dsh-context 0.35.0 → 0.37.0

package/README.md

@@ -119,6 +119,14 @@ Pick **Live (next request)** or any retained step from the picker, and browse wh
 
 Six collapsible category sections (system prompt, tool schemas, user messages, injected context, assistant replies, tool results) expand into one row per element — each with its token price — and every element expands again into its **actual content**: the full system prompt, each tool's description and JSON schema, message text, reasoning, tool-call arguments, and tool outputs.
 
+- **Who provides each tool** — tool-schema rows carry a best-effort source tag when attribution is knowable: the pinned first-party package map (`@deepseek-ai/dsh-tool-*`) for shipped tools, `mcp:<server>` for MCP-proxied tools (their public names encode the server), or a harness-logged `plugin` field. Beyond those, the host also watches live `tools.register()` calls and tags each tool to the plugin that registered it — third-party, agent-scoped and dynamically named tools included, even plugins installed from a local path or npm/pnpm link (their package name is resolved from the registering code's directory). The session log itself records only each tool's name/description/parameters, so a tool that was **already registered when the context plugin mounted** (e.g. local-link plugins that load earlier in the bundle) shows an **Unknown plugin** tag instead: the registering plugin is gone from every observable surface by the time attribution starts. Tools not in the current registry at all (older sessions) stay untagged.
+
+![Context browser tool schemas with per-plugin source chips, a text filter, and the size/name sort](https://raw.githubusercontent.com/bowenliang123/dsh-context/main/docs/context-browser-tools.png)
+
+- **Filter every category, sort the tools** — every category body carries a text filter that scans exactly what its rows show: tool schemas match name/description/parameters/plugin; message rows match the preview and the tag chip (call breadcrumbs, call-argument summaries) — full bodies stay out of scope, they load on demand. Each filter box names its own searchable fields, and the tools category adds a **size / name** sort with size (token price) as the default, mirroring the overview's Top chips. The filter is a lens on the open category: switching categories resets it, picking steps keeps it.
+
+![Row filtering in the Context browser: tool results narrowed by "fileCard"](https://raw.githubusercontent.com/bowenliang123/dsh-context/main/docs/context-browser-search.png)
+
 - **Linked with the trend chart** — hover any bar in the Context trend card and the browser previews that step instantly; leave the chart and it returns to your own pick. Keep a category open while scrubbing to compare one category across steps. Clicking a step-brief row (**User** / **In** / **Response**) opens that exact message here, expanded and scrolled into view.
 - **Honest about coverage** — steps before a compaction are reconstructed from the removed-message archive, and the card says so when a step's makeup is only approximate. Elements older than the loaded chat window page older history in automatically when you expand them, and live injections (AGENTS.md, session-start context, …) are always listed — never a token sum without its items.
 - **Diff against the previous turn** — switch the picker to **vs previous turn** and every category gets signed delta badges (`+N` items, `+Nk` tokens), so one glance tells you what the conversation added since the end of the last turn.