bowenliang123/dsh-context
bowenliang123/dsh-contextDSH 上下文洞察面板:Context 仪表盘 + /context命令 + Context 浏览器,查看 Context的分类组成、内容详情、演进趋势、压缩/注入事件、统计等一站式 Context 全生命周期管理。
DSH 按模型的能力画像:把本地会话历史变成逐模型的会话数、工具使用、错误率与 Top 错误签名看板,支持时间范围过滤与多模型对比。
Per-model capability profiling — a plugin for DSH (DeepSeek Harness).
The more sessions you run, the harder it is to answer "which model is reliable at what, and where does it tend to fail". Reading raw session logs is slow, and manual tallying misses things.
This plugin turns the session history in ~/.dsh/sessions into per-model capability profiles, directly answering: per-model session count, tool-call volume, error count, error rate, most-used tools, and top error signatures.
Prerequisites: DSH (with web) + pnpm; Node ≥ 24 (zstd decoding uses the built-in zstdDecompressSync).
cd ~/.dsh/profiles/web # your DSH web profile directory
pnpm add github:Ansonfishing/dsh-cap-profile
Then add "dsh-cap-profile" to the dsh.profile.bundles array in package.json and restart dsh. A "Capability Profile" tab appears in the conversation view; the first open triggers a background initial scan (~20s for large histories), during which mock data is shown.
bash: [exit code: 137] OOM) reveal where a model tends to fail.Clone this repo and open test/harness/index.html in a browser — a zero-dependency render harness (mock data; ?scenario=mock|live|empty|error switches scenarios, ?chrome=0 hides the harness bar).
npm test # node --test test/*.test.mjs
Local development: after cloning, use pnpm add link:../path/to/dsh-cap-profile in your profile. Client-only changes need a browser refresh; Node-side changes (index.js / lib/*.js) need a dsh restart.
MIT © Ansonfishing