@vizuh/sabi 0.1.1 → 0.1.2

package/README.md

@@ -8,6 +8,19 @@ round 2 on, a read round goes cheap, edits and tests go mid, and a failing tool
 This package is the **mod only**. The local proxy (BYOK, for any harness that accepts a `baseURL`)
 lives in the [repository](https://github.com/vizuh/sabi).
 
+The two paths are independent:
+
+- **Command Code mod:** no Sabi provider key and no proxy; it routes the subscription already
+  available to Command Code.
+- **Local proxy:** works with OpenCode, Hermes, Kilo and other OpenAI-compatible clients, using
+  OpenRouter, Ollama or another configured upstream.
+
+For the proxy, Sabi loads only the credential names referenced by `sabi.config.json`. Existing
+environment variables win, followed by `SABI_SECRETS_FILE`, the nearest workspace `secrets/.env`,
+and `~/.config/sabi/secrets.env` or `~/.config/sabi/.env`. It never copies secret values into a
+harness config, terminal, worktree, log or Git. Users who do not use Jev can set `judge.enabled` to
+`false`; users without a central secrets file can keep exporting provider variables normally.
+
 ## Install
 
 ```bash

package/mod/sabi.mjs

@@ -1,4 +1,4 @@
-// @vizuh/sabi 0.1.1 — generated by pack.mjs from packages/adapters/command-code/mod/sabi.ts.
+// @vizuh/sabi 0.1.2 — generated by pack.mjs from packages/adapters/command-code/mod/sabi.ts.
 // Source and docs: https://github.com/vizuh/sabi
 
 // packages/core/src/state.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vizuh/sabi",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "Adaptive inference scheduling for Command Code: one bundled mod that routes each continuing round by model, effort and trajectory state.",
   "type": "module",
   "license": "MIT",