@mindstudio-ai/agent 0.1.100 → 0.1.101
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.
- package/README.md +54 -0
- package/dist/cli.js +90 -13
- package/dist/index.d.ts +351 -7
- package/dist/index.js +198 -6
- package/dist/index.js.map +1 -1
- package/dist/postinstall.js +90 -13
- package/llms.txt +12 -8
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -428,6 +428,56 @@ roles. Pass `onEvent` to stream progress (text chunks, tool calls, the final `do
|
|
|
428
428
|
> `outputSchema` typing uses a `const` type parameter — consuming projects need
|
|
429
429
|
> TypeScript 5.0+.
|
|
430
430
|
|
|
431
|
+
## Jewels
|
|
432
|
+
|
|
433
|
+
A jewel is an agentic shadow companion for an app method — `foo.jewel.ts` beside `foo.ts`,
|
|
434
|
+
built with `defineJewel(method, { subject, propose, grade? })`. It proposes the method
|
|
435
|
+
input a careful operator would submit (or abstains with `input: null`), gets graded
|
|
436
|
+
against what a human actually did, and climbs an autonomy ladder (`shadow` → `approve` →
|
|
437
|
+
`auto`) declared in the app manifest.
|
|
438
|
+
|
|
439
|
+
For decision moments the app detects itself (an ingest branch that lands a row in a
|
|
440
|
+
pending state), hand the moment to the jewels from backend code:
|
|
441
|
+
|
|
442
|
+
```ts
|
|
443
|
+
mindstudio.waitUntil((async () => {
|
|
444
|
+
const merge = await mindstudio.jewels.propose(
|
|
445
|
+
'merge-issues', { sourceId: issue.id }, { idempotencyKey: issue.id });
|
|
446
|
+
if (merge.outcome !== 'committed') {
|
|
447
|
+
await mindstudio.jewels.propose(
|
|
448
|
+
'triage-issue', { issueId: issue.id }, { idempotencyKey: issue.id });
|
|
449
|
+
}
|
|
450
|
+
})());
|
|
451
|
+
```
|
|
452
|
+
|
|
453
|
+
The platform routes each proposal by the method's autonomy — `recorded` (shadow),
|
|
454
|
+
`queued` (approve), `committed` (auto: the method ran) — and `idempotencyKey` has Stripe
|
|
455
|
+
semantics (a replayed key returns the original outcome, so webhook retries are safe).
|
|
456
|
+
Backend/managed contexts only.
|
|
457
|
+
|
|
458
|
+
For `approve`-mode methods, build the review inbox natively in the app with
|
|
459
|
+
`jewels.queue`:
|
|
460
|
+
|
|
461
|
+
```ts
|
|
462
|
+
// Backend method — gate with the app's own reviewer role.
|
|
463
|
+
export async function listDraftReplies() {
|
|
464
|
+
auth.requireRole('support_lead');
|
|
465
|
+
return mindstudio.jewels.queue.list({ methodId: 'send-reply' });
|
|
466
|
+
}
|
|
467
|
+
|
|
468
|
+
export async function reviewDraft({ itemId, action, edited }) {
|
|
469
|
+
auth.requireRole('support_lead');
|
|
470
|
+
return mindstudio.jewels.queue.resolve(itemId, {
|
|
471
|
+
action, // 'approve' | 'dismiss'
|
|
472
|
+
...(edited ? { input: edited } : {}), // an edit is captured with the pair
|
|
473
|
+
});
|
|
474
|
+
}
|
|
475
|
+
```
|
|
476
|
+
|
|
477
|
+
Approving **applies the target method as the signed-in reviewer** — the effect belongs
|
|
478
|
+
to the human who clicked, and the target method's own auth checks are the real gate on
|
|
479
|
+
who may approve. Unresolved items expire at the method's `attributionWindow`.
|
|
480
|
+
|
|
431
481
|
## Thread persistence
|
|
432
482
|
|
|
433
483
|
Steps execute within threads. Pass `$threadId` and `$appId` from a previous call to maintain state:
|
|
@@ -585,6 +635,10 @@ import type {
|
|
|
585
635
|
JsonSchema,
|
|
586
636
|
JsonObjectSchema,
|
|
587
637
|
FromSchema,
|
|
638
|
+
JewelConfig,
|
|
639
|
+
JewelPairRecord,
|
|
640
|
+
JewelProposeOutcome,
|
|
641
|
+
JewelProposeResult,
|
|
588
642
|
} from '@mindstudio-ai/agent';
|
|
589
643
|
```
|
|
590
644
|
|