@xiaozhiclaw/qlogicagent-core 0.3.0 → 0.4.1

package/README.md

@@ -48,6 +48,10 @@ its own UI. Tool names and paths do not imply artifacts or business success.
 This library has no production dependencies, CLI, desktop host, local tool
 runtime, native session store or per-process credential/policy discovery.
 Unknown tool effects require product reconciliation before recovery.
+Optional transport `bindModelRequest` selects an opaque account/origin reference
+before dispatch; it is saved with the attempt and passed back during execution.
+The product transport must query that same origin during recovery. It must not
+switch accounts and treat a missing receipt as proof of rejection.
 Cloud tasks persist inference attempts before network I/O. Unknown model attempts
 require request-bound model-service evidence; acknowledged original responses can
 be replayed without another inference. Sensitive response bodies are excluded from
@@ -56,10 +60,32 @@ progress events. Provider usage is distinct from settlement evidence.
 Optional `assistantContext` binds module schema/prerequisites, enterprise configuration
 and source state, employee permissions, and versioned platform Skill/examples.
 Optional `goal` requires a product `verifyGoal` callback before semantic completion.
+Products may resolve a user-confirmed goal from receipts with `resolveGoal`, and
+return independent verification evidence to the same loop through `goalFeedback`.
+The model can correct its work or use existing input/confirmation wait tools;
+the product continues to own the acceptance criteria and verification rules.
+Criteria may carry opaque product `parameters`; verification can return `expected`,
+`observed`, `revision` and `detail` alongside source references. Core does not parse
+domain rules or declare unsupported criteria satisfied.
+Context anchors carry explicit revision/supersession/source references through
+compaction, keeping unresolved conflicts visible. Product tools retrieve material.
+Tool text is complete by default. The removed `maxToolResultChars` option is
+replaced by optional product-owned `runtime.projectToolResult`; persisted receipts
+retain the original result. No quality score controls execution.
 Task request/deadline budgets survive resume; products own atomic financial quota
 reservation through `reserveModelBudget`. Optional delta batching preserves control
 event order and cursor continuity. These contracts do not create new product stores,
 business rules, queues or accounting authorities.
+`appendInput(context, taskId, {id, messages, data?})` lets a product append an authenticated
+user amendment to a resolved completed/stopped task. The amendment ID is durable
+and idempotent; a conflicting repeat is rejected. It preserves the original
+request, deadline, cumulative model attempts, monetary ceiling and receipts.
+Unresolved effects, pending waits and cancellation must use their existing
+reconciliation/wait/cancellation paths first. The product's store must atomically
+requeue the task with the amendment receipt and progress event.
+Optional `data` stays in the durable product receipt and is not added to model
+messages. Products can use it to retain source references or tool inputs without
+repeating complete materials in every amendment prompt.
 Model retries require all three transport-error fields `retryable`, `safeToRetry`
 and `automaticRetryAuthorized` to be explicitly true (`ModelRetryAuthorization`).
 An unknown model outcome pauses a durable task for product reconciliation;