The Promise.withResolvers signal resolved inside the scripted model
response fires before the loop can possibly dispatch the tool, so the
parked-state assertions passed even with parking disabled (verified by
simulation: 20/20 green with waitUntilResumed stubbed out). Resolve the
readiness signal from a test-local wrap of agentPauseGate.waitUntilResumed
instead: deterministic (no wall-clock race with the cold yieldIfDue
timer), immune to sibling restoreAllMocks (manual patch, restored in
finally), and a non-parking regression now hangs the await and fails
the test.
Resolved tool interruptibility from each call's raw arguments so mixed-operation tools can keep side-effecting calls non-interruptible.
Restricted the unified hub to interrupt passive waits and followed logs while preserving start, send, and lifecycle operation results.
Fixes#5995
Long sessions re-walked the full live AgentMessage[] every turn: convertToLlm
re-converted the unchanged prefix and estimateTokens re-tokenized settled tool
results and assistants, redoing work only the newest suffix can change.
- Added a per-message estimate cache in agent-core keyed by identity, with a
settle gate (assistants cache only with real usage + terminal non-error
stopReason; streaming partials bypass) and dual option-split WeakMaps for the
default vs compaction-floor estimates.
- Memoized convertToLlm per message identity + assistant interruptedNext flag,
with an exact-repeat outer-array reuse and slice-on-growth for append-only
turns, guarded by a boundary-identity check against interior splice-replaces.
- Invalidated both caches at the mutation seams: prune, shake, strip-images, and
the prewalk plan-nudge scrub, via invalidateMessageCache /
registerMessageCacheInvalidator across the package boundary.
- Added the llm-assembly bench (N=5000, robust MAD-noise gate): steady/append
convert and repeat estimate are all >10x faster with noise under 20%.
Fixes#5934
Forwarded the session xd registry into Cursor provider tool contexts.
Routed Cursor MCP execution through the mounted registry fallback and added regression coverage for built-in devices and external MCP tools.
Fixes#5650
Print-mode assistant-error/aborted exit, RPC pi.shutdown() and stdin-EOF
shutdowns, and the extension command-context shutdown() called
process.exit() before (or racing) session.dispose(), skipping the bounded
browser reaper (releaseTabsForOwner) installed in dispose(). An OMP-owned
Chromium could survive the parent and reparent to PID 1.
Route all four graceful paths through the idempotent, promise-memoized
session.dispose() and await it before the final exit. The RPC
performShutdown no longer emits session_shutdown directly (dispose() emits
it), avoiding a double emit.
Fixes#5643
A dropped stream that emitted toolcall_start/delta but never toolcall_end
leaves an incomplete toolCall block in content. The prior guard bailed on
any toolCall block, so the outer loop dispatched empty or partially
parsed arguments instead of retrying the transport failure.
Track streamed tool-call ids (toolcall_start/delta) alongside completed
ones (toolcall_end): a call streamed but never completed is incomplete.
reclassifyEmptyToolUseStop now reclassifies unless a usable (atomic or
completed) tool call remains, stripping incomplete blocks first. Atomic
deliveries (single done/end(result) message, e.g. Cursor) emit no
granular events and stay usable.
Fixes#5600
Streams finalized via end(result) with no terminal done/error event fall
through to the trailing-result branch, which returned response.result()
unchanged. An empty toolUse turn completing that way stayed a silent
success and never retried. Apply the same
retainCompletedToolCalls/recoverTransientErrorToolTurn/
reclassifyEmptyToolUseStop chain to the trailing branch.
Fixes#5600
A provider stream that closes after the thinking block but before the
tool call JSON is emitted finalizes with stopReason=toolUse and zero
toolCall content blocks. The loop treated this as a successful turn:
dispatched no tools, rendered an empty tool widget, and never retried.
reclassifyEmptyToolUseStop stamps such a turn as stopReason=error with
the Transient classifier bit set explicitly, so AgentSession's standard
retry-with-backoff path fires regardless of message-text matching.
Fixes#5600
- Added the `xd://` virtual device protocol (`internal-urls/xd-protocol.ts`, `tools/xdev.ts`): tools declaring `loadMode: "discoverable"` are unmounted from the request tools array and driven via `read xd://` (list/docs+schema) and `write xd://<tool>` (execute), gated by the `tools.xdev` setting (default on) and inlined into the system prompt.
- Merged the `irc`, `job`, and `launch` tools into a single `hub` tool (`tools/hub/`, `async/job-manager.ts`): messaging keeps `send`/`inbox`/`list`, job control maps to `wait`/`cancel`/`jobs`, process supervision keeps `start`/`logs`/`stop`/`restart`/`describe` with `ps`, and the unified `wait` races background jobs against peer messages; SDK `IrcTool`/`JobTool`/`LaunchTool` are replaced by `HubTool`.
- Removed the hidden `resolve` tool in favor of the `xd://resolve`/`xd://reject`/`xd://propose` resolution devices, auto-including `write` whenever a deferrable tool or plan mode is present.
- Removed the BM25 tool-discovery system: the `search_tool_bm25` tool, the `tool-discovery` module, the `tools.discoveryMode`/`mcp.discoveryMode`/`mcp.discoveryDefaultServers`/`tools.essentialOverride` settings, per-tool MCP selection, and the `mcp_tool_selection` message type.
- Unified tool presentation on `ToolLoadMode` (`essential`|`discoverable`), replacing the custom-tool `xdev?: boolean` opt-out; custom, extension, MCP, RPC host, image-generation, and TTS tools now default to `discoverable`, and added a `satisfies` predicate to `SoftToolRequirement`.
- Removed the standalone `ssh` command tool and `ssh/ssh-executor` (the `ssh://` read/write/search protocol stays), and made `--tools` address hidden built-ins.
- Updated collab-web to render `xd://` dispatches and `hub` op families, dropped the `search_tool_bm25`/`ssh`/`report-finding` renderers, refreshed tool docs and prompts, and migrated the affected tests and changelogs.
The interrupt-clobber branch in executeToolCalls replaced a tool's real
result with the "Skipped due to queued user message" placeholder whenever
an interrupt fired, the tool's own signal aborted, and the result was an
error. It ignored whether tool.execute() actually completed, so steering
while a tool was in flight could discard a genuine error result (e.g. a
command exiting non-zero) that the tool had already produced.
Gate the clobber on !completedToolExecution so a tool that ran to
completion keeps its real result; only tools cut off before returning are
reported as skipped. Align the aborted telemetry status the same way.
Fixes#4752
- Disabled the eval watchdog when timeout is explicitly zero.
- Classified session deadline aborts as TimeoutError while preserving their message.
- Documented and tested both timeout contracts.
Fixes#5250
- Introduced an `AgentPauseGate` mechanism to suspend and resume agent loops and tool executions safely.
- Added a `/pause` slash command to trigger a new fullscreen UI that manages agent suspension and lifecycle.
- Integrated pause checks into the core agent loop and tool execution pipeline to ensure responsive state handling.
- Provided a new pause screen component to facilitate user interaction and resume control during suspension.
Escaped Harmony control-token markers when compaction serializes transcripts into plain summary prompts so Copilot gpt-5.6 models do not reject analysis-channel text.
Added regression coverage for summary-bound Harmony serialization while keeping native transcript rendering unchanged.
Fixes#5184
- Introduced `stringifyJson` helper to preserve bigint precision by serializing them as decimal strings.
- Replaced native `JSON.stringify` across compaction and session management modules to prevent serialization errors when handling bigint values in tool arguments.
- Added regression tests in `agent` and `coding-agent` packages to ensure bigint tool arguments remain intact through compaction and persistence flows.
- Force `reasoning.context` to `all_turns` in OpenAI Codex requests when `responsesLite` is enabled.
- Include `reasoning.encrypted_content` in the `include` header for lite responses.
- Update request transformation logic to ensure these fields are populated even when reasoning effort is not explicitly set.
- Enabled Codex Responses Lite for GPT-5.6 models by integrating model discovery flags and wire contract updates.
- Implemented request transformations for streaming and remote compaction, including header injection and image detail stripping.
- Introduced sequential-cutoff logic and atomic reasoning summary events for concurrent stream processing.
- Added comprehensive test suites to validate remote compaction, image handling, and reasoning summary delivery.
Retried handoff generation with toolChoice auto when a provider rejects the cache-preserving toolChoice none request as auto-only.
Kept unrelated provider 400s terminal so bad request failures still surface without masking the cause.
Fixes#4715
- Sent remoteCompaction.model or requestModelId in chat-completions remote compaction requests instead of the local catalog id.
- Covered both direct requestRemoteCompaction formatting and end-to-end openai-completions compaction with wire model ids.
Fixes#4630
- Sent OpenAI-compatible chat messages when compaction.remoteEndpoint targets /chat/completions while preserving the existing custom summarizer payload elsewhere.
- Added regressions for direct wire formatting and end-to-end openai-completions compaction against a configured chat endpoint.
Fixes#4630
calculateContextTokens returned usage.totalTokens which, with the new
Usage.orchestration sidecar, folds provider-side orchestration back into the
context size used by auto-compaction/context promotion thresholds. Subtract
the orchestration sidecar so context sizing stays conversation-only while
cost and totalTokens keep the orchestration spend visible.
Refs #4469
Codex review on PR #4351: synthesizing toolCall content blocks for
Cursor's exec-channel native tools made the shared agent loop treat the
finalized assistant message as a fresh runnable tool turn. Because
executeToolCalls filters message.content for any toolCall block on
stop/toolUse, bash/write/delete/etc. ran a second time after Cursor
already executed them server-side via the bridge, duplicating side
effects and appending conflicting toolResults.
- packages/ai/src/utils/block-symbols.ts: add `kCursorExecResolved`
symbol and `CursorExecResolvedCarrier` carrier type. Symbol-keyed so
the marker never leaks into JSONL; rebuild pairs blocks with toolResult
messages by id.
- packages/ai/src/providers/cursor.ts: stamp the marker onto every
block `synthesizeCursorExecToolCall` emits and extend `ToolCallState`.
- packages/agent/src/agent-loop.ts: filter marked blocks out of the
runnable-toolCall extraction in both the main runnable path and the
error/aborted placeholder path, plus defense-in-depth inside
`executeToolCalls`. Marked blocks stay in `assistantMessage.content`
for persistence + rebuild rendering; they just never re-execute.
- packages/agent/test/agent-loop.test.ts: two regression tests — one
proves a marked block yields zero `tool.execute` calls and no
`tool_execution_*` events from the loop, the other verifies mixed
batches still run the unmarked blocks unchanged.
Fixes#4348
When an assistant turn ends with stopReason="error" after a tool call
was already streamed, agent-loop synthesizes a placeholder tool result
via createAbortedToolResult() to preserve the tool_use / tool_result
pairing the provider API requires. The previous wording ("Tool
execution failed due to an error: <upstream>") and event shape
(normal tool_execution_start / tool_execution_end with empty details)
were indistinguishable from a real local tool failure — a Codex
websocket close mid-turn showed up in the CLI as a broken Edit panel,
misattributing provider-transport faults to the local tool.
Reword the "error" placeholder to state explicitly that the tool
never ran ("Tool call was not executed because the provider stream
ended with an error before the tool could run: <upstream>") and thread
a SyntheticToolResultDetails discriminator ({ __synthetic: true,
source: "assistant_stop_error" | "assistant_stop_aborted" |
"assistant_stop_skipped" | "assistant_stop_length", executed: false,
upstreamError }) through both the ToolResultMessage.details and the
tool_execution_end event's result.details, so downstream UI/telemetry/
ACP consumers can render "provider transport failed, tool not
executed" without string-matching content.
Fixes#4321
- Stopped calling the consuming `getSteeringMessages` getter during mid-batch interrupt polls to prevent stranding or dropping messages before they reach the injection boundary.
- Skip subsequent steering checks in the poll loop once an interrupt has already triggered.
- Added a regression test to ensure legacy steering remains queued until the injection boundary when no non-consuming peek exists.
- Made CompactionSettings.reserveTokens optional so field presence carries provenance; the proportional small-window fallback only applies to genuinely defaulted reserves.
- Clamped the fallback reserve to >= 1 and the derived threshold strictly below the context window.
- Changed the coding-agent settings-schema default from 16384 to unset so Settings.get() no longer materializes a default that masks provenance.
can1357