@mrciphersmith/keryx 0.2.113 → 0.2.114

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (4) hide show
  1. package/README.md +13 -0
  2. package/dist/cli.js +2700 -867
  3. package/dist/core.js +82 -2
  4. package/package.json +1 -1
package/README.md CHANGED
@@ -228,6 +228,19 @@ What is in it today:
228
228
  LAN hosts (RFC1918/CGNAT) as an explicit operator-trust boundary — built-in
229
229
  providers never get it, and loopback/link-local metadata addresses stay
230
230
  denied regardless.
231
+ - **Reasoning effort, provider-neutral.** `/reasoning
232
+ [off|minimal|low|medium|high|xhigh|max]` sets the session's thinking effort
233
+ for the main turn (no argument shows the resolved level and its source);
234
+ `KERYX_REASONING_EFFORT` and a persisted shell-config setting sit below it
235
+ in precedence, defaulting to `off`. Anthropic, OpenAI and Gemini each map
236
+ the level onto their own reasoning knob, clamping a level a model does not
237
+ support rather than sending it unrecognized. A custom OpenAI-compatible
238
+ provider ignores this control — its reasoning shape (inline `<think>` tags,
239
+ an out-of-band field, a gateway-specific request flag) is configured
240
+ per-provider in `llm-providers.json` instead. Reasoning streams live into
241
+ the shell and collapses to a `◆ thought for 12s · 1.8k tokens` block;
242
+ `/think auto|expand|hide` chooses how it is shown. See [the CLI
243
+ reference](docs/docs/cli-reference.md#reasoning-effort-and-output-budget).
231
244
  - **Durable sessions, per project.** JSONL transcripts on disk, resume across a
232
245
  process restart, and context compaction that keeps the full archive.
233
246
  `/resume`, `/sessions`, `/status`, `/flows`, `/compact`, `/new`, and