feat: sync upstream SDK to runtime 1.0.84-8

.copilot-schema-version

@@ -1 +1 @@
-1.0.84-5
+1.0.84-8

CHANGELOG.md

@@ -3,6 +3,29 @@ All notable changes to this project will be documented in this file. This change
 
 ## [Unreleased]
 
+### Added (v1.0.84-8 sync)
+- Added stable event provenance for assistant originating messages, configured
+  MCP server identity/source, subagent model selection, and managed MCP server
+  display names, and added the `"error"` outcome to the public MCP
+  headers-refresh completion contract.
+  ([upstream PR #2658](https://github.com/github/copilot-sdk/pull/2658))
+
+### Changed (v1.0.84-8 sync)
+- Updated the runtime and schema pin to `1.0.84-8` and recertified the complete
+  stable Node SDK public surface through upstream commit
+  [`e9df3938b0f2bb028b203f4095d48b75f155c008`](https://github.com/github/copilot-sdk/commit/e9df3938b0f2bb028b203f4095d48b75f155c008).
+  Factory pause/checkpoint APIs, shell completion facts, managed catalog and
+  workspace RPC additions, and internal skill-reference events remain outside
+  the stable Clojure API.
+  ([upstream PR #2658](https://github.com/github/copilot-sdk/pull/2658),
+  [upstream PR #2662](https://github.com/github/copilot-sdk/pull/2662))
+
+### Fixed (v1.0.84-8 sync)
+- Factory abort requests now cancel only the matching execution token, so a
+  stale attempt cannot cancel another attempt sharing the durable run ID.
+  The new factory pause/checkpoint API remains intentionally excluded.
+  ([upstream PR #2537](https://github.com/github/copilot-sdk/pull/2537))
+
 ### Added (v1.0.84-5 sync)
 - Added stable `:mcp-transport` metadata to
   `:copilot/tool.execution_start` events. The optional field accepts `"stdio"`,

doc/api/agent-factories.html

@@ -154,7 +154,7 @@ <h2><a href="#cancellation" id="cancellation"></a>Cancellation</h2>
 <p>There are three distinct cancellation mechanisms, and they don’t overlap:</p>
 <ol>
 <li><strong>Cooperative, inside <code>:run</code></strong> — react to <code>:cancel-chan</code> closing or poll <code>:cancelled?</code> to stop early. Cancellation does not interrupt <code>:run</code> automatically; the factory must check for it.</li>
-<li><strong>From outside, server-side</strong> - <code>cancel-factory-run!</code> asks the CLI to cancel a run. The runtime then sends a reverse <code>factory.abort</code> request, which closes the run’s <code>:cancel-chan</code> and eventually delivers a <code>:cancelled</code> terminal status.</li>
+<li><strong>From outside, server-side</strong> - <code>cancel-factory-run!</code> asks the CLI to cancel a run. The runtime then sends a reverse <code>factory.abort</code> request for the active execution token, which closes only that attempt’s <code>:cancel-chan</code> and eventually delivers a <code>:cancelled</code> terminal status. An older attempt cannot cancel a replacement that shares the durable run ID.</li>
 <li><strong>Aborting a local wait, not the run</strong> — <code>wait-for-factory-run!</code> accepts an optional <code>:cancel-chan</code>; closing it makes the <em>wait</em> throw <code>ex-info</code> with <code>{:type :factory-wait-cancelled}</code> without cancelling the run itself. Use this to stop blocking on a run you still want to keep executing.</li>
 </ol>
 <h2><a href="#limits" id="limits"></a>Limits</h2>

doc/api/github.copilot-sdk.session.html

@@ -58,7 +58,8 @@
 </div></div></div><div class="public anchor" id="var-handle-elicitation-request.21"><h3>handle-elicitation-request!</h3><div class="usage"><code>(handle-elicitation-request! client session-id context)</code></div><div class="doc"><div class="markdown"><p>Handle an incoming elicitation.requested broadcast event. Calls the session’s elicitation handler with a single ElicitationContext arg (includes :session-id alongside request fields). Returns a channel with the result. If the handler fails, returns {:action “cancel”} to avoid hanging requests.</p>
 </div></div></div><div class="public anchor" id="var-handle-exit-plan-mode-request.21"><h3>handle-exit-plan-mode-request!</h3><div class="usage"><code>(handle-exit-plan-mode-request! client session-id request)</code></div><div class="doc"><div class="markdown"><p>Handle an incoming exitPlanMode.request RPC (upstream PR #1228). The server asks the client whether to exit plan mode. Calls the session’s :exit-plan-mode-handler with <code>(request, {:session-id ...})</code>. If the handler returns a channel, awaits its value. Returns a channel with a wire-shaped response wrapped in {:result …} or {:error …}.</p>
 <p>Default behavior (no handler): {:result {:approved true}}. Handler must return an idiomatic map containing :approved? (required boolean), optional :selected-action (string), :feedback (string). If the handler result is malformed (non-map, missing :approved?, or :approved? is not a boolean), a warning is logged and the response falls back to {:result {:approved true}} (matching the no-handler default). Exceptions thrown by the handler are converted to {:error {:code -32603 :message …}}.</p>
-</div></div></div><div class="public anchor" id="var-handle-factory-abort.21"><h3>handle-factory-abort!</h3><div class="usage"><code>(handle-factory-abort! client session-id run-id)</code></div><div class="doc"><div class="markdown"><p>Cooperatively cancel active executions for a durable factory run.</p>
+</div></div></div><div class="public anchor" id="var-handle-factory-abort.21"><h3>handle-factory-abort!</h3><div class="usage"><code>(handle-factory-abort! client session-id run-id)</code><code>(handle-factory-abort! client session-id run-id execution-token)</code></div><div class="doc"><div class="markdown"><p>Cooperatively cancel factory executions.</p>
+<p>The four-arity form targets one execution attempt. The three-arity form retains run-wide cancellation for direct callers of this low-level handler.</p>
 </div></div></div><div class="public anchor" id="var-handle-factory-execute.21"><h3>handle-factory-execute!</h3><div class="usage"><code>(handle-factory-execute! client session-id {:keys [name run-id execution-token], :as params})</code></div><div class="doc"><div class="markdown"><p>Execute an extension-authored factory for a runtime reverse RPC.</p>
 </div></div></div><div class="public anchor" id="var-handle-hooks-invoke.21"><h3>handle-hooks-invoke!</h3><div class="usage"><code>(handle-hooks-invoke! client session-id hook-type input)</code></div><div class="doc"><div class="markdown"><p>Handle an incoming hooks invocation. Returns a channel with the result. PR #269 feature.</p>
 </div></div></div><div class="public anchor" id="var-handle-mcp-auth-request.21"><h3>handle-mcp-auth-request!</h3><div class="usage"><code>(handle-mcp-auth-request! client session-id request)</code></div><div class="doc"><div class="markdown"><p>Handle an <code>mcp.oauth_required</code> MCP OAuth request (upstream PR #1669). Returns a channel yielding the wire <code>result</code> map for session.mcp.oauth.handlePendingRequest.</p>

doc/guides/agent-factories.md

@@ -196,7 +196,7 @@ A thunk or stage that throws is either **fatal** or **ordinary**:
 There are three distinct cancellation mechanisms, and they don't overlap:
 
 1. **Cooperative, inside `:run`** — react to `:cancel-chan` closing or poll `:cancelled?` to stop early. Cancellation does not interrupt `:run` automatically; the factory must check for it.
-2. **From outside, server-side** - `cancel-factory-run!` asks the CLI to cancel a run. The runtime then sends a reverse `factory.abort` request, which closes the run's `:cancel-chan` and eventually delivers a `:cancelled` terminal status.
+2. **From outside, server-side** - `cancel-factory-run!` asks the CLI to cancel a run. The runtime then sends a reverse `factory.abort` request for the active execution token, which closes only that attempt's `:cancel-chan` and eventually delivers a `:cancelled` terminal status. An older attempt cannot cancel a replacement that shares the durable run ID.
 3. **Aborting a local wait, not the run** — `wait-for-factory-run!` accepts an optional `:cancel-chan`; closing it makes the *wait* throw `ex-info` with `{:type :factory-wait-cancelled}` without cancelling the run itself. Use this to stop blocking on a run you still want to keep executing.
 
 ## Limits