwen3 embedding/reranker profiles, installed with dsh-tmcra-memory local-install. The repository vendors the reviewable TypeScript client and lifecycle modules under src/sdk/; hosted API, billing, control-plane and production memory-engine code are not included.
Details
- Repo: reshuibuduo/tmcra-plugin-deepseek-harness
- Category: Web UI Enhancements
- Stars: 1
- Version: GitHub release tarball v1.0.0-rc.1 (asset verified live 2026-09-17; repo 1★, Apache-2.0, last push 2026-09-05)
- Last push: 2026-09-05
- First seen: 2026-08-14
Recent updates
The README documents the lifecycle, the verification steps, the local provider profiles and the current limits rather than a release-by-release table; the latest release when checked on 2026-09-17 was v1.0.0-rc.1.
FAQ
- How do I install it?
- The README's preview route is the prebuilt release tarball — dsh plugin --profile web add https://github.com/reshuibuduo/tmcra-plugin-deepseek-harness/releases/download/v1.0.0-rc.1/dsh-tmcra-memory-1.0.0-rc.1.tgz — followed by dsh plugin --profile web exec dsh-tmcra-memory login.
- Does my memory follow me between apps?
- Per the README yes, provided both tools use the same TMCRA account and resolve the same project identity: TMCRA gives each account a global scope and each project an isolated scope, so a Harness project can be recalled in Codex and vice versa.
- Can it run fully locally?
- The README documents a local path: the package includes E5/MiniLM, BGE-M3/BGE reranker and Qwen3 embedding/reranker profiles, and dsh-tmcra-memory local-install picks a profile. Separate provider APIs are also supported.